IO Tools (Text, CSV, HDF5, ...)¶
The Pandas I/O api is a set of top level reader functions accessed like pd.read_csv() that generally return a pandas object.
- read_csv
- read_excel
- read_hdf
- read_sql
- read_json
- read_msgpack (experimental)
- read_html
- read_gbq (experimental)
- read_stata
- read_clipboard
- read_pickle
The corresponding writer functions are object methods that are accessed like df.to_csv()
CSV & Text files¶
The two workhorse functions for reading text files (a.k.a. flat files) are read_csv() and read_table(). They both use the same parsing code to intelligently convert tabular data into a DataFrame object. See the cookbook for some advanced strategies
They can take a number of arguments:
- filepath_or_buffer: Either a string path to a file, url (including http, ftp, and s3 locations), or any object with a read method (such as an open file or StringIO).
- sep or delimiter: A delimiter / separator to split fields on. read_csv is capable of inferring the delimiter automatically in some cases by “sniffing.” The separator may be specified as a regular expression; for instance you may use ‘|\s*’ to indicate a pipe plus arbitrary whitespace.
- delim_whitespace: Parse whitespace-delimited (spaces or tabs) file (much faster than using a regular expression)
- compression: decompress 'gzip' and 'bz2' formats on the fly.
- dialect: string or csv.Dialect instance to expose more ways to specify the file format
- dtype: A data type name or a dict of column name to data type. If not specified, data types will be inferred.
- header: row number(s) to use as the column names, and the start of the data. Defaults to 0 if no names passed, otherwise None. Explicitly pass header=0 to be able to replace existing names. The header can be a list of integers that specify row locations for a multi-index on the columns E.g. [0,1,3]. Intervening rows that are not specified will be skipped. (E.g. 2 in this example are skipped)
- skiprows: A collection of numbers for rows in the file to skip. Can also be an integer to skip the first n rows
- index_col: column number, column name, or list of column numbers/names, to use as the index (row labels) of the resulting DataFrame. By default, it will number the rows without using any column, unless there is one more data column than there are headers, in which case the first column is taken as the index.
- names: List of column names to use as column names. To replace header existing in file, explicitly pass header=0.
- na_values: optional list of strings to recognize as NaN (missing values), either in addition to or in lieu of the default set.
- true_values: list of strings to recognize as True
- false_values: list of strings to recognize as False
- keep_default_na: whether to include the default set of missing values in addition to the ones specified in na_values
- parse_dates: if True then index will be parsed as dates (False by default). You can specify more complicated options to parse a subset of columns or a combination of columns into a single date column (list of ints or names, list of lists, or dict) [1, 2, 3] -> try parsing columns 1, 2, 3 each as a separate date column [[1, 3]] -> combine columns 1 and 3 and parse as a single date column {‘foo’ : [1, 3]} -> parse columns 1, 3 as date and call result ‘foo’
- keep_date_col: if True, then date component columns passed into parse_dates will be retained in the output (False by default).
- date_parser: function to use to parse strings into datetime objects. If parse_dates is True, it defaults to the very robust dateutil.parser. Specifying this implicitly sets parse_dates as True. You can also use functions from community supported date converters from date_converters.py
- dayfirst: if True then uses the DD/MM international/European date format (This is False by default)
- thousands: specifies the thousands separator. If not None, this character will be stripped from numeric dtypes. However, if it is the first character in a field, that column will be imported as a string. In the PythonParser, if not None, then parser will try to look for it in the output and parse relevant data to numeric dtypes. Because it has to essentially scan through the data again, this causes a significant performance hit so only use if necessary.
- lineterminator : string (length 1), default None, Character to break file into lines. Only valid with C parser
- quotechar : string, The character to used to denote the start and end of a quoted item. Quoted items can include the delimiter and it will be ignored.
- quoting : int, Controls whether quotes should be recognized. Values are taken from csv.QUOTE_* values. Acceptable values are 0, 1, 2, and 3 for QUOTE_MINIMAL, QUOTE_ALL, QUOTE_NONE, and QUOTE_NONNUMERIC, respectively.
- skipinitialspace : boolean, default False, Skip spaces after delimiter
- escapechar : string, to specify how to escape quoted data
- comment: denotes the start of a comment and ignores the rest of the line. Currently line commenting is not supported.
- nrows: Number of rows to read out of the file. Useful to only read a small portion of a large file
- iterator: If True, return a TextFileReader to enable reading a file into memory piece by piece
- chunksize: An number of rows to be used to “chunk” a file into pieces. Will cause an TextFileReader object to be returned. More on this below in the section on iterating and chunking
- skip_footer: number of lines to skip at bottom of file (default 0)
- converters: a dictionary of functions for converting values in certain columns, where keys are either integers or column labels
- encoding: a string representing the encoding to use for decoding unicode data, e.g. 'utf-8` or 'latin-1'.
- verbose: show number of NA values inserted in non-numeric columns
- squeeze: if True then output with only one column is turned into Series
- error_bad_lines: if False then any lines causing an error will be skipped bad lines
- usecols: a subset of columns to return, results in much faster parsing time and lower memory usage.
- mangle_dupe_cols: boolean, default True, then duplicate columns will be specified as ‘X.0’...’X.N’, rather than ‘X’...’X’
- tupleize_cols: boolean, default False, if False, convert a list of tuples to a multi-index of columns, otherwise, leave the column index as a list of tuples
Consider a typical CSV file containing, in this case, some time series data:
In [1]: print(open('foo.csv').read())
date,A,B,C
20090101,a,1,2
20090102,b,3,4
20090103,c,4,5
The default for read_csv is to create a DataFrame with simple numbered rows:
In [2]: pd.read_csv('foo.csv')
date A B C
0 20090101 a 1 2
1 20090102 b 3 4
2 20090103 c 4 5
[3 rows x 4 columns]
In the case of indexed data, you can pass the column number or column name you wish to use as the index:
In [3]: pd.read_csv('foo.csv', index_col=0)
A B C
date
20090101 a 1 2
20090102 b 3 4
20090103 c 4 5
[3 rows x 3 columns]
In [4]: pd.read_csv('foo.csv', index_col='date')
A B C
date
20090101 a 1 2
20090102 b 3 4
20090103 c 4 5
[3 rows x 3 columns]
You can also use a list of columns to create a hierarchical index:
In [5]: pd.read_csv('foo.csv', index_col=[0, 'A'])
B C
date A
20090101 a 1 2
20090102 b 3 4
20090103 c 4 5
[3 rows x 2 columns]
The dialect keyword gives greater flexibility in specifying the file format. By default it uses the Excel dialect but you can specify either the dialect name or a csv.Dialect instance.
Suppose you had data with unenclosed quotes:
In [6]: print(data)
label1,label2,label3
index1,"a,c,e
index2,b,d,f
By default, read_csv uses the Excel dialect and treats the double quote as the quote character, which causes it to fail when it finds a newline before it finds the closing double quote.
We can get around this using dialect
In [7]: dia = csv.excel()
In [8]: dia.quoting = csv.QUOTE_NONE
In [9]: pd.read_csv(StringIO(data), dialect=dia)
label1 label2 label3
index1 "a c e
index2 b d f
[2 rows x 3 columns]
All of the dialect options can be specified separately by keyword arguments:
In [10]: data = 'a,b,c~1,2,3~4,5,6'
In [11]: pd.read_csv(StringIO(data), lineterminator='~')
a b c
0 1 2 3
1 4 5 6
[2 rows x 3 columns]
Another common dialect option is skipinitialspace, to skip any whitespace after a delimiter:
In [12]: data = 'a, b, c\n1, 2, 3\n4, 5, 6'
In [13]: print(data)
a, b, c
1, 2, 3
4, 5, 6
In [14]: pd.read_csv(StringIO(data), skipinitialspace=True)
a b c
0 1 2 3
1 4 5 6
[2 rows x 3 columns]
The parsers make every attempt to “do the right thing” and not be very fragile. Type inference is a pretty big deal. So if a column can be coerced to integer dtype without altering the contents, it will do so. Any non-numeric columns will come through as object dtype as with the rest of pandas objects.
Specifying column data types¶
Starting with v0.10, you can indicate the data type for the whole DataFrame or individual columns:
In [15]: data = 'a,b,c\n1,2,3\n4,5,6\n7,8,9'
In [16]: print(data)
a,b,c
1,2,3
4,5,6
7,8,9
In [17]: df = pd.read_csv(StringIO(data), dtype=object)
In [18]: df
a b c
0 1 2 3
1 4 5 6
2 7 8 9
[3 rows x 3 columns]
In [19]: df['a'][0]
'1'
In [20]: df = pd.read_csv(StringIO(data), dtype={'b': object, 'c': np.float64})
In [21]: df.dtypes
a int64
b object
c float64
dtype: object
Handling column names¶
A file may or may not have a header row. pandas assumes the first row should be used as the column names:
In [22]: from StringIO import StringIO
In [23]: data = 'a,b,c\n1,2,3\n4,5,6\n7,8,9'
In [24]: print(data)
a,b,c
1,2,3
4,5,6
7,8,9
In [25]: pd.read_csv(StringIO(data))
a b c
0 1 2 3
1 4 5 6
2 7 8 9
[3 rows x 3 columns]
By specifying the names argument in conjunction with header you can indicate other names to use and whether or not to throw away the header row (if any):
In [26]: print(data)
a,b,c
1,2,3
4,5,6
7,8,9
In [27]: pd.read_csv(StringIO(data), names=['foo', 'bar', 'baz'], header=0)
foo bar baz
0 1 2 3
1 4 5 6
2 7 8 9
[3 rows x 3 columns]
In [28]: pd.read_csv(StringIO(data), names=['foo', 'bar', 'baz'], header=None)
foo bar baz
0 a b c
1 1 2 3
2 4 5 6
3 7 8 9
[4 rows x 3 columns]
If the header is in a row other than the first, pass the row number to header. This will skip the preceding rows:
In [29]: data = 'skip this skip it\na,b,c\n1,2,3\n4,5,6\n7,8,9'
In [30]: pd.read_csv(StringIO(data), header=1)
a b c
0 1 2 3
1 4 5 6
2 7 8 9
[3 rows x 3 columns]
Filtering columns (usecols)¶
The usecols argument allows you to select any subset of the columns in a file, either using the column names or position numbers:
In [31]: data = 'a,b,c,d\n1,2,3,foo\n4,5,6,bar\n7,8,9,baz'
In [32]: pd.read_csv(StringIO(data))
a b c d
0 1 2 3 foo
1 4 5 6 bar
2 7 8 9 baz
[3 rows x 4 columns]
In [33]: pd.read_csv(StringIO(data), usecols=['b', 'd'])
b d
0 2 foo
1 5 bar
2 8 baz
[3 rows x 2 columns]
In [34]: pd.read_csv(StringIO(data), usecols=[0, 2, 3])
a c d
0 1 3 foo
1 4 6 bar
2 7 9 baz
[3 rows x 3 columns]
Dealing with Unicode Data¶
The encoding argument should be used for encoded unicode data, which will result in byte strings being decoded to unicode in the result:
In [35]: data = 'word,length\nTr\xe4umen,7\nGr\xfc\xdfe,5'
In [36]: df = pd.read_csv(StringIO(data), encoding='latin-1')
In [37]: df
word length
0 Träumen 7
1 Grüße 5
[2 rows x 2 columns]
In [38]: df['word'][1]
u'Gr\xfc\xdfe'
Some formats which encode all characters as multiple bytes, like UTF-16, won’t parse correctly at all without specifying the encoding.
Index columns and trailing delimiters¶
If a file has one more column of data than the number of column names, the first column will be used as the DataFrame’s row names:
In [39]: data = 'a,b,c\n4,apple,bat,5.7\n8,orange,cow,10'
In [40]: pd.read_csv(StringIO(data))
a b c
4 apple bat 5.7
8 orange cow 10.0
[2 rows x 3 columns]
In [41]: data = 'index,a,b,c\n4,apple,bat,5.7\n8,orange,cow,10'
In [42]: pd.read_csv(StringIO(data), index_col=0)
a b c
index
4 apple bat 5.7
8 orange cow 10.0
[2 rows x 3 columns]
Ordinarily, you can achieve this behavior using the index_col option.
There are some exception cases when a file has been prepared with delimiters at the end of each data line, confusing the parser. To explicitly disable the index column inference and discard the last column, pass index_col=False:
In [43]: data = 'a,b,c\n4,apple,bat,\n8,orange,cow,'
In [44]: print(data)
a,b,c
4,apple,bat,
8,orange,cow,
In [45]: pd.read_csv(StringIO(data))
a b c
4 apple bat NaN
8 orange cow NaN
[2 rows x 3 columns]
In [46]: pd.read_csv(StringIO(data), index_col=False)
a b c
0 4 apple bat
1 8 orange cow
[2 rows x 3 columns]
Specifying Date Columns¶
To better facilitate working with datetime data, read_csv() and read_table() uses the keyword arguments parse_dates and date_parser to allow users to specify a variety of columns and date/time formats to turn the input text data into datetime objects.
The simplest case is to just pass in parse_dates=True:
# Use a column as an index, and parse it as dates.
In [47]: df = pd.read_csv('foo.csv', index_col=0, parse_dates=True)
In [48]: df
A B C
date
2009-01-01 a 1 2
2009-01-02 b 3 4
2009-01-03 c 4 5
[3 rows x 3 columns]
# These are python datetime objects
In [49]: df.index
<class 'pandas.tseries.index.DatetimeIndex'>
[2009-01-01 00:00:00, ..., 2009-01-03 00:00:00]
Length: 3, Freq: None, Timezone: None
It is often the case that we may want to store date and time data separately, or store various date fields separately. the parse_dates keyword can be used to specify a combination of columns to parse the dates and/or times from.
You can specify a list of column lists to parse_dates, the resulting date columns will be prepended to the output (so as to not affect the existing column order) and the new column names will be the concatenation of the component column names:
In [50]: print(open('tmp.csv').read())
KORD,19990127, 19:00:00, 18:56:00, 0.8100
KORD,19990127, 20:00:00, 19:56:00, 0.0100
KORD,19990127, 21:00:00, 20:56:00, -0.5900
KORD,19990127, 21:00:00, 21:18:00, -0.9900
KORD,19990127, 22:00:00, 21:56:00, -0.5900
KORD,19990127, 23:00:00, 22:56:00, -0.5900
In [51]: df = pd.read_csv('tmp.csv', header=None, parse_dates=[[1, 2], [1, 3]])
In [52]: df
1_2 1_3 0 4
0 1999-01-27 19:00:00 1999-01-27 18:56:00 KORD 0.81
1 1999-01-27 20:00:00 1999-01-27 19:56:00 KORD 0.01
2 1999-01-27 21:00:00 1999-01-27 20:56:00 KORD -0.59
3 1999-01-27 21:00:00 1999-01-27 21:18:00 KORD -0.99
4 1999-01-27 22:00:00 1999-01-27 21:56:00 KORD -0.59
5 1999-01-27 23:00:00 1999-01-27 22:56:00 KORD -0.59
[6 rows x 4 columns]
By default the parser removes the component date columns, but you can choose to retain them via the keep_date_col keyword:
In [53]: df = pd.read_csv('tmp.csv', header=None, parse_dates=[[1, 2], [1, 3]],
....: keep_date_col=True)
....:
In [54]: df
1_2 1_3 0 1 2 \
0 1999-01-27 19:00:00 1999-01-27 18:56:00 KORD 19990127 19:00:00
1 1999-01-27 20:00:00 1999-01-27 19:56:00 KORD 19990127 20:00:00
2 1999-01-27 21:00:00 1999-01-27 20:56:00 KORD 19990127 21:00:00
3 1999-01-27 21:00:00 1999-01-27 21:18:00 KORD 19990127 21:00:00
4 1999-01-27 22:00:00 1999-01-27 21:56:00 KORD 19990127 22:00:00
5 1999-01-27 23:00:00 1999-01-27 22:56:00 KORD 19990127 23:00:00
3 4
0 18:56:00 0.81
1 19:56:00 0.01
2 20:56:00 -0.59
3 21:18:00 -0.99
4 21:56:00 -0.59
5 22:56:00 -0.59
[6 rows x 7 columns]
Note that if you wish to combine multiple columns into a single date column, a nested list must be used. In other words, parse_dates=[1, 2] indicates that the second and third columns should each be parsed as separate date columns while parse_dates=[[1, 2]] means the two columns should be parsed into a single column.
You can also use a dict to specify custom name columns:
In [55]: date_spec = {'nominal': [1, 2], 'actual': [1, 3]}
In [56]: df = pd.read_csv('tmp.csv', header=None, parse_dates=date_spec)
In [57]: df
nominal actual 0 4
0 1999-01-27 19:00:00 1999-01-27 18:56:00 KORD 0.81
1 1999-01-27 20:00:00 1999-01-27 19:56:00 KORD 0.01
2 1999-01-27 21:00:00 1999-01-27 20:56:00 KORD -0.59
3 1999-01-27 21:00:00 1999-01-27 21:18:00 KORD -0.99
4 1999-01-27 22:00:00 1999-01-27 21:56:00 KORD -0.59
5 1999-01-27 23:00:00 1999-01-27 22:56:00 KORD -0.59
[6 rows x 4 columns]
It is important to remember that if multiple text columns are to be parsed into a single date column, then a new column is prepended to the data. The index_col specification is based off of this new set of columns rather than the original data columns:
In [58]: date_spec = {'nominal': [1, 2], 'actual': [1, 3]}
In [59]: df = pd.read_csv('tmp.csv', header=None, parse_dates=date_spec,
....: index_col=0) #index is the nominal column
....:
In [60]: df
actual 0 4
nominal
1999-01-27 19:00:00 1999-01-27 18:56:00 KORD 0.81
1999-01-27 20:00:00 1999-01-27 19:56:00 KORD 0.01
1999-01-27 21:00:00 1999-01-27 20:56:00 KORD -0.59
1999-01-27 21:00:00 1999-01-27 21:18:00 KORD -0.99
1999-01-27 22:00:00 1999-01-27 21:56:00 KORD -0.59
1999-01-27 23:00:00 1999-01-27 22:56:00 KORD -0.59
[6 rows x 3 columns]
Note: When passing a dict as the parse_dates argument, the order of the columns prepended is not guaranteed, because dict objects do not impose an ordering on their keys. On Python 2.7+ you may use collections.OrderedDict instead of a regular dict if this matters to you. Because of this, when using a dict for ‘parse_dates’ in conjunction with the index_col argument, it’s best to specify index_col as a column label rather then as an index on the resulting frame.
Date Parsing Functions¶
Finally, the parser allows you can specify a custom date_parser function to take full advantage of the flexiblity of the date parsing API:
In [61]: import pandas.io.date_converters as conv
In [62]: df = pd.read_csv('tmp.csv', header=None, parse_dates=date_spec,
....: date_parser=conv.parse_date_time)
....:
In [63]: df
nominal actual 0 4
0 1999-01-27 19:00:00 1999-01-27 18:56:00 KORD 0.81
1 1999-01-27 20:00:00 1999-01-27 19:56:00 KORD 0.01
2 1999-01-27 21:00:00 1999-01-27 20:56:00 KORD -0.59
3 1999-01-27 21:00:00 1999-01-27 21:18:00 KORD -0.99
4 1999-01-27 22:00:00 1999-01-27 21:56:00 KORD -0.59
5 1999-01-27 23:00:00 1999-01-27 22:56:00 KORD -0.59
[6 rows x 4 columns]
You can explore the date parsing functionality in date_converters.py and add your own. We would love to turn this module into a community supported set of date/time parsers. To get you started, date_converters.py contains functions to parse dual date and time columns, year/month/day columns, and year/month/day/hour/minute/second columns. It also contains a generic_parser function so you can curry it with a function that deals with a single date rather than the entire array.
International Date Formats¶
While US date formats tend to be MM/DD/YYYY, many international formats use DD/MM/YYYY instead. For convenience, a dayfirst keyword is provided:
In [64]: print(open('tmp.csv').read())
date,value,cat
1/6/2000,5,a
2/6/2000,10,b
3/6/2000,15,c
In [65]: pd.read_csv('tmp.csv', parse_dates=[0])
date value cat
0 2000-01-06 00:00:00 5 a
1 2000-02-06 00:00:00 10 b
2 2000-03-06 00:00:00 15 c
[3 rows x 3 columns]
In [66]: pd.read_csv('tmp.csv', dayfirst=True, parse_dates=[0])
date value cat
0 2000-06-01 00:00:00 5 a
1 2000-06-02 00:00:00 10 b
2 2000-06-03 00:00:00 15 c
[3 rows x 3 columns]
Thousand Separators¶
For large numbers that have been written with a thousands separator, you can set the thousands keyword to a string of length 1 so that integers will be parsed correctly:
By default, numbers with a thousands separator will be parsed as strings
In [67]: print(open('tmp.csv').read())
ID|level|category
Patient1|123,000|x
Patient2|23,000|y
Patient3|1,234,018|z
In [68]: df = pd.read_csv('tmp.csv', sep='|')
In [69]: df
ID level category
0 Patient1 123,000 x
1 Patient2 23,000 y
2 Patient3 1,234,018 z
[3 rows x 3 columns]
In [70]: df.level.dtype
dtype('O')
The thousands keyword allows integers to be parsed correctly
In [71]: print(open('tmp.csv').read())
ID|level|category
Patient1|123,000|x
Patient2|23,000|y
Patient3|1,234,018|z
In [72]: df = pd.read_csv('tmp.csv', sep='|', thousands=',')
In [73]: df
ID level category
0 Patient1 123000 x
1 Patient2 23000 y
2 Patient3 1234018 z
[3 rows x 3 columns]
In [74]: df.level.dtype
dtype('int64')
NA Values¶
To control which values are parsed as missing values (which are signified by NaN), specifiy a list of strings in na_values. If you specify a number (a float, like 5.0 or an integer like 5), the corresponding equivalent values will also imply a missing value (in this case effectively [5.0,5] are recognized as NaN.
To completely override the default values that are recognized as missing, specify keep_default_na=False. The default NaN recognized values are ['-1.#IND', '1.#QNAN', '1.#IND', '-1.#QNAN', '#N/A','N/A', 'NA', '#NA', 'NULL', 'NaN', 'nan'].
read_csv(path, na_values=[5])
the default values, in addition to 5 , 5.0 when interpreted as numbers are recognized as NaN
read_csv(path, keep_default_na=False, na_values=[""])
only an empty field will be NaN
read_csv(path, keep_default_na=False, na_values=["NA", "0"])
only NA and 0 as strings are NaN
read_csv(path, na_values=["Nope"])
the default values, in addition to the string "Nope" are recognized as NaN
Infinity¶
inf like values will be parsed as np.inf (positive infinity), and -inf as -np.inf (negative infinity). These will ignore the case of the value, meaning Inf, will also be parsed as np.inf.
Comments¶
Sometimes comments or meta data may be included in a file:
In [75]: print(open('tmp.csv').read())
ID,level,category
Patient1,123000,x # really unpleasant
Patient2,23000,y # wouldn't take his medicine
Patient3,1234018,z # awesome
By default, the parse includes the comments in the output:
In [76]: df = pd.read_csv('tmp.csv')
In [77]: df
ID level category
0 Patient1 123000 x # really unpleasant
1 Patient2 23000 y # wouldn't take his medicine
2 Patient3 1234018 z # awesome
[3 rows x 3 columns]
We can suppress the comments using the comment keyword:
In [78]: df = pd.read_csv('tmp.csv', comment='#')
In [79]: df
ID level category
0 Patient1 123000 x
1 Patient2 23000 y
2 Patient3 1234018 z
[3 rows x 3 columns]
Returning Series¶
Using the squeeze keyword, the parser will return output with a single column as a Series:
In [80]: print(open('tmp.csv').read())
level
Patient1,123000
Patient2,23000
Patient3,1234018
In [81]: output = pd.read_csv('tmp.csv', squeeze=True)
In [82]: output
Patient1 123000
Patient2 23000
Patient3 1234018
Name: level, dtype: int64
In [83]: type(output)
pandas.core.series.Series
Boolean values¶
The common values True, False, TRUE, and FALSE are all recognized as boolean. Sometime you would want to recognize some other values as being boolean. To do this use the true_values and false_values options:
In [84]: data= 'a,b,c\n1,Yes,2\n3,No,4'
In [85]: print(data)
a,b,c
1,Yes,2
3,No,4
In [86]: pd.read_csv(StringIO(data))
a b c
0 1 Yes 2
1 3 No 4
[2 rows x 3 columns]
In [87]: pd.read_csv(StringIO(data), true_values=['Yes'], false_values=['No'])
a b c
0 1 True 2
1 3 False 4
[2 rows x 3 columns]
Handling “bad” lines¶
Some files may have malformed lines with too few fields or too many. Lines with too few fields will have NA values filled in the trailing fields. Lines with too many will cause an error by default:
In [27]: data = 'a,b,c\n1,2,3\n4,5,6,7\n8,9,10'
In [28]: pd.read_csv(StringIO(data))
---------------------------------------------------------------------------
CParserError Traceback (most recent call last)
CParserError: Error tokenizing data. C error: Expected 3 fields in line 3, saw 4
You can elect to skip bad lines:
In [29]: pd.read_csv(StringIO(data), error_bad_lines=False)
Skipping line 3: expected 3 fields, saw 4
Out[29]:
a b c
0 1 2 3
1 8 9 10
Quoting and Escape Characters¶
Quotes (and other escape characters) in embedded fields can be handled in any number of ways. One way is to use backslashes; to properly parse this data, you should pass the escapechar option:
In [88]: data = 'a,b\n"hello, \\"Bob\\", nice to see you",5'
In [89]: print(data)
a,b
"hello, \"Bob\", nice to see you",5
In [90]: pd.read_csv(StringIO(data), escapechar='\\')
a b
0 hello, "Bob", nice to see you 5
[1 rows x 2 columns]
Files with Fixed Width Columns¶
While read_csv reads delimited data, the read_fwf() function works with data files that have known and fixed column widths. The function parameters to read_fwf are largely the same as read_csv with two extra parameters:
- colspecs: A list of pairs (tuples) giving the extents of the fixed-width fields of each line as half-open intervals (i.e., [from, to[ ). String value ‘infer’ can be used to instruct the parser to try detecting the column specifications from the first 100 rows of the data. Default behaviour, if not specified, is to infer.
- widths: A list of field widths which can be used instead of ‘colspecs’ if the intervals are contiguous.
Consider a typical fixed-width data file:
In [91]: print(open('bar.csv').read())
id8141 360.242940 149.910199 11950.7
id1594 444.953632 166.985655 11788.4
id1849 364.136849 183.628767 11806.2
id1230 413.836124 184.375703 11916.8
id1948 502.953953 173.237159 12468.3
In order to parse this file into a DataFrame, we simply need to supply the column specifications to the read_fwf function along with the file name:
#Column specifications are a list of half-intervals
In [92]: colspecs = [(0, 6), (8, 20), (21, 33), (34, 43)]
In [93]: df = pd.read_fwf('bar.csv', colspecs=colspecs, header=None, index_col=0)
In [94]: df
1 2 3
0
id8141 360.242940 149.910199 11950.7
id1594 444.953632 166.985655 11788.4
id1849 364.136849 183.628767 11806.2
id1230 413.836124 184.375703 11916.8
id1948 502.953953 173.237159 12468.3
[5 rows x 3 columns]
Note how the parser automatically picks column names X.<column number> when header=None argument is specified. Alternatively, you can supply just the column widths for contiguous columns:
#Widths are a list of integers
In [95]: widths = [6, 14, 13, 10]
In [96]: df = pd.read_fwf('bar.csv', widths=widths, header=None)
In [97]: df
0 1 2 3
0 id8141 360.242940 149.910199 11950.7
1 id1594 444.953632 166.985655 11788.4
2 id1849 364.136849 183.628767 11806.2
3 id1230 413.836124 184.375703 11916.8
4 id1948 502.953953 173.237159 12468.3
[5 rows x 4 columns]
The parser will take care of extra white spaces around the columns so it’s ok to have extra separation between the columns in the file.
New in version 0.13.0.
By default, read_fwf will try to infer the file’s colspecs by using the first 100 rows of the file. It can do it only in cases when the columns are aligned and correctly separated by the provided delimiter (default delimiter is whitespace).
In [98]: df = pd.read_fwf('bar.csv', header=None, index_col=0)
In [99]: df
1 2 3
0
id8141 360.242940 149.910199 11950.7
id1594 444.953632 166.985655 11788.4
id1849 364.136849 183.628767 11806.2
id1230 413.836124 184.375703 11916.8
id1948 502.953953 173.237159 12468.3
[5 rows x 3 columns]
Files with an “implicit” index column¶
Consider a file with one less entry in the header than the number of data column:
In [100]: print(open('foo.csv').read())
A,B,C
20090101,a,1,2
20090102,b,3,4
20090103,c,4,5
In this special case, read_csv assumes that the first column is to be used as the index of the DataFrame:
In [101]: pd.read_csv('foo.csv')
A B C
20090101 a 1 2
20090102 b 3 4
20090103 c 4 5
[3 rows x 3 columns]
Note that the dates weren’t automatically parsed. In that case you would need to do as before:
In [102]: df = pd.read_csv('foo.csv', parse_dates=True)
In [103]: df.index
<class 'pandas.tseries.index.DatetimeIndex'>
[2009-01-01 00:00:00, ..., 2009-01-03 00:00:00]
Length: 3, Freq: None, Timezone: None
Reading an index with a MultiIndex¶
Suppose you have data indexed by two columns:
In [104]: print(open('data/mindex_ex.csv').read())
year,indiv,zit,xit
1977,"A",1.2,.6
1977,"B",1.5,.5
1977,"C",1.7,.8
1978,"A",.2,.06
1978,"B",.7,.2
1978,"C",.8,.3
1978,"D",.9,.5
1978,"E",1.4,.9
1979,"C",.2,.15
1979,"D",.14,.05
1979,"E",.5,.15
1979,"F",1.2,.5
1979,"G",3.4,1.9
1979,"H",5.4,2.7
1979,"I",6.4,1.2
The index_col argument to read_csv and read_table can take a list of column numbers to turn multiple columns into a MultiIndex for the index of the returned object:
In [105]: df = pd.read_csv("data/mindex_ex.csv", index_col=[0,1])
In [106]: df
zit xit
year indiv
1977 A 1.20 0.60
B 1.50 0.50
C 1.70 0.80
1978 A 0.20 0.06
B 0.70 0.20
C 0.80 0.30
D 0.90 0.50
E 1.40 0.90
1979 C 0.20 0.15
D 0.14 0.05
E 0.50 0.15
F 1.20 0.50
G 3.40 1.90
H 5.40 2.70
I 6.40 1.20
[15 rows x 2 columns]
In [107]: df.ix[1978]
zit xit
indiv
A 0.2 0.06
B 0.7 0.20
C 0.8 0.30
D 0.9 0.50
E 1.4 0.90
[5 rows x 2 columns]
Reading columns with a MultiIndex¶
By specifying list of row locations for the header argument, you can read in a MultiIndex for the columns. Specifying non-consecutive rows will skip the interveaning rows. In order to have the pre-0.13 behavior of tupleizing columns, specify tupleize_cols=True.
In [108]: from pandas.util.testing import makeCustomDataframe as mkdf
In [109]: df = mkdf(5,3,r_idx_nlevels=2,c_idx_nlevels=4)
In [110]: df.to_csv('mi.csv')
In [111]: print(open('mi.csv').read())
C0,,C_l0_g0,C_l0_g1,C_l0_g2
C1,,C_l1_g0,C_l1_g1,C_l1_g2
C2,,C_l2_g0,C_l2_g1,C_l2_g2
C3,,C_l3_g0,C_l3_g1,C_l3_g2
R0,R1,,,
R_l0_g0,R_l1_g0,R0C0,R0C1,R0C2
R_l0_g1,R_l1_g1,R1C0,R1C1,R1C2
R_l0_g2,R_l1_g2,R2C0,R2C1,R2C2
R_l0_g3,R_l1_g3,R3C0,R3C1,R3C2
R_l0_g4,R_l1_g4,R4C0,R4C1,R4C2
In [112]: pd.read_csv('mi.csv',header=[0,1,2,3],index_col=[0,1])
C0 C_l0_g0 C_l0_g1 C_l0_g2
C1 C_l1_g0 C_l1_g1 C_l1_g2
C2 C_l2_g0 C_l2_g1 C_l2_g2
C3 C_l3_g0 C_l3_g1 C_l3_g2
R0 R1
R_l0_g0 R_l1_g0 R0C0 R0C1 R0C2
R_l0_g1 R_l1_g1 R1C0 R1C1 R1C2
R_l0_g2 R_l1_g2 R2C0 R2C1 R2C2
R_l0_g3 R_l1_g3 R3C0 R3C1 R3C2
R_l0_g4 R_l1_g4 R4C0 R4C1 R4C2
[5 rows x 3 columns]
Starting in 0.13.0, read_csv will be able to interpret a more common format of multi-columns indices.
In [113]: print(open('mi2.csv').read())
,a,a,a,b,c,c
,q,r,s,t,u,v
one,1,2,3,4,5,6
two,7,8,9,10,11,12
In [114]: pd.read_csv('mi2.csv',header=[0,1],index_col=0)
a b c
q r s t u v
one 1 2 3 4 5 6
two 7 8 9 10 11 12
[2 rows x 6 columns]
Note: If an index_col is not specified (e.g. you don’t have an index, or wrote it with df.to_csv(..., index=False), then any names on the columns index will be lost.
Automatically “sniffing” the delimiter¶
read_csv is capable of inferring delimited (not necessarily comma-separated) files. YMMV, as pandas uses the csv.Sniffer class of the csv module.
In [115]: print(open('tmp2.sv').read())
:0:1:2:3
0:0.4691122999071863:-0.2828633443286633:-1.5090585031735124:-1.1356323710171934
1:1.2121120250208506:-0.17321464905330858:0.11920871129693428:-1.0442359662799567
2:-0.8618489633477999:-2.1045692188948086:-0.4949292740687813:1.071803807037338
3:0.7215551622443669:-0.7067711336300845:-1.0395749851146963:0.27185988554282986
4:-0.42497232978883753:0.567020349793672:0.27623201927771873:-1.0874006912859915
5:-0.6736897080883706:0.1136484096888855:-1.4784265524372235:0.5249876671147047
6:0.4047052186802365:0.5770459859204836:-1.7150020161146375:-1.0392684835147725
7:-0.3706468582364464:-1.1578922506419993:-1.344311812731667:0.8448851414248841
8:1.0757697837155533:-0.10904997528022223:1.6435630703622064:-1.4693879595399115
9:0.35702056413309086:-0.6746001037299882:-1.776903716971867:-0.9689138124473498
In [116]: pd.read_csv('tmp2.sv')
:0:1:2:3
0 0:0.4691122999071863:-0.2828633443286633:-1.50...
1 1:1.2121120250208506:-0.17321464905330858:0.11...
2 2:-0.8618489633477999:-2.1045692188948086:-0.4...
3 3:0.7215551622443669:-0.7067711336300845:-1.03...
4 4:-0.42497232978883753:0.567020349793672:0.276...
5 5:-0.6736897080883706:0.1136484096888855:-1.47...
6 6:0.4047052186802365:0.5770459859204836:-1.715...
7 7:-0.3706468582364464:-1.1578922506419993:-1.3...
8 8:1.0757697837155533:-0.10904997528022223:1.64...
9 9:0.35702056413309086:-0.6746001037299882:-1.7...
[10 rows x 1 columns]
Iterating through files chunk by chunk¶
Suppose you wish to iterate through a (potentially very large) file lazily rather than reading the entire file into memory, such as the following:
In [117]: print(open('tmp.sv').read())
|0|1|2|3
0|0.4691122999071863|-0.2828633443286633|-1.5090585031735124|-1.1356323710171934
1|1.2121120250208506|-0.17321464905330858|0.11920871129693428|-1.0442359662799567
2|-0.8618489633477999|-2.1045692188948086|-0.4949292740687813|1.071803807037338
3|0.7215551622443669|-0.7067711336300845|-1.0395749851146963|0.27185988554282986
4|-0.42497232978883753|0.567020349793672|0.27623201927771873|-1.0874006912859915
5|-0.6736897080883706|0.1136484096888855|-1.4784265524372235|0.5249876671147047
6|0.4047052186802365|0.5770459859204836|-1.7150020161146375|-1.0392684835147725
7|-0.3706468582364464|-1.1578922506419993|-1.344311812731667|0.8448851414248841
8|1.0757697837155533|-0.10904997528022223|1.6435630703622064|-1.4693879595399115
9|0.35702056413309086|-0.6746001037299882|-1.776903716971867|-0.9689138124473498
In [118]: table = pd.read_table('tmp.sv', sep='|')
In [119]: table
Unnamed: 0 0 1 2 3
0 0 0.469112 -0.282863 -1.509059 -1.135632
1 1 1.212112 -0.173215 0.119209 -1.044236
2 2 -0.861849 -2.104569 -0.494929 1.071804
3 3 0.721555 -0.706771 -1.039575 0.271860
4 4 -0.424972 0.567020 0.276232 -1.087401
5 5 -0.673690 0.113648 -1.478427 0.524988
6 6 0.404705 0.577046 -1.715002 -1.039268
7 7 -0.370647 -1.157892 -1.344312 0.844885
8 8 1.075770 -0.109050 1.643563 -1.469388
9 9 0.357021 -0.674600 -1.776904 -0.968914
[10 rows x 5 columns]
By specifiying a chunksize to read_csv or read_table, the return value will be an iterable object of type TextFileReader:
In [120]: reader = pd.read_table('tmp.sv', sep='|', chunksize=4)
In [121]: reader
<pandas.io.parsers.TextFileReader at 0xce290d0>
In [122]: for chunk in reader:
.....: print(chunk)
.....:
Unnamed: 0 0 1 2 3
0 0 0.469112 -0.282863 -1.509059 -1.135632
1 1 1.212112 -0.173215 0.119209 -1.044236
2 2 -0.861849 -2.104569 -0.494929 1.071804
3 3 0.721555 -0.706771 -1.039575 0.271860
[4 rows x 5 columns]
Unnamed: 0 0 1 2 3
0 4 -0.424972 0.567020 0.276232 -1.087401
1 5 -0.673690 0.113648 -1.478427 0.524988
2 6 0.404705 0.577046 -1.715002 -1.039268
3 7 -0.370647 -1.157892 -1.344312 0.844885
[4 rows x 5 columns]
Unnamed: 0 0 1 2 3
0 8 1.075770 -0.10905 1.643563 -1.469388
1 9 0.357021 -0.67460 -1.776904 -0.968914
[2 rows x 5 columns]
Specifying iterator=True will also return the TextFileReader object:
In [123]: reader = pd.read_table('tmp.sv', sep='|', iterator=True)
In [124]: reader.get_chunk(5)
Unnamed: 0 0 1 2 3
0 0 0.469112 -0.282863 -1.509059 -1.135632
1 1 1.212112 -0.173215 0.119209 -1.044236
2 2 -0.861849 -2.104569 -0.494929 1.071804
3 3 0.721555 -0.706771 -1.039575 0.271860
4 4 -0.424972 0.567020 0.276232 -1.087401
[5 rows x 5 columns]
Writing to CSV format¶
The Series and DataFrame objects have an instance method to_csv which allows storing the contents of the object as a comma-separated-values file. The function takes a number of arguments. Only the first is required.
- path: A string path to the file to write
- na_rep: A string representation of a missing value (default ‘’)
- cols: Columns to write (default None)
- header: Whether to write out the column names (default True)
- index: whether to write row (index) names (default True)
- index_label: Column label(s) for index column(s) if desired. If None (default), and header and index are True, then the index names are used. (A sequence should be given if the DataFrame uses MultiIndex).
- mode : Python write mode, default ‘w’
- sep : Field delimiter for the output file (default ”,”)
- encoding: a string representing the encoding to use if the contents are non-ascii, for python versions prior to 3
- tupleize_cols: boolean, default False, if False, write as a list of tuples, otherwise write in an expanded line format suitable for read_csv
Writing a formatted string¶
The DataFrame object has an instance method to_string which allows control over the string representation of the object. All arguments are optional:
- buf default None, for example a StringIO object
- columns default None, which columns to write
- col_space default None, minimum width of each column.
- na_rep default NaN, representation of NA value
- formatters default None, a dictionary (by column) of functions each of which takes a single argument and returns a formatted string
- float_format default None, a function which takes a single (float) argument and returns a formatted string; to be applied to floats in the DataFrame.
- sparsify default True, set to False for a DataFrame with a hierarchical index to print every multiindex key at each row.
- index_names default True, will print the names of the indices
- index default True, will print the index (ie, row labels)
- header default True, will print the column labels
- justify default left, will print column headers left- or right-justified
The Series object also has a to_string method, but with only the buf, na_rep, float_format arguments. There is also a length argument which, if set to True, will additionally output the length of the Series.
JSON¶
Read and write JSON format files and strings.
Writing JSON¶
A Series or DataFrame can be converted to a valid JSON string. Use to_json with optional parameters:
path_or_buf : the pathname or buffer to write the output This can be None in which case a JSON string is returned
orient :
- Series :
- default is index
- allowed values are {split, records, index}
- DataFrame
- default is columns
- allowed values are {split, records, index, columns, values}
The format of the JSON string
split dict like {index -> [index], columns -> [columns], data -> [values]} records list like [{column -> value}, ... , {column -> value}] index dict like {index -> {column -> value}} columns dict like {column -> {index -> value}} values just the values array date_format : string, type of date conversion, ‘epoch’ for timestamp, ‘iso’ for ISO8601.
double_precision : The number of decimal places to use when encoding floating point values, default 10.
force_ascii : force encoded string to be ASCII, default True.
date_unit : The time unit to encode to, governs timestamp and ISO8601 precision. One of ‘s’, ‘ms’, ‘us’ or ‘ns’ for seconds, milliseconds, microseconds and nanoseconds respectively. Default ‘ms’.
default_handler : The handler to call if an object cannot otherwise be converted to a suitable format for JSON. Takes a single argument, which is the object to convert, and returns a serialisable object.
Note NaN‘s, NaT‘s and None will be converted to null and datetime objects will be converted based on the date_format and date_unit parameters.
In [125]: dfj = DataFrame(randn(5, 2), columns=list('AB'))
In [126]: json = dfj.to_json()
In [127]: json
'{"A":{"0":-1.2945235903,"1":0.2766617129,"2":-0.0139597524,"3":-0.0061535699,"4":0.8957173022},"B":{"0":0.4137381054,"1":-0.472034511,"2":-0.3625429925,"3":-0.923060654,"4":0.8052440254}}'
Orient Options¶
There are a number of different options for the format of the resulting JSON file / string. Consider the following DataFrame and Series:
In [128]: dfjo = DataFrame(dict(A=range(1, 4), B=range(4, 7), C=range(7, 10)),
.....: columns=list('ABC'), index=list('xyz'))
.....:
In [129]: dfjo
A B C
x 1 4 7
y 2 5 8
z 3 6 9
[3 rows x 3 columns]
In [130]: sjo = Series(dict(x=15, y=16, z=17), name='D')
In [131]: sjo
x 15
y 16
z 17
Name: D, dtype: int64
Column oriented (the default for DataFrame) serialises the data as nested JSON objects with column labels acting as the primary index:
In [132]: dfjo.to_json(orient="columns")
'{"A":{"x":1,"y":2,"z":3},"B":{"x":4,"y":5,"z":6},"C":{"x":7,"y":8,"z":9}}'
Index oriented (the default for Series) similar to column oriented but the index labels are now primary:
In [133]: dfjo.to_json(orient="index")
'{"x":{"A":1,"B":4,"C":7},"y":{"A":2,"B":5,"C":8},"z":{"A":3,"B":6,"C":9}}'
In [134]: sjo.to_json(orient="index")
'{"x":15,"y":16,"z":17}'
Record oriented serialises the data to a JSON array of column -> value records, index labels are not included. This is useful for passing DataFrame data to plotting libraries, for example the JavaScript library d3.js:
In [135]: dfjo.to_json(orient="records")
'[{"A":1,"B":4,"C":7},{"A":2,"B":5,"C":8},{"A":3,"B":6,"C":9}]'
In [136]: sjo.to_json(orient="records")
'[15,16,17]'
Value oriented is a bare-bones option which serialises to nested JSON arrays of values only, column and index labels are not included:
In [137]: dfjo.to_json(orient="values")
'[[1,4,7],[2,5,8],[3,6,9]]'
Split oriented serialises to a JSON object containing separate entries for values, index and columns. Name is also included for Series:
In [138]: dfjo.to_json(orient="split")
'{"columns":["A","B","C"],"index":["x","y","z"],"data":[[1,4,7],[2,5,8],[3,6,9]]}'
In [139]: sjo.to_json(orient="split")
'{"name":"D","index":["x","y","z"],"data":[15,16,17]}'
Note
Any orient option that encodes to a JSON object will not preserve the ordering of index and column labels during round-trip serialisation. If you wish to preserve label ordering use the split option as it uses ordered containers.
Date Handling¶
Writing in iso date format
In [140]: dfd = DataFrame(randn(5, 2), columns=list('AB'))
In [141]: dfd['date'] = Timestamp('20130101')
In [142]: dfd = dfd.sort_index(1, ascending=False)
In [143]: json = dfd.to_json(date_format='iso')
In [144]: json
'{"date":{"0":"2013-01-01T00:00:00.000Z","1":"2013-01-01T00:00:00.000Z","2":"2013-01-01T00:00:00.000Z","3":"2013-01-01T00:00:00.000Z","4":"2013-01-01T00:00:00.000Z"},"B":{"0":2.5656459463,"1":1.3403088498,"2":-0.2261692849,"3":0.8138502857,"4":-0.8273169356},"A":{"0":-1.2064117817,"1":1.4312559863,"2":-1.1702987971,"3":0.4108345112,"4":0.1320031703}}'
Writing in iso date format, with microseconds
In [145]: json = dfd.to_json(date_format='iso', date_unit='us')
In [146]: json
'{"date":{"0":"2013-01-01T00:00:00.000000Z","1":"2013-01-01T00:00:00.000000Z","2":"2013-01-01T00:00:00.000000Z","3":"2013-01-01T00:00:00.000000Z","4":"2013-01-01T00:00:00.000000Z"},"B":{"0":2.5656459463,"1":1.3403088498,"2":-0.2261692849,"3":0.8138502857,"4":-0.8273169356},"A":{"0":-1.2064117817,"1":1.4312559863,"2":-1.1702987971,"3":0.4108345112,"4":0.1320031703}}'
Epoch timestamps, in seconds
In [147]: json = dfd.to_json(date_format='epoch', date_unit='s')
In [148]: json
'{"date":{"0":1356998400,"1":1356998400,"2":1356998400,"3":1356998400,"4":1356998400},"B":{"0":2.5656459463,"1":1.3403088498,"2":-0.2261692849,"3":0.8138502857,"4":-0.8273169356},"A":{"0":-1.2064117817,"1":1.4312559863,"2":-1.1702987971,"3":0.4108345112,"4":0.1320031703}}'
Writing to a file, with a date index and a date column
In [149]: dfj2 = dfj.copy()
In [150]: dfj2['date'] = Timestamp('20130101')
In [151]: dfj2['ints'] = list(range(5))
In [152]: dfj2['bools'] = True
In [153]: dfj2.index = date_range('20130101', periods=5)
In [154]: dfj2.to_json('test.json')
In [155]: open('test.json').read()
'{"A":{"1356998400000":-1.2945235903,"1357084800000":0.2766617129,"1357171200000":-0.0139597524,"1357257600000":-0.0061535699,"1357344000000":0.8957173022},"B":{"1356998400000":0.4137381054,"1357084800000":-0.472034511,"1357171200000":-0.3625429925,"1357257600000":-0.923060654,"1357344000000":0.8052440254},"date":{"1356998400000":1356998400000,"1357084800000":1356998400000,"1357171200000":1356998400000,"1357257600000":1356998400000,"1357344000000":1356998400000},"ints":{"1356998400000":0,"1357084800000":1,"1357171200000":2,"1357257600000":3,"1357344000000":4},"bools":{"1356998400000":true,"1357084800000":true,"1357171200000":true,"1357257600000":true,"1357344000000":true}}'
Fallback Behavior¶
If the JSON serialiser cannot handle the container contents directly it will fallback in the following manner:
- if a toDict method is defined by the unrecognised object then that will be called and its returned dict will be JSON serialised.
- if a default_handler has been passed to to_json that will be called to convert the object.
- otherwise an attempt is made to convert the object to a dict by parsing its contents. However if the object is complex this will often fail with an OverflowError.
Your best bet when encountering OverflowError during serialisation is to specify a default_handler. For example timedelta can cause problems:
In [141]: from datetime import timedelta
In [142]: dftd = DataFrame([timedelta(23), timedelta(seconds=5), 42])
In [143]: dftd.to_json()
---------------------------------------------------------------------------
OverflowError Traceback (most recent call last)
OverflowError: Maximum recursion level reached
which can be dealt with by specifying a simple default_handler:
In [156]: dftd.to_json(default_handler=str)
'{"0":{"0":"23 days, 0:00:00","1":"0:00:05","2":42}}'
In [157]: def my_handler(obj):
.....: return obj.total_seconds()
.....:
In [158]: dftd.to_json(default_handler=my_handler)
'{"0":{"0":1987200.0,"1":5.0,"2":42}}'
Reading JSON¶
Reading a JSON string to pandas object can take a number of parameters. The parser will try to parse a DataFrame if typ is not supplied or is None. To explicity force Series parsing, pass typ=series
filepath_or_buffer : a VALID JSON string or file handle / StringIO. The string could be a URL. Valid URL schemes include http, ftp, s3, and file. For file URLs, a host is expected. For instance, a local file could be file ://localhost/path/to/table.json
typ : type of object to recover (series or frame), default ‘frame’
orient :
- Series :
- default is index
- allowed values are {split, records, index}
- DataFrame
- default is columns
- allowed values are {split, records, index, columns, values}
The format of the JSON string
split dict like {index -> [index], columns -> [columns], data -> [values]} records list like [{column -> value}, ... , {column -> value}] index dict like {index -> {column -> value}} columns dict like {column -> {index -> value}} values just the values array dtype : if True, infer dtypes, if a dict of column to dtype, then use those, if False, then don’t infer dtypes at all, default is True, apply only to the data
convert_axes : boolean, try to convert the axes to the proper dtypes, default is True
convert_dates : a list of columns to parse for dates; If True, then try to parse datelike columns, default is True
keep_default_dates : boolean, default True. If parsing dates, then parse the default datelike columns
numpy : direct decoding to numpy arrays. default is False; Supports numeric data only, although labels may be non-numeric. Also note that the JSON ordering MUST be the same for each term if numpy=True
precise_float : boolean, default False. Set to enable usage of higher precision (strtod) function when decoding string to double values. Default (False) is to use fast but less precise builtin functionality
date_unit : string, the timestamp unit to detect if converting dates. Default None. By default the timestamp precision will be detected, if this is not desired then pass one of ‘s’, ‘ms’, ‘us’ or ‘ns’ to force timestamp precision to seconds, milliseconds, microseconds or nanoseconds respectively.
The parser will raise one of ValueError/TypeError/AssertionError if the JSON is not parsable.
If a non-default orient was used when encoding to JSON be sure to pass the same option here so that decoding produces sensible results, see Orient Options for an overview.
Data Conversion¶
The default of convert_axes=True, dtype=True, and convert_dates=True will try to parse the axes, and all of the data into appropriate types, including dates. If you need to override specific dtypes, pass a dict to dtype. convert_axes should only be set to False if you need to preserve string-like numbers (e.g. ‘1’, ‘2’) in an axes.
Note
Large integer values may be converted to dates if convert_dates=True and the data and / or column labels appear ‘date-like’. The exact threshold depends on the date_unit specified.
Warning
When reading JSON data, automatic coercing into dtypes has some quirks:
- an index can be reconstructed in a different order from serialization, that is, the returned order is not guaranteed to be the same as before serialization
- a column that was float data will be converted to integer if it can be done safely, e.g. a column of 1.
- bool columns will be converted to integer on reconstruction
Thus there are times where you may want to specify specific dtypes via the dtype keyword argument.
Reading from a JSON string:
In [159]: pd.read_json(json)
A B date
0 -1.206412 2.565646 2013-01-01 00:00:00
1 1.431256 1.340309 2013-01-01 00:00:00
2 -1.170299 -0.226169 2013-01-01 00:00:00
3 0.410835 0.813850 2013-01-01 00:00:00
4 0.132003 -0.827317 2013-01-01 00:00:00
[5 rows x 3 columns]
Reading from a file:
In [160]: pd.read_json('test.json')
A B bools date ints
2013-01-01 -1.294524 0.413738 True 2013-01-01 00:00:00 0
2013-01-02 0.276662 -0.472035 True 2013-01-01 00:00:00 1
2013-01-03 -0.013960 -0.362543 True 2013-01-01 00:00:00 2
2013-01-04 -0.006154 -0.923061 True 2013-01-01 00:00:00 3
2013-01-05 0.895717 0.805244 True 2013-01-01 00:00:00 4
[5 rows x 5 columns]
Don’t convert any data (but still convert axes and dates):
In [161]: pd.read_json('test.json', dtype=object).dtypes
A object
B object
bools object
date datetime64[ns]
ints object
dtype: object
Specify dtypes for conversion:
In [162]: pd.read_json('test.json', dtype={'A' : 'float32', 'bools' : 'int8'}).dtypes
A float32
B float64
bools int8
date datetime64[ns]
ints int64
dtype: object
Preserve string indicies:
In [163]: si = DataFrame(np.zeros((4, 4)),
.....: columns=list(range(4)),
.....: index=[str(i) for i in range(4)])
.....:
In [164]: si
0 1 2 3
0 0 0 0 0
1 0 0 0 0
2 0 0 0 0
3 0 0 0 0
[4 rows x 4 columns]
In [165]: si.index
Index([u'0', u'1', u'2', u'3'], dtype='object')
In [166]: si.columns
Int64Index([0, 1, 2, 3], dtype='int64')
In [167]: json = si.to_json()
In [168]: sij = pd.read_json(json, convert_axes=False)
In [169]: sij
0 1 2 3
0 0 0 0 0
1 0 0 0 0
2 0 0 0 0
3 0 0 0 0
[4 rows x 4 columns]
In [170]: sij.index
Index([u'0', u'1', u'2', u'3'], dtype='object')
In [171]: sij.columns
Index([u'0', u'1', u'2', u'3'], dtype='object')
Dates written in nanoseconds need to be read back in nanoseconds:
In [172]: json = dfj2.to_json(date_unit='ns')
# Try to parse timestamps as millseconds -> Won't Work
In [173]: dfju = pd.read_json(json, date_unit='ms')
In [174]: dfju
A B bools date ints
1356998400000000000 -1.294524 0.413738 True 1356998400000000000 0
1357084800000000000 0.276662 -0.472035 True 1356998400000000000 1
1357171200000000000 -0.013960 -0.362543 True 1356998400000000000 2
1357257600000000000 -0.006154 -0.923061 True 1356998400000000000 3
1357344000000000000 0.895717 0.805244 True 1356998400000000000 4
[5 rows x 5 columns]
# Let Pandas detect the correct precision
In [175]: dfju = pd.read_json(json)
In [176]: dfju
A B bools date ints
2013-01-01 -1.294524 0.413738 True 2013-01-01 00:00:00 0
2013-01-02 0.276662 -0.472035 True 2013-01-01 00:00:00 1
2013-01-03 -0.013960 -0.362543 True 2013-01-01 00:00:00 2
2013-01-04 -0.006154 -0.923061 True 2013-01-01 00:00:00 3
2013-01-05 0.895717 0.805244 True 2013-01-01 00:00:00 4
[5 rows x 5 columns]
# Or specify that all timestamps are in nanoseconds
In [177]: dfju = pd.read_json(json, date_unit='ns')
In [178]: dfju
A B bools date ints
2013-01-01 -1.294524 0.413738 True 2013-01-01 00:00:00 0
2013-01-02 0.276662 -0.472035 True 2013-01-01 00:00:00 1
2013-01-03 -0.013960 -0.362543 True 2013-01-01 00:00:00 2
2013-01-04 -0.006154 -0.923061 True 2013-01-01 00:00:00 3
2013-01-05 0.895717 0.805244 True 2013-01-01 00:00:00 4
[5 rows x 5 columns]
The Numpy Parameter¶
Note
This supports numeric data only. Index and columns labels may be non-numeric, e.g. strings, dates etc.
If numpy=True is passed to read_json an attempt will be made to sniff an appropriate dtype during deserialisation and to subsequently decode directly to numpy arrays, bypassing the need for intermediate Python objects.
This can provide speedups if you are deserialising a large amount of numeric data:
In [179]: randfloats = np.random.uniform(-100, 1000, 10000)
In [180]: randfloats.shape = (1000, 10)
In [181]: dffloats = DataFrame(randfloats, columns=list('ABCDEFGHIJ'))
In [182]: jsonfloats = dffloats.to_json()
In [183]: timeit read_json(jsonfloats)
100 loops, best of 3: 9.72 ms per loop
In [184]: timeit read_json(jsonfloats, numpy=True)
100 loops, best of 3: 5.28 ms per loop
The speedup is less noticable for smaller datasets:
In [185]: jsonfloats = dffloats.head(100).to_json()
In [186]: timeit read_json(jsonfloats)
100 loops, best of 3: 3.76 ms per loop
In [187]: timeit read_json(jsonfloats, numpy=True)
100 loops, best of 3: 2.84 ms per loop
Warning
Direct numpy decoding makes a number of assumptions and may fail or produce unexpected output if these assumptions are not satisfied:
- data is numeric.
- data is uniform. The dtype is sniffed from the first value decoded. A ValueError may be raised, or incorrect output may be produced if this condition is not satisfied.
- labels are ordered. Labels are only read from the first container, it is assumed that each subsequent row / column has been encoded in the same order. This should be satisfied if the data was encoded using to_json but may not be the case if the JSON is from another source.
Normalization¶
New in version 0.13.0.
Pandas provides a utility function to take a dict or list of dicts and normalize this semi-structured data into a flat table.
In [188]: from pandas.io.json import json_normalize
In [189]: data = [{'state': 'Florida',
.....: 'shortname': 'FL',
.....: 'info': {
.....: 'governor': 'Rick Scott'
.....: },
.....: 'counties': [{'name': 'Dade', 'population': 12345},
.....: {'name': 'Broward', 'population': 40000},
.....: {'name': 'Palm Beach', 'population': 60000}]},
.....: {'state': 'Ohio',
.....: 'shortname': 'OH',
.....: 'info': {
.....: 'governor': 'John Kasich'
.....: },
.....: 'counties': [{'name': 'Summit', 'population': 1234},
.....: {'name': 'Cuyahoga', 'population': 1337}]}]
.....:
In [190]: json_normalize(data, 'counties', ['state', 'shortname', ['info', 'governor']])
name population info.governor state shortname
0 Dade 12345 Rick Scott Florida FL
1 Broward 40000 Rick Scott Florida FL
2 Palm Beach 60000 Rick Scott Florida FL
3 Summit 1234 John Kasich Ohio OH
4 Cuyahoga 1337 John Kasich Ohio OH
[5 rows x 5 columns]
HTML¶
Reading HTML Content¶
Warning
We highly encourage you to read the HTML parsing gotchas regarding the issues surrounding the BeautifulSoup4/html5lib/lxml parsers.
New in version 0.12.0.
The top-level read_html() function can accept an HTML string/file/url and will parse HTML tables into list of pandas DataFrames. Let’s look at a few examples.
Note
read_html returns a list of DataFrame objects, even if there is only a single table contained in the HTML content
Read a URL with no options
In [191]: url = 'http://www.fdic.gov/bank/individual/failed/banklist.html'
In [192]: dfs = read_html(url)
In [193]: dfs
[ Bank Name City ST \
0 Syringa Bank Boise ID
1 The Bank of Union El Reno OK
2 DuPage National Bank West Chicago IL
3 Texas Community Bank, National Association The Woodlands TX
4 Bank of Jackson County Graceville FL
5 First National Bank also operating as The Nat... Edinburg TX
6 The Community's Bank Bridgeport CT
7 Sunrise Bank of Arizona Phoenix AZ
8 Community South Bank Parsons TN
9 Bank of Wausau Wausau WI
10 First Community Bank of Southwest Florida (als... Fort Myers FL
11 Mountain National Bank Sevierville TN
12 1st Commerce Bank North Las Vegas NV
13 Banks of Wisconsin d/b/a Bank of Kenosha Kenosha WI
14 Central Arizona Bank Scottsdale AZ
... ... ...
CERT Acquiring Institution Closing Date \
0 34296 Sunwest Bank 2014-01-31 00:00:00
1 17967 BancFirst 2014-01-24 00:00:00
2 5732 Republic Bank of Chicago 2014-01-17 00:00:00
3 57431 Spirit of Texas Bank, SSB 2013-12-13 00:00:00
4 14794 First Federal Bank of Florida 2013-10-30 00:00:00
5 14318 PlainsCapital Bank 2013-09-13 00:00:00
6 57041 No Acquirer 2013-09-13 00:00:00
7 34707 First Fidelity Bank, National Association 2013-08-23 00:00:00
8 19849 CB&S Bank, Inc. 2013-08-23 00:00:00
9 35016 Nicolet National Bank 2013-08-09 00:00:00
10 34943 C1 Bank 2013-08-02 00:00:00
11 34789 First Tennessee Bank, National Association 2013-06-07 00:00:00
12 58358 Plaza Bank 2013-06-06 00:00:00
13 35386 North Shore Bank, FSB 2013-05-31 00:00:00
14 34527 Western State Bank 2013-05-14 00:00:00
... ... ...
Updated Date Loss Share Type
0 2014-01-31 00:00:00 NaN
1 2014-01-28 00:00:00 NaN
2 2014-01-27 00:00:00 NaN
3 2014-01-13 00:00:00 NaN
4 2013-12-09 00:00:00 none
5 2013-11-01 00:00:00 SFR/NSF
6 2013-12-20 00:00:00 none
7 2013-11-01 00:00:00 none
8 2013-11-01 00:00:00 none
9 2013-10-24 00:00:00 none
10 2013-10-24 00:00:00 none
11 2013-07-12 00:00:00 none
12 2013-07-12 00:00:00 NSF
13 2013-10-29 00:00:00 none
14 2013-07-12 00:00:00 none
... ...
[519 rows x 8 columns]]
Note
The data from the above URL changes every Monday so the resulting data above and the data below may be slightly different.
Read in the content of the file from the above URL and pass it to read_html as a string
In [194]: with open(file_path, 'r') as f:
.....: dfs = read_html(f.read())
.....:
In [195]: dfs
[ Bank Name City ST CERT \
0 Banks of Wisconsin d/b/a Bank of Kenosha Kenosha WI 35386
1 Central Arizona Bank Scottsdale AZ 34527
2 Sunrise Bank Valdosta GA 58185
3 Pisgah Community Bank Asheville NC 58701
4 Douglas County Bank Douglasville GA 21649
5 Parkway Bank Lenoir NC 57158
6 Chipola Community Bank Marianna FL 58034
7 Heritage Bank of North Florida Orange Park FL 26680
8 First Federal Bank Lexington KY 29594
9 Gold Canyon Bank Gold Canyon AZ 58066
10 Frontier Bank LaGrange GA 16431
11 Covenant Bank Chicago IL 22476
12 1st Regents Bank Andover MN 57157
13 Westside Community Bank University Place WA 33997
14 Community Bank of the Ozarks Sunrise Beach MO 27331
... ... ... ...
Acquiring Institution Closing Date \
0 North Shore Bank, FSB 2013-05-31 00:00:00
1 Western State Bank 2013-05-14 00:00:00
2 Synovus Bank 2013-05-10 00:00:00
3 Capital Bank, N.A. 2013-05-10 00:00:00
4 Hamilton State Bank 2013-04-26 00:00:00
5 CertusBank, National Association 2013-04-26 00:00:00
6 First Federal Bank of Florida 2013-04-19 00:00:00
7 FirstAtlantic Bank 2013-04-19 00:00:00
8 Your Community Bank 2013-04-19 00:00:00
9 First Scottsdale Bank, National Association 2013-04-05 00:00:00
10 HeritageBank of the South 2013-03-08 00:00:00
11 Liberty Bank and Trust Company 2013-02-15 00:00:00
12 First Minnesota Bank 2013-01-18 00:00:00
13 Sunwest Bank 2013-01-11 00:00:00
14 Bank of Sullivan 2012-12-14 00:00:00
... ...
Updated Date
0 2013-05-31 00:00:00
1 2013-05-20 00:00:00
2 2013-05-21 00:00:00
3 2013-05-14 00:00:00
4 2013-05-16 00:00:00
5 2013-05-17 00:00:00
6 2013-05-16 00:00:00
7 2013-05-16 00:00:00
8 2013-04-23 00:00:00
9 2013-04-09 00:00:00
10 2013-03-26 00:00:00
11 2013-03-04 00:00:00
12 2013-02-28 00:00:00
13 2013-01-24 00:00:00
14 2013-01-24 00:00:00
...
[506 rows x 7 columns]]
You can even pass in an instance of StringIO if you so desire
In [196]: from cStringIO import StringIO
In [197]: with open(file_path, 'r') as f:
.....: sio = StringIO(f.read())
.....:
In [198]: dfs = read_html(sio)
In [199]: dfs
[ Bank Name City ST CERT \
0 Banks of Wisconsin d/b/a Bank of Kenosha Kenosha WI 35386
1 Central Arizona Bank Scottsdale AZ 34527
2 Sunrise Bank Valdosta GA 58185
3 Pisgah Community Bank Asheville NC 58701
4 Douglas County Bank Douglasville GA 21649
5 Parkway Bank Lenoir NC 57158
6 Chipola Community Bank Marianna FL 58034
7 Heritage Bank of North Florida Orange Park FL 26680
8 First Federal Bank Lexington KY 29594
9 Gold Canyon Bank Gold Canyon AZ 58066
10 Frontier Bank LaGrange GA 16431
11 Covenant Bank Chicago IL 22476
12 1st Regents Bank Andover MN 57157
13 Westside Community Bank University Place WA 33997
14 Community Bank of the Ozarks Sunrise Beach MO 27331
... ... ... ...
Acquiring Institution Closing Date \
0 North Shore Bank, FSB 2013-05-31 00:00:00
1 Western State Bank 2013-05-14 00:00:00
2 Synovus Bank 2013-05-10 00:00:00
3 Capital Bank, N.A. 2013-05-10 00:00:00
4 Hamilton State Bank 2013-04-26 00:00:00
5 CertusBank, National Association 2013-04-26 00:00:00
6 First Federal Bank of Florida 2013-04-19 00:00:00
7 FirstAtlantic Bank 2013-04-19 00:00:00
8 Your Community Bank 2013-04-19 00:00:00
9 First Scottsdale Bank, National Association 2013-04-05 00:00:00
10 HeritageBank of the South 2013-03-08 00:00:00
11 Liberty Bank and Trust Company 2013-02-15 00:00:00
12 First Minnesota Bank 2013-01-18 00:00:00
13 Sunwest Bank 2013-01-11 00:00:00
14 Bank of Sullivan 2012-12-14 00:00:00
... ...
Updated Date
0 2013-05-31 00:00:00
1 2013-05-20 00:00:00
2 2013-05-21 00:00:00
3 2013-05-14 00:00:00
4 2013-05-16 00:00:00
5 2013-05-17 00:00:00
6 2013-05-16 00:00:00
7 2013-05-16 00:00:00
8 2013-04-23 00:00:00
9 2013-04-09 00:00:00
10 2013-03-26 00:00:00
11 2013-03-04 00:00:00
12 2013-02-28 00:00:00
13 2013-01-24 00:00:00
14 2013-01-24 00:00:00
...
[506 rows x 7 columns]]
Note
The following examples are not run by the IPython evaluator due to the fact that having so many network-accessing functions slows down the documentation build. If you spot an error or an example that doesn’t run, please do not hesitate to report it over on pandas GitHub issues page.
Read a URL and match a table that contains specific text
match = 'Metcalf Bank'
df_list = read_html(url, match=match)
Specify a header row (by default <th> elements are used to form the column index); if specified, the header row is taken from the data minus the parsed header elements (<th> elements).
dfs = read_html(url, header=0)
Specify an index column
dfs = read_html(url, index_col=0)
Specify a number of rows to skip
dfs = read_html(url, skiprows=0)
Specify a number of rows to skip using a list (xrange (Python 2 only) works as well)
dfs = read_html(url, skiprows=range(2))
Don’t infer numeric and date types
dfs = read_html(url, infer_types=False)
Specify an HTML attribute
dfs1 = read_html(url, attrs={'id': 'table'})
dfs2 = read_html(url, attrs={'class': 'sortable'})
print(np.array_equal(dfs1[0], dfs2[0])) # Should be True
Use some combination of the above
dfs = read_html(url, match='Metcalf Bank', index_col=0)
Read in pandas to_html output (with some loss of floating point precision)
df = DataFrame(randn(2, 2))
s = df.to_html(float_format='{0:.40g}'.format)
dfin = read_html(s, index_col=0)
The lxml backend will raise an error on a failed parse if that is the only parser you provide (if you only have a single parser you can provide just a string, but it is considered good practice to pass a list with one string if, for example, the function expects a sequence of strings)
dfs = read_html(url, 'Metcalf Bank', index_col=0, flavor=['lxml'])
or
dfs = read_html(url, 'Metcalf Bank', index_col=0, flavor='lxml')
However, if you have bs4 and html5lib installed and pass None or ['lxml', 'bs4'] then the parse will most likely succeed. Note that as soon as a parse succeeds, the function will return.
dfs = read_html(url, 'Metcalf Bank', index_col=0, flavor=['lxml', 'bs4'])
Writing to HTML files¶
DataFrame objects have an instance method to_html which renders the contents of the DataFrame as an HTML table. The function arguments are as in the method to_string described above.
Note
Not all of the possible options for DataFrame.to_html are shown here for brevity’s sake. See to_html() for the full set of options.
In [200]: df = DataFrame(randn(2, 2))
In [201]: df
0 1
0 -0.184744 0.496971
1 -0.856240 1.857977
[2 rows x 2 columns]
In [202]: print(df.to_html()) # raw html
<table border="1" class="dataframe">
<thead>
<tr style="text-align: right;">
<th></th>
<th>0</th>
<th>1</th>
</tr>
</thead>
<tbody>
<tr>
<th>0</th>
<td>-0.184744</td>
<td> 0.496971</td>
</tr>
<tr>
<th>1</th>
<td>-0.856240</td>
<td> 1.857977</td>
</tr>
</tbody>
</table>
HTML:
| 0 | 1 | |
|---|---|---|
| 0 | -0.184744 | 0.496971 |
| 1 | -0.856240 | 1.857977 |
The columns argument will limit the columns shown
In [203]: print(df.to_html(columns=[0]))
<table border="1" class="dataframe">
<thead>
<tr style="text-align: right;">
<th></th>
<th>0</th>
</tr>
</thead>
<tbody>
<tr>
<th>0</th>
<td>-0.184744</td>
</tr>
<tr>
<th>1</th>
<td>-0.856240</td>
</tr>
</tbody>
</table>
HTML:
| 0 | |
|---|---|
| 0 | -0.184744 |
| 1 | -0.856240 |
float_format takes a Python callable to control the precision of floating point values
In [204]: print(df.to_html(float_format='{0:.10f}'.format))
<table border="1" class="dataframe">
<thead>
<tr style="text-align: right;">
<th></th>
<th>0</th>
<th>1</th>
</tr>
</thead>
<tbody>
<tr>
<th>0</th>
<td>-0.1847438576</td>
<td>0.4969711327</td>
</tr>
<tr>
<th>1</th>
<td>-0.8562396763</td>
<td>1.8579766508</td>
</tr>
</tbody>
</table>
HTML:
| 0 | 1 | |
|---|---|---|
| 0 | -0.1847438576 | 0.4969711327 |
| 1 | -0.8562396763 | 1.8579766508 |
bold_rows will make the row labels bold by default, but you can turn that off
In [205]: print(df.to_html(bold_rows=False))
<table border="1" class="dataframe">
<thead>
<tr style="text-align: right;">
<th></th>
<th>0</th>
<th>1</th>
</tr>
</thead>
<tbody>
<tr>
<td>0</td>
<td>-0.184744</td>
<td> 0.496971</td>
</tr>
<tr>
<td>1</td>
<td>-0.856240</td>
<td> 1.857977</td>
</tr>
</tbody>
</table>
| 0 | 1 | |
|---|---|---|
| 0 | -0.184744 | 0.496971 |
| 1 | -0.856240 | 1.857977 |
The classes argument provides the ability to give the resulting HTML table CSS classes. Note that these classes are appended to the existing 'dataframe' class.
In [206]: print(df.to_html(classes=['awesome_table_class', 'even_more_awesome_class']))
<table border="1" class="dataframe awesome_table_class even_more_awesome_class">
<thead>
<tr style="text-align: right;">
<th></th>
<th>0</th>
<th>1</th>
</tr>
</thead>
<tbody>
<tr>
<th>0</th>
<td>-0.184744</td>
<td> 0.496971</td>
</tr>
<tr>
<th>1</th>
<td>-0.856240</td>
<td> 1.857977</td>
</tr>
</tbody>
</table>
Finally, the escape argument allows you to control whether the “<”, “>” and “&” characters escaped in the resulting HTML (by default it is True). So to get the HTML without escaped characters pass escape=False
In [207]: df = DataFrame({'a': list('&<>'), 'b': randn(3)})
Escaped:
In [208]: print(df.to_html())
<table border="1" class="dataframe">
<thead>
<tr style="text-align: right;">
<th></th>
<th>a</th>
<th>b</th>
</tr>
</thead>
<tbody>
<tr>
<th>0</th>
<td> &</td>
<td>-0.474063</td>
</tr>
<tr>
<th>1</th>
<td> <</td>
<td>-0.230305</td>
</tr>
<tr>
<th>2</th>
<td> ></td>
<td>-0.400654</td>
</tr>
</tbody>
</table>
| a | b | |
|---|---|---|
| 0 | & | -0.474063 |
| 1 | < | -0.230305 |
| 2 | > | -0.400654 |
Not escaped:
In [209]: print(df.to_html(escape=False))
<table border="1" class="dataframe">
<thead>
<tr style="text-align: right;">
<th></th>
<th>a</th>
<th>b</th>
</tr>
</thead>
<tbody>
<tr>
<th>0</th>
<td> &</td>
<td>-0.474063</td>
</tr>
<tr>
<th>1</th>
<td> <</td>
<td>-0.230305</td>
</tr>
<tr>
<th>2</th>
<td> ></td>
<td>-0.400654</td>
</tr>
</tbody>
</table>
| a | b | |
|---|---|---|
| 0 | & | -0.474063 |
| 1 | < | -0.230305 |
| 2 | > | -0.400654 |
Note
Some browsers may not show a difference in the rendering of the previous two HTML tables.
Excel files¶
The read_excel method can read Excel 2003 (.xls) and Excel 2007 (.xlsx) files using the xlrd Python module and use the same parsing code as the above to convert tabular data into a DataFrame. See the cookbook for some advanced strategies
Note
The prior method of accessing Excel is now deprecated as of 0.12.0, this will work but will be removed in a future version.
from pandas.io.parsers import ExcelFile xls = ExcelFile('path_to_file.xls') xls.parse('Sheet1', index_col=None, na_values=['NA'])
Replaced by
read_excel('path_to_file.xls', 'Sheet1', index_col=None, na_values=['NA'])
New in version 0.13.
There are now two ways to read in sheets from an Excel file. You can provide either the index of a sheet or its name. If the value provided is an integer then it is assumed that the integer refers to the index of a sheet, otherwise if a string is passed then it is assumed that the string refers to the name of a particular sheet in the file.
Using the sheet name:
read_excel('path_to_file.xls', 'Sheet1', index_col=None, na_values=['NA'])
Using the sheet index:
read_excel('path_to_file.xls', 0, index_col=None, na_values=['NA'])
It is often the case that users will insert columns to do temporary computations in Excel and you may not want to read in those columns. read_excel takes a parse_cols keyword to allow you to specify a subset of columns to parse.
If parse_cols is an integer, then it is assumed to indicate the last column to be parsed.
read_excel('path_to_file.xls', 'Sheet1', parse_cols=2, index_col=None, na_values=['NA'])
If parse_cols is a list of integers, then it is assumed to be the file column indices to be parsed.
read_excel('path_to_file.xls', 'Sheet1', parse_cols=[0, 2, 3], index_col=None, na_values=['NA'])
To write a DataFrame object to a sheet of an Excel file, you can use the to_excel instance method. The arguments are largely the same as to_csv described above, the first argument being the name of the excel file, and the optional second argument the name of the sheet to which the DataFrame should be written. For example:
df.to_excel('path_to_file.xlsx', sheet_name='Sheet1')
Files with a .xls extension will be written using xlwt and those with a .xlsx extension will be written using xlsxwriter (if available) or openpyxl.
The Panel class also has a to_excel instance method, which writes each DataFrame in the Panel to a separate sheet.
In order to write separate DataFrames to separate sheets in a single Excel file, one can pass an ExcelWriter.
with ExcelWriter('path_to_file.xlsx') as writer:
df1.to_excel(writer, sheet_name='Sheet1')
df2.to_excel(writer, sheet_name='Sheet2')
Note
Wringing a little more performance out of read_excel Internally, Excel stores all numeric data as floats. Because this can produce unexpected behavior when reading in data, pandas defaults to trying to convert integers to floats if it doesn’t lose information (1.0 --> 1). You can pass convert_float=False to disable this behavior, which may give a slight performance improvement.
Excel writer engines¶
New in version 0.13.
pandas chooses an Excel writer via two methods:
- the engine keyword argument
- the filename extension (via the default specified in config options)
By default, pandas uses the XlsxWriter for .xlsx and openpyxl for .xlsm files and xlwt for .xls files. If you have multiple engines installed, you can set the default engine through setting the config options io.excel.xlsx.writer and io.excel.xls.writer. pandas will fall back on openpyxl for .xlsx files if Xlsxwriter is not available.
To specify which writer you want to use, you can pass an engine keyword argument to to_excel and to ExcelWriter.
# By setting the 'engine' in the DataFrame and Panel 'to_excel()' methods.
df.to_excel('path_to_file.xlsx', sheet_name='Sheet1', engine='xlsxwriter')
# By setting the 'engine' in the ExcelWriter constructor.
writer = ExcelWriter('path_to_file.xlsx', engine='xlsxwriter')
# Or via pandas configuration.
from pandas import options
options.io.excel.xlsx.writer = 'xlsxwriter'
df.to_excel('path_to_file.xlsx', sheet_name='Sheet1')
Clipboard¶
A handy way to grab data is to use the read_clipboard method, which takes the contents of the clipboard buffer and passes them to the read_table method. For instance, you can copy the following text to the clipboard (CTRL-C on many operating systems):
A B C
x 1 4 p
y 2 5 q
z 3 6 r
And then import the data directly to a DataFrame by calling:
clipdf = pd.read_clipboard()
In [210]: clipdf
A B C
x 1 4 p
y 2 5 q
z 3 6 r
[3 rows x 3 columns]
The to_clipboard method can be used to write the contents of a DataFrame to the clipboard. Following which you can paste the clipboard contents into other applications (CTRL-V on many operating systems). Here we illustrate writing a DataFrame into clipboard and reading it back.
In [211]: df=pd.DataFrame(randn(5,3))
In [212]: df
0 1 2
0 -0.288267 -0.084905 0.004772
1 1.382989 0.343635 -1.253994
2 -0.124925 0.212244 0.496654
3 0.525417 1.238640 -1.210543
4 -1.175743 -0.172372 -0.734129
[5 rows x 3 columns]
In [213]: df.to_clipboard()
In [214]: pd.read_clipboard()
0 1 2
0 -0.288267 -0.084905 0.004772
1 1.382989 0.343635 -1.253994
2 -0.124925 0.212244 0.496654
3 0.525417 1.238640 -1.210543
4 -1.175743 -0.172372 -0.734129
[5 rows x 3 columns]
We can see that we got the same content back, which we had earlier written to the clipboard.
Note
You may need to install xclip or xsel (with gtk or PyQt4 modules) on Linux to use these methods.
Pickling¶
All pandas objects are equipped with to_pickle methods which use Python’s cPickle module to save data structures to disk using the pickle format.
In [215]: df
0 1 2
0 -0.288267 -0.084905 0.004772
1 1.382989 0.343635 -1.253994
2 -0.124925 0.212244 0.496654
3 0.525417 1.238640 -1.210543
4 -1.175743 -0.172372 -0.734129
[5 rows x 3 columns]
In [216]: df.to_pickle('foo.pkl')
The read_pickle function in the pandas namespace can be used to load any pickled pandas object (or any other pickled object) from file:
In [217]: read_pickle('foo.pkl')
0 1 2
0 -0.288267 -0.084905 0.004772
1 1.382989 0.343635 -1.253994
2 -0.124925 0.212244 0.496654
3 0.525417 1.238640 -1.210543
4 -1.175743 -0.172372 -0.734129
[5 rows x 3 columns]
Warning
Loading pickled data received from untrusted sources can be unsafe.
Warning
In 0.13, pickle preserves compatibility with pickles created prior to 0.13. These must be read with pd.read_pickle, rather than the default python pickle.load. See this question for a detailed explanation.
Note
These methods were previously pd.save and pd.load, prior to 0.12.0, and are now deprecated.
msgpack (experimental)¶
New in version 0.13.0.
Starting in 0.13.0, pandas is supporting the msgpack format for object serialization. This is a lightweight portable binary format, similar to binary JSON, that is highly space efficient, and provides good performance both on the writing (serialization), and reading (deserialization).
Warning
This is a very new feature of pandas. We intend to provide certain optimizations in the io of the msgpack data. Since this is marked as an EXPERIMENTAL LIBRARY, the storage format may not be stable until a future release.
In [218]: df = DataFrame(np.random.rand(5,2),columns=list('AB'))
In [219]: df.to_msgpack('foo.msg')
In [220]: pd.read_msgpack('foo.msg')
A B
0 0.154336 0.710999
1 0.398096 0.765220
2 0.586749 0.293052
3 0.290293 0.710783
4 0.988593 0.062106
[5 rows x 2 columns]
In [221]: s = Series(np.random.rand(5),index=date_range('20130101',periods=5))
You can pass a list of objects and you will receive them back on deserialization.
In [222]: pd.to_msgpack('foo.msg', df, 'foo', np.array([1,2,3]), s)
In [223]: pd.read_msgpack('foo.msg')
[ A B
0 0.154336 0.710999
1 0.398096 0.765220
2 0.586749 0.293052
3 0.290293 0.710783
4 0.988593 0.062106
[5 rows x 2 columns],
u'foo',
array([1, 2, 3]),
2013-01-01 0.690810
2013-01-02 0.235907
2013-01-03 0.712756
2013-01-04 0.119599
2013-01-05 0.023493
Freq: D, dtype: float64]
You can pass iterator=True to iterate over the unpacked results
In [224]: for o in pd.read_msgpack('foo.msg',iterator=True):
.....: print o
.....:
A B
0 0.154336 0.710999
1 0.398096 0.765220
2 0.586749 0.293052
3 0.290293 0.710783
4 0.988593 0.062106
[5 rows x 2 columns]
foo
[1 2 3]
2013-01-01 0.690810
2013-01-02 0.235907
2013-01-03 0.712756
2013-01-04 0.119599
2013-01-05 0.023493
Freq: D, dtype: float64
You can pass append=True to the writer to append to an existing pack
In [225]: df.to_msgpack('foo.msg',append=True)
In [226]: pd.read_msgpack('foo.msg')
[ A B
0 0.154336 0.710999
1 0.398096 0.765220
2 0.586749 0.293052
3 0.290293 0.710783
4 0.988593 0.062106
[5 rows x 2 columns],
u'foo',
array([1, 2, 3]),
2013-01-01 0.690810
2013-01-02 0.235907
2013-01-03 0.712756
2013-01-04 0.119599
2013-01-05 0.023493
Freq: D, dtype: float64,
A B
0 0.154336 0.710999
1 0.398096 0.765220
2 0.586749 0.293052
3 0.290293 0.710783
4 0.988593 0.062106
[5 rows x 2 columns]]
Unlike other io methods, to_msgpack is available on both a per-object basis, df.to_msgpack() and using the top-level pd.to_msgpack(...) where you can pack arbitrary collections of python lists, dicts, scalars, while intermixing pandas objects.
In [227]: pd.to_msgpack('foo2.msg', { 'dict' : [ { 'df' : df }, { 'string' : 'foo' }, { 'scalar' : 1. }, { 's' : s } ] })
In [228]: pd.read_msgpack('foo2.msg')
{u'dict': ({u'df': A B
0 0.154336 0.710999
1 0.398096 0.765220
2 0.586749 0.293052
3 0.290293 0.710783
4 0.988593 0.062106
[5 rows x 2 columns]},
{u'string': u'foo'},
{u'scalar': 1.0},
{u's': 2013-01-01 0.690810
2013-01-02 0.235907
2013-01-03 0.712756
2013-01-04 0.119599
2013-01-05 0.023493
Freq: D, dtype: float64})}
Read/Write API¶
Msgpacks can also be read from and written to strings.
In [229]: df.to_msgpack()
'\x84\xa6blocks\x91\x86\xa5items\x85\xa5dtype\x11\xa3typ\xa5index\xa5klass\xa5Index\xa4data\x92\xa1A\xa1B\xa4name\xc0\xa8compress\xc0\xa5shape\x92\x02\x05\xa6values\xda\x00P\xa0\xab\xfb6H\xc1\xc3?\x98(oMgz\xd9?\x17\xaed\\\xa5\xc6\xe2?\xdc\xd0\x1bd(\x94\xd2?\xb5\xe8\xf5\x0e\x8d\xa2\xef?\x02D\xebO\x80\xc0\xe6?\x16\xbddQ\xae|\xe8?\x10?Ya[\xc1\xd2?\xa8\xfd\xcf\xa0\xbc\xbe\xe6? Z\xe1\ti\xcc\xaf?\xa5klass\xaaFloatBlock\xa5dtype\x0c\xa4axes\x92\x85\xa5dtype\x11\xa3typ\xa5index\xa5klass\xa5Index\xa4data\x92\xa1A\xa1B\xa4name\xc0\x85\xa5dtype\x07\xa3typ\xa5index\xa5klass\xaaInt64Index\xa4data\xda\x00(\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x02\x00\x00\x00\x00\x00\x00\x00\x03\x00\x00\x00\x00\x00\x00\x00\x04\x00\x00\x00\x00\x00\x00\x00\xa4name\xc0\xa3typ\xadblock_manager\xa5klass\xa9DataFrame'
Furthermore you can concatenate the strings to produce a list of the original objects.
In [230]: pd.read_msgpack(df.to_msgpack() + s.to_msgpack())
[ A B
0 0.154336 0.710999
1 0.398096 0.765220
2 0.586749 0.293052
3 0.290293 0.710783
4 0.988593 0.062106
[5 rows x 2 columns],
2013-01-01 0.690810
2013-01-02 0.235907
2013-01-03 0.712756
2013-01-04 0.119599
2013-01-05 0.023493
Freq: D, dtype: float64]
HDF5 (PyTables)¶
HDFStore is a dict-like object which reads and writes pandas using the high performance HDF5 format using the excellent PyTables library. See the cookbook for some advanced strategies
Note
PyTables 3.0.0 was recently released to enable support for Python 3. Pandas should be fully compatible (and previously written stores should be backwards compatible) with all PyTables >= 2.3. For python >= 3.2, pandas >= 0.12.0 is required for compatibility.
In [231]: store = HDFStore('store.h5')
In [232]: print(store)
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
Empty
Objects can be written to the file just like adding key-value pairs to a dict:
In [233]: np.random.seed(1234)
In [234]: index = date_range('1/1/2000', periods=8)
In [235]: s = Series(randn(5), index=['a', 'b', 'c', 'd', 'e'])
In [236]: df = DataFrame(randn(8, 3), index=index,
.....: columns=['A', 'B', 'C'])
.....:
In [237]: wp = Panel(randn(2, 5, 4), items=['Item1', 'Item2'],
.....: major_axis=date_range('1/1/2000', periods=5),
.....: minor_axis=['A', 'B', 'C', 'D'])
.....:
# store.put('s', s) is an equivalent method
In [238]: store['s'] = s
In [239]: store['df'] = df
In [240]: store['wp'] = wp
# the type of stored data
In [241]: store.root.wp._v_attrs.pandas_type
'wide'
In [242]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame (shape->[8,3])
/s series (shape->[5])
/wp wide (shape->[2,5,4])
In a current or later Python session, you can retrieve stored objects:
# store.get('df') is an equivalent method
In [243]: store['df']
A B C
2000-01-01 0.887163 0.859588 -0.636524
2000-01-02 0.015696 -2.242685 1.150036
2000-01-03 0.991946 0.953324 -2.021255
2000-01-04 -0.334077 0.002118 0.405453
2000-01-05 0.289092 1.321158 -1.546906
2000-01-06 -0.202646 -0.655969 0.193421
2000-01-07 0.553439 1.318152 -0.469305
2000-01-08 0.675554 -1.817027 -0.183109
[8 rows x 3 columns]
# dotted (attribute) access provides get as well
In [244]: store.df
A B C
2000-01-01 0.887163 0.859588 -0.636524
2000-01-02 0.015696 -2.242685 1.150036
2000-01-03 0.991946 0.953324 -2.021255
2000-01-04 -0.334077 0.002118 0.405453
2000-01-05 0.289092 1.321158 -1.546906
2000-01-06 -0.202646 -0.655969 0.193421
2000-01-07 0.553439 1.318152 -0.469305
2000-01-08 0.675554 -1.817027 -0.183109
[8 rows x 3 columns]
Deletion of the object specified by the key
# store.remove('wp') is an equivalent method
In [245]: del store['wp']
In [246]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame (shape->[8,3])
/s series (shape->[5])
Closing a Store, Context Manager
In [247]: store.close()
In [248]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
File is CLOSED
In [249]: store.is_open
False
# Working with, and automatically closing the store with the context
# manager
In [250]: with get_store('store.h5') as store:
.....: store.keys()
.....:
Read/Write API¶
HDFStore supports an top-level API using read_hdf for reading and to_hdf for writing, similar to how read_csv and to_csv work. (new in 0.11.0)
In [251]: df_tl = DataFrame(dict(A=list(range(5)), B=list(range(5))))
In [252]: df_tl.to_hdf('store_tl.h5','table',append=True)
In [253]: read_hdf('store_tl.h5', 'table', where = ['index>2'])
A B
3 3 3
4 4 4
[2 rows x 2 columns]
Fixed Format¶
Note
This was prior to 0.13.0 the Storer format.
The examples above show storing using put, which write the HDF5 to PyTables in a fixed array format, called the fixed format. These types of stores are are not appendable once written (though you can simply remove them and rewrite). Nor are they queryable; they must be retrieved in their entirety. These offer very fast writing and slightly faster reading than table stores. This format is specified by default when using put or to_hdf or by format='fixed' or format='f'
Warning
A fixed format will raise a TypeError if you try to retrieve using a where .
DataFrame(randn(10,2)).to_hdf('test_fixed.h5','df')
pd.read_hdf('test_fixed.h5','df',where='index>5')
TypeError: cannot pass a where specification when reading a fixed format.
this store must be selected in its entirety
Table Format¶
HDFStore supports another PyTables format on disk, the table format. Conceptually a table is shaped very much like a DataFrame, with rows and columns. A table may be appended to in the same or other sessions. In addition, delete & query type operations are supported. This format is specified by format='table' or format='t' to append or put or to_hdf
This format can be set as an option as well pd.set_option('io.hdf.default_format','table') to enable put/append/to_hdf to by default store in the table format.
In [254]: store = HDFStore('store.h5')
In [255]: df1 = df[0:4]
In [256]: df2 = df[4:]
# append data (creates a table automatically)
In [257]: store.append('df', df1)
In [258]: store.append('df', df2)
In [259]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
# select the entire object
In [260]: store.select('df')
A B C
2000-01-01 0.887163 0.859588 -0.636524
2000-01-02 0.015696 -2.242685 1.150036
2000-01-03 0.991946 0.953324 -2.021255
2000-01-04 -0.334077 0.002118 0.405453
2000-01-05 0.289092 1.321158 -1.546906
2000-01-06 -0.202646 -0.655969 0.193421
2000-01-07 0.553439 1.318152 -0.469305
2000-01-08 0.675554 -1.817027 -0.183109
[8 rows x 3 columns]
# the type of stored data
In [261]: store.root.df._v_attrs.pandas_type
'frame_table'
Note
You can also create a table by passing format='table' or format='t' to a put operation.
Hierarchical Keys¶
Keys to a store can be specified as a string. These can be in a hierarchical path-name like format (e.g. foo/bar/bah), which will generate a hierarchy of sub-stores (or Groups in PyTables parlance). Keys can be specified with out the leading ‘/’ and are ALWAYS absolute (e.g. ‘foo’ refers to ‘/foo’). Removal operations can remove everying in the sub-store and BELOW, so be careful.
In [262]: store.put('foo/bar/bah', df)
In [263]: store.append('food/orange', df)
In [264]: store.append('food/apple', df)
In [265]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/food/apple frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/food/orange frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/foo/bar/bah frame (shape->[8,3])
# a list of keys are returned
In [266]: store.keys()
['/df', '/food/apple', '/food/orange', '/foo/bar/bah']
# remove all nodes under this level
In [267]: store.remove('food')
In [268]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/foo/bar/bah frame (shape->[8,3])
Storing Mixed Types in a Table¶
Storing mixed-dtype data is supported. Strings are stored as a fixed-width using the maximum size of the appended column. Subsequent appends will truncate strings at this length.
Passing min_itemsize={`values`: size} as a parameter to append will set a larger minimum for the string columns. Storing floats, strings, ints, bools, datetime64 are currently supported. For string columns, passing nan_rep = 'nan' to append will change the default nan representation on disk (which converts to/from np.nan), this defaults to nan.
In [269]: df_mixed = DataFrame({ 'A' : randn(8),
.....: 'B' : randn(8),
.....: 'C' : np.array(randn(8),dtype='float32'),
.....: 'string' :'string',
.....: 'int' : 1,
.....: 'bool' : True,
.....: 'datetime64' : Timestamp('20010102')},
.....: index=list(range(8)))
.....:
In [270]: df_mixed.ix[3:5,['A', 'B', 'string', 'datetime64']] = np.nan
In [271]: store.append('df_mixed', df_mixed, min_itemsize = {'values': 50})
In [272]: df_mixed1 = store.select('df_mixed')
In [273]: df_mixed1
A B C bool datetime64 int string
0 0.704721 -1.152659 -0.430096 True 2001-01-02 00:00:00 1 string
1 -0.785435 0.631979 0.767369 True 2001-01-02 00:00:00 1 string
2 0.462060 0.039513 0.984920 True 2001-01-02 00:00:00 1 string
3 NaN NaN 0.270836 True NaT 1 NaN
4 NaN NaN 1.391986 True NaT 1 NaN
5 NaN NaN 0.079842 True NaT 1 NaN
6 2.007843 0.152631 -0.399965 True 2001-01-02 00:00:00 1 string
7 0.226963 0.164530 -1.027851 True 2001-01-02 00:00:00 1 string
[8 rows x 7 columns]
In [274]: df_mixed1.get_dtype_counts()
bool 1
datetime64[ns] 1
float32 1
float64 2
int64 1
object 1
dtype: int64
# we have provided a minimum string column size
In [275]: store.root.df_mixed.table
/df_mixed/table (Table(8,)) ''
description := {
"index": Int64Col(shape=(), dflt=0, pos=0),
"values_block_0": Float64Col(shape=(2,), dflt=0.0, pos=1),
"values_block_1": Float32Col(shape=(1,), dflt=0.0, pos=2),
"values_block_2": Int64Col(shape=(1,), dflt=0, pos=3),
"values_block_3": Int64Col(shape=(1,), dflt=0, pos=4),
"values_block_4": BoolCol(shape=(1,), dflt=False, pos=5),
"values_block_5": StringCol(itemsize=50, shape=(1,), dflt='', pos=6)}
byteorder := 'little'
chunkshape := (689,)
autoindex := True
colindexes := {
"index": Index(6, medium, shuffle, zlib(1)).is_csi=False}
Storing Multi-Index DataFrames¶
Storing multi-index dataframes as tables is very similar to storing/selecting from homogeneous index DataFrames.
In [276]: index = MultiIndex(levels=[['foo', 'bar', 'baz', 'qux'],
.....: ['one', 'two', 'three']],
.....: labels=[[0, 0, 0, 1, 1, 2, 2, 3, 3, 3],
.....: [0, 1, 2, 0, 1, 1, 2, 0, 1, 2]],
.....: names=['foo', 'bar'])
.....:
In [277]: df_mi = DataFrame(np.random.randn(10, 3), index=index,
.....: columns=['A', 'B', 'C'])
.....:
In [278]: df_mi
A B C
foo bar
foo one -0.584718 0.816594 -0.081947
two -0.344766 0.528288 -1.068989
three -0.511881 0.291205 0.566534
bar one 0.503592 0.285296 0.484288
two 1.363482 -0.781105 -0.468018
baz two 1.224574 -1.281108 0.875476
three -1.710715 -0.450765 0.749164
qux one -0.203933 -0.182175 0.680656
two -1.818499 0.047072 0.394844
three -0.248432 -0.617707 -0.682884
[10 rows x 3 columns]
In [279]: store.append('df_mi',df_mi)
In [280]: store.select('df_mi')
A B C
foo bar
foo one -0.584718 0.816594 -0.081947
two -0.344766 0.528288 -1.068989
three -0.511881 0.291205 0.566534
bar one 0.503592 0.285296 0.484288
two 1.363482 -0.781105 -0.468018
baz two 1.224574 -1.281108 0.875476
three -1.710715 -0.450765 0.749164
qux one -0.203933 -0.182175 0.680656
two -1.818499 0.047072 0.394844
three -0.248432 -0.617707 -0.682884
[10 rows x 3 columns]
# the levels are automatically included as data columns
In [281]: store.select('df_mi', 'foo=bar')
A B C
foo bar
bar one 0.503592 0.285296 0.484288
two 1.363482 -0.781105 -0.468018
[2 rows x 3 columns]
Querying a Table¶
Warning
This query capabilities have changed substantially starting in 0.13.0. Queries from prior version are accepted (with a DeprecationWarning) printed if its not string-like.
select and delete operations have an optional criterion that can be specified to select/delete only a subset of the data. This allows one to have a very large on-disk table and retrieve only a portion of the data.
A query is specified using the Term class under the hood, as a boolean expression.
- index and columns are supported indexers of a DataFrame
- major_axis, minor_axis, and items are supported indexers of the Panel
- if data_columns are specified, these can be used as additional indexers
Valid comparison operators are:
- =, ==, !=, >, >=, <, <=
Valid boolean expressions are combined with:
- | : or
- & : and
- ( and ) : for grouping
These rules are similar to how boolean expressions are used in pandas for indexing.
Note
- = will be automatically expanded to the comparison operator ==
- ~ is the not operator, but can only be used in very limited circumstances
- If a list/tuple of expressions is passed they will be combined via &
The following are valid expressions:
- 'index>=date'
- "columns=['A', 'D']"
- "columns in ['A', 'D']"
- 'columns=A'
- 'columns==A'
- "~(columns=['A','B'])"
- 'index>df.index[3] & string="bar"'
- '(index>df.index[3] & index<=df.index[6]) | string="bar"'
- "ts>=Timestamp('2012-02-01')"
- "major_axis>=20130101"
The indexers are on the left-hand side of the sub-expression:
- columns, major_axis, ts
The right-hand side of the sub-expression (after a comparsion operator) can be:
- functions that will be evaluated, e.g. Timestamp('2012-02-01')
- strings, e.g. "bar"
- date-like, e.g. 20130101, or "20130101"
- lists, e.g. "['A','B']"
- variables that are defined in the local names space, e.g. date
Here are some examples:
In [282]: dfq = DataFrame(randn(10,4),columns=list('ABCD'),index=date_range('20130101',periods=10))
In [283]: store.append('dfq',dfq,format='table',data_columns=True)
Use boolean expressions, with in-line function evaluation.
In [284]: store.select('dfq',"index>Timestamp('20130104') & columns=['A', 'B']")
A B
2013-01-05 1.210384 0.797435
2013-01-06 -0.850346 1.176812
2013-01-07 0.984188 -0.121728
2013-01-08 0.796595 -0.474021
2013-01-09 -0.804834 -2.123620
2013-01-10 0.334198 0.536784
[6 rows x 2 columns]
Use and inline column reference
In [285]: store.select('dfq',where="A>0 or C>0")
A B C D
2013-01-01 0.436258 -1.703013 0.393711 -0.479324
2013-01-02 -0.299016 0.694103 0.678630 0.239556
2013-01-03 0.151227 0.816127 1.893534 0.639633
2013-01-04 -0.962029 -2.085266 1.930247 -1.735349
2013-01-05 1.210384 0.797435 -0.379811 0.702562
2013-01-07 0.984188 -0.121728 2.365769 0.496143
2013-01-08 0.796595 -0.474021 -0.056696 1.357797
2013-01-10 0.334198 0.536784 -0.743830 -0.320204
[8 rows x 4 columns]
Works with a Panel as well.
In [286]: store.append('wp',wp)
In [287]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/df_mi frame_table (typ->appendable_multi,nrows->10,ncols->5,indexers->[index],dc->[bar,foo])
/df_mixed frame_table (typ->appendable,nrows->8,ncols->7,indexers->[index])
/dfq frame_table (typ->appendable,nrows->10,ncols->4,indexers->[index],dc->[A,B,C,D])
/wp wide_table (typ->appendable,nrows->20,ncols->2,indexers->[major_axis,minor_axis])
/foo/bar/bah frame (shape->[8,3])
In [288]: store.select('wp', "major_axis>Timestamp('20000102') & minor_axis=['A', 'B']")
<class 'pandas.core.panel.Panel'>
Dimensions: 2 (items) x 3 (major_axis) x 2 (minor_axis)
Items axis: Item1 to Item2
Major_axis axis: 2000-01-03 00:00:00 to 2000-01-05 00:00:00
Minor_axis axis: A to B
The columns keyword can be supplied to select a list of columns to be returned, this is equivalent to passing a 'columns=list_of_columns_to_filter':
In [289]: store.select('df', "columns=['A', 'B']")
A B
2000-01-01 0.887163 0.859588
2000-01-02 0.015696 -2.242685
2000-01-03 0.991946 0.953324
2000-01-04 -0.334077 0.002118
2000-01-05 0.289092 1.321158
2000-01-06 -0.202646 -0.655969
2000-01-07 0.553439 1.318152
2000-01-08 0.675554 -1.817027
[8 rows x 2 columns]
start and stop parameters can be specified to limit the total search space. These are in terms of the total number of rows in a table.
# this is effectively what the storage of a Panel looks like
In [290]: wp.to_frame()
Item1 Item2
major minor
2000-01-01 A 1.058969 0.215269
B -0.397840 0.841009
C 0.337438 -1.445810
D 1.047579 -1.401973
2000-01-02 A 1.045938 -0.100918
B 0.863717 -0.548242
C -0.122092 -0.144620
D 0.124713 0.354020
2000-01-03 A -0.322795 -0.035513
B 0.841675 0.565738
C 2.390961 1.545659
D 0.076200 -0.974236
2000-01-04 A -0.566446 -0.070345
B 0.036142 0.307969
C -2.074978 -0.208499
... ...
[20 rows x 2 columns]
# limiting the search
In [291]: store.select('wp',"major_axis>20000102 & minor_axis=['A','B']",
.....: start=0, stop=10)
.....:
<class 'pandas.core.panel.Panel'>
Dimensions: 2 (items) x 1 (major_axis) x 2 (minor_axis)
Items axis: Item1 to Item2
Major_axis axis: 2000-01-03 00:00:00 to 2000-01-03 00:00:00
Minor_axis axis: A to B
Note
select will raise a ValueError if the query expression has an unknown variable reference. Usually this means that you are trying to select on a column that is not a data_column.
select will raise a SyntaxError if the query expression is not valid.
Using timedelta64[ns]
New in version 0.13.
Beginning in 0.13.0, you can store and query using the timedelta64[ns] type. Terms can be specified in the format: <float>(<unit>), where float may be signed (and fractional), and unit can be D,s,ms,us,ns for the timedelta. Here’s an example:
Warning
This requires numpy >= 1.7
In [292]: from datetime import timedelta
In [293]: dftd = DataFrame(dict(A = Timestamp('20130101'), B = [ Timestamp('20130101') + timedelta(days=i,seconds=10) for i in range(10) ]))
In [294]: dftd['C'] = dftd['A']-dftd['B']
In [295]: dftd
A B C
0 2013-01-01 00:00:00 2013-01-01 00:00:10 -00:00:10
1 2013-01-01 00:00:00 2013-01-02 00:00:10 -1 days, 00:00:10
2 2013-01-01 00:00:00 2013-01-03 00:00:10 -2 days, 00:00:10
3 2013-01-01 00:00:00 2013-01-04 00:00:10 -3 days, 00:00:10
4 2013-01-01 00:00:00 2013-01-05 00:00:10 -4 days, 00:00:10
5 2013-01-01 00:00:00 2013-01-06 00:00:10 -5 days, 00:00:10
6 2013-01-01 00:00:00 2013-01-07 00:00:10 -6 days, 00:00:10
7 2013-01-01 00:00:00 2013-01-08 00:00:10 -7 days, 00:00:10
8 2013-01-01 00:00:00 2013-01-09 00:00:10 -8 days, 00:00:10
9 2013-01-01 00:00:00 2013-01-10 00:00:10 -9 days, 00:00:10
[10 rows x 3 columns]
In [296]: store.append('dftd',dftd,data_columns=True)
In [297]: store.select('dftd',"C<'-3.5D'")
A B C
4 2013-01-01 00:00:00 2013-01-05 00:00:10 -4 days, 00:00:10
5 2013-01-01 00:00:00 2013-01-06 00:00:10 -5 days, 00:00:10
6 2013-01-01 00:00:00 2013-01-07 00:00:10 -6 days, 00:00:10
7 2013-01-01 00:00:00 2013-01-08 00:00:10 -7 days, 00:00:10
8 2013-01-01 00:00:00 2013-01-09 00:00:10 -8 days, 00:00:10
9 2013-01-01 00:00:00 2013-01-10 00:00:10 -9 days, 00:00:10
[6 rows x 3 columns]
Indexing¶
You can create/modify an index for a table with create_table_index after data is already in the table (after and append/put operation). Creating a table index is highly encouraged. This will speed your queries a great deal when you use a select with the indexed dimension as the where.
Note
Indexes are automagically created (starting 0.10.1) on the indexables and any data columns you specify. This behavior can be turned off by passing index=False to append.
# we have automagically already created an index (in the first section)
In [298]: i = store.root.df.table.cols.index.index
In [299]: i.optlevel, i.kind
(6, 'medium')
# change an index by passing new parameters
In [300]: store.create_table_index('df', optlevel=9, kind='full')
In [301]: i = store.root.df.table.cols.index.index
In [302]: i.optlevel, i.kind
(9, 'full')
See here for how to create a completely-sorted-index (CSI) on an existing store.
Query via Data Columns¶
You can designate (and index) certain columns that you want to be able to perform queries (other than the indexable columns, which you can always query). For instance say you want to perform this common operation, on-disk, and return just the frame that matches this query. You can specify data_columns = True to force all columns to be data_columns
In [303]: df_dc = df.copy()
In [304]: df_dc['string'] = 'foo'
In [305]: df_dc.ix[4:6,'string'] = np.nan
In [306]: df_dc.ix[7:9,'string'] = 'bar'
In [307]: df_dc['string2'] = 'cool'
In [308]: df_dc.ix[1:3,['B','C']] = 1.0
In [309]: df_dc
A B C string string2
2000-01-01 0.887163 0.859588 -0.636524 foo cool
2000-01-02 0.015696 1.000000 1.000000 foo cool
2000-01-03 0.991946 1.000000 1.000000 foo cool
2000-01-04 -0.334077 0.002118 0.405453 foo cool
2000-01-05 0.289092 1.321158 -1.546906 NaN cool
2000-01-06 -0.202646 -0.655969 0.193421 NaN cool
2000-01-07 0.553439 1.318152 -0.469305 foo cool
2000-01-08 0.675554 -1.817027 -0.183109 bar cool
[8 rows x 5 columns]
# on-disk operations
In [310]: store.append('df_dc', df_dc, data_columns = ['B', 'C', 'string', 'string2'])
In [311]: store.select('df_dc', [ Term('B>0') ])
A B C string string2
2000-01-01 0.887163 0.859588 -0.636524 foo cool
2000-01-02 0.015696 1.000000 1.000000 foo cool
2000-01-03 0.991946 1.000000 1.000000 foo cool
2000-01-04 -0.334077 0.002118 0.405453 foo cool
2000-01-05 0.289092 1.321158 -1.546906 NaN cool
2000-01-07 0.553439 1.318152 -0.469305 foo cool
[6 rows x 5 columns]
# getting creative
In [312]: store.select('df_dc', 'B > 0 & C > 0 & string == foo')
A B C string string2
2000-01-02 0.015696 1.000000 1.000000 foo cool
2000-01-03 0.991946 1.000000 1.000000 foo cool
2000-01-04 -0.334077 0.002118 0.405453 foo cool
[3 rows x 5 columns]
# this is in-memory version of this type of selection
In [313]: df_dc[(df_dc.B > 0) & (df_dc.C > 0) & (df_dc.string == 'foo')]
A B C string string2
2000-01-02 0.015696 1.000000 1.000000 foo cool
2000-01-03 0.991946 1.000000 1.000000 foo cool
2000-01-04 -0.334077 0.002118 0.405453 foo cool
[3 rows x 5 columns]
# we have automagically created this index and the B/C/string/string2
# columns are stored separately as ``PyTables`` columns
In [314]: store.root.df_dc.table
/df_dc/table (Table(8,)) ''
description := {
"index": Int64Col(shape=(), dflt=0, pos=0),
"values_block_0": Float64Col(shape=(1,), dflt=0.0, pos=1),
"B": Float64Col(shape=(), dflt=0.0, pos=2),
"C": Float64Col(shape=(), dflt=0.0, pos=3),
"string": StringCol(itemsize=3, shape=(), dflt='', pos=4),
"string2": StringCol(itemsize=4, shape=(), dflt='', pos=5)}
byteorder := 'little'
chunkshape := (1680,)
autoindex := True
colindexes := {
"index": Index(6, medium, shuffle, zlib(1)).is_csi=False,
"C": Index(6, medium, shuffle, zlib(1)).is_csi=False,
"B": Index(6, medium, shuffle, zlib(1)).is_csi=False,
"string2": Index(6, medium, shuffle, zlib(1)).is_csi=False,
"string": Index(6, medium, shuffle, zlib(1)).is_csi=False}
There is some performance degredation by making lots of columns into data columns, so it is up to the user to designate these. In addition, you cannot change data columns (nor indexables) after the first append/put operation (Of course you can simply read in the data and create a new table!)
Iterator¶
Starting in 0.11.0, you can pass, iterator=True or chunksize=number_in_a_chunk to select and select_as_multiple to return an iterator on the results. The default is 50,000 rows returned in a chunk.
In [315]: for df in store.select('df', chunksize=3):
.....: print(df)
.....:
A B C
2000-01-01 0.887163 0.859588 -0.636524
2000-01-02 0.015696 -2.242685 1.150036
2000-01-03 0.991946 0.953324 -2.021255
[3 rows x 3 columns]
A B C
2000-01-04 -0.334077 0.002118 0.405453
2000-01-05 0.289092 1.321158 -1.546906
2000-01-06 -0.202646 -0.655969 0.193421
[3 rows x 3 columns]
A B C
2000-01-07 0.553439 1.318152 -0.469305
2000-01-08 0.675554 -1.817027 -0.183109
[2 rows x 3 columns]
Note
New in version 0.12.0.
You can also use the iterator with read_hdf which will open, then automatically close the store when finished iterating.
for df in read_hdf('store.h5','df', chunsize=3):
print(df)
Note, that the chunksize keyword applies to the returned rows. So if you are doing a query, then that set will be subdivided and returned in the iterator. Keep in mind that if you do not pass a where selection criteria then the nrows of the table are considered.
Advanced Queries¶
Select a Single Column
To retrieve a single indexable or data column, use the method select_column. This will, for example, enable you to get the index very quickly. These return a Series of the result, indexed by the row number. These do not currently accept the where selector.
In [316]: store.select_column('df_dc', 'index')
0 2000-01-01 00:00:00
1 2000-01-02 00:00:00
2 2000-01-03 00:00:00
3 2000-01-04 00:00:00
4 2000-01-05 00:00:00
5 2000-01-06 00:00:00
6 2000-01-07 00:00:00
7 2000-01-08 00:00:00
dtype: datetime64[ns]
In [317]: store.select_column('df_dc', 'string')
0 foo
1 foo
2 foo
3 foo
4 NaN
5 NaN
6 foo
7 bar
dtype: object
Selecting coordinates
Sometimes you want to get the coordinates (a.k.a the index locations) of your query. This returns an Int64Index of the resulting locations. These coordinates can also be passed to subsequent where operations.
In [318]: df_coord = DataFrame(np.random.randn(1000,2),index=date_range('20000101',periods=1000))
In [319]: store.append('df_coord',df_coord)
In [320]: c = store.select_as_coordinates('df_coord','index>20020101')
In [321]: c.summary()
u'Int64Index: 268 entries, 732 to 999'
In [322]: store.select('df_coord',where=c)
0 1
2002-01-02 -0.667994 -0.368175
2002-01-03 0.020119 -0.823208
2002-01-04 -0.165481 0.720866
2002-01-05 1.295919 -0.527767
2002-01-06 -0.463393 -0.150792
2002-01-07 -1.139341 -0.954387
2002-01-08 0.051837 -0.147048
2002-01-09 -0.383978 1.209025
2002-01-10 0.213923 -0.113980
2002-01-11 0.944945 -0.183393
2002-01-12 1.714323 0.024600
2002-01-13 0.454133 0.272278
2002-01-14 0.305823 -0.390413
2002-01-15 0.424165 0.208513
2002-01-16 0.429386 1.357697
... ...
[268 rows x 2 columns]
Selecting using a where mask
Sometime your query can involve creating a list of rows to select. Usually this mask would be a resulting index from an indexing operation. This example selects the months of a datetimeindex which are 5.
In [323]: df_mask = DataFrame(np.random.randn(1000,2),index=date_range('20000101',periods=1000))
In [324]: store.append('df_mask',df_mask)
In [325]: c = store.select_column('df_mask','index')
In [326]: where = c[DatetimeIndex(c).month==5].index
In [327]: store.select('df_mask',where=where)
0 1
2000-05-01 -0.098554 -0.280782
2000-05-02 0.739851 1.627182
2000-05-03 0.030132 -0.145601
2000-05-04 0.227530 1.048856
2000-05-05 1.773939 1.116887
2000-05-06 1.081251 1.509416
2000-05-07 -0.498694 -0.913155
2000-05-08 -0.126945 0.079686
2000-05-09 1.020345 -0.790110
2000-05-10 -1.155447 -0.367505
2000-05-11 1.263914 0.827049
2000-05-12 -0.572469 1.211678
2000-05-13 1.792988 1.379125
2000-05-14 2.049917 0.992011
2000-05-15 -0.930035 -0.249394
... ...
[93 rows x 2 columns]
Storer Object
If you want to inspect the stored object, retrieve via get_storer. You could use this programmatically to say get the number of rows in an object.
In [328]: store.get_storer('df_dc').nrows
8
Multiple Table Queries¶
New in 0.10.1 are the methods append_to_multiple and select_as_multiple, that can perform appending/selecting from multiple tables at once. The idea is to have one table (call it the selector table) that you index most/all of the columns, and perform your queries. The other table(s) are data tables with an index matching the selector table’s index. You can then perform a very fast query on the selector table, yet get lots of data back. This method is similar to having a very wide table, but enables more efficient queries.
The append_to_multiple method splits a given single DataFrame into multiple tables according to d, a dictionary that maps the table names to a list of ‘columns’ you want in that table. If None is used in place of a list, that table will have the remaining unspecified columns of the given DataFrame. The argument selector defines which table is the selector table (which you can make queries from). The argument dropna will drop rows from the input DataFrame to ensure tables are synchronized. This means that if a row for one of the tables being written to is entirely np.NaN, that row will be dropped from all tables.
If dropna is False, THE USER IS RESPONSIBLE FOR SYNCHRONIZING THE TABLES. Remember that entirely np.Nan rows are not written to the HDFStore, so if you choose to call dropna=False, some tables may have more rows than others, and therefore select_as_multiple may not work or it may return unexpected results.
In [329]: df_mt = DataFrame(randn(8, 6), index=date_range('1/1/2000', periods=8),
.....: columns=['A', 'B', 'C', 'D', 'E', 'F'])
.....:
In [330]: df_mt['foo'] = 'bar'
In [331]: df_mt.ix[1, ('A', 'B')] = np.nan
# you can also create the tables individually
In [332]: store.append_to_multiple({'df1_mt': ['A', 'B'], 'df2_mt': None },
.....: df_mt, selector='df1_mt')
.....:
In [333]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/df1_mt frame_table (typ->appendable,nrows->7,ncols->2,indexers->[index],dc->[A,B])
/df2_mt frame_table (typ->appendable,nrows->7,ncols->5,indexers->[index])
/df_coord frame_table (typ->appendable,nrows->1000,ncols->2,indexers->[index])
/df_dc frame_table (typ->appendable,nrows->8,ncols->5,indexers->[index],dc->[B,C,string,string2])
/df_mask frame_table (typ->appendable,nrows->1000,ncols->2,indexers->[index])
/df_mi frame_table (typ->appendable_multi,nrows->10,ncols->5,indexers->[index],dc->[bar,foo])
/df_mixed frame_table (typ->appendable,nrows->8,ncols->7,indexers->[index])
/dfq frame_table (typ->appendable,nrows->10,ncols->4,indexers->[index],dc->[A,B,C,D])
/dftd frame_table (typ->appendable,nrows->10,ncols->3,indexers->[index],dc->[A,B,C])
/wp wide_table (typ->appendable,nrows->20,ncols->2,indexers->[major_axis,minor_axis])
/foo/bar/bah frame (shape->[8,3])
# individual tables were created
In [334]: store.select('df1_mt')
A B
2000-01-01 -0.816310 1.282296
2000-01-03 0.684353 -1.755306
2000-01-04 -1.315814 1.455079
2000-01-05 -0.027564 0.046757
2000-01-06 -0.416244 -0.821168
2000-01-07 0.665090 1.084344
2000-01-08 0.607460 0.790907
[7 rows x 2 columns]
In [335]: store.select('df2_mt')
C D E F foo
2000-01-01 -1.521825 -0.428670 -1.550209 0.826839 bar
2000-01-03 1.236974 -1.328279 0.662291 1.894976 bar
2000-01-04 -0.746478 0.851039 1.415686 -0.929096 bar
2000-01-05 -1.452287 1.575492 -0.197377 -0.219901 bar
2000-01-06 1.190342 2.115021 0.148762 1.073931 bar
2000-01-07 -0.709897 -2.022441 0.714697 0.318215 bar
2000-01-08 0.852225 0.096696 -0.379903 0.929313 bar
[7 rows x 5 columns]
# as a multiple
In [336]: store.select_as_multiple(['df1_mt', 'df2_mt'], where=['A>0', 'B>0'],
.....: selector = 'df1_mt')
.....:
A B C D E F foo
2000-01-07 0.66509 1.084344 -0.709897 -2.022441 0.714697 0.318215 bar
2000-01-08 0.60746 0.790907 0.852225 0.096696 -0.379903 0.929313 bar
[2 rows x 7 columns]
Delete from a Table¶
You can delete from a table selectively by specifying a where. In deleting rows, it is important to understand the PyTables deletes rows by erasing the rows, then moving the following data. Thus deleting can potentially be a very expensive operation depending on the orientation of your data. This is especially true in higher dimensional objects (Panel and Panel4D). To get optimal performance, it’s worthwhile to have the dimension you are deleting be the first of the indexables.
Data is ordered (on the disk) in terms of the indexables. Here’s a simple use case. You store panel-type data, with dates in the major_axis and ids in the minor_axis. The data is then interleaved like this:
- date_1
- id_1
- id_2
- .
- id_n
- date_2
- id_1
- .
- id_n
It should be clear that a delete operation on the major_axis will be fairly quick, as one chunk is removed, then the following data moved. On the other hand a delete operation on the minor_axis will be very expensive. In this case it would almost certainly be faster to rewrite the table using a where that selects all but the missing data.
# returns the number of rows deleted
In [337]: store.remove('wp', 'major_axis>20000102' )
12
In [338]: store.select('wp')
<class 'pandas.core.panel.Panel'>
Dimensions: 2 (items) x 2 (major_axis) x 4 (minor_axis)
Items axis: Item1 to Item2
Major_axis axis: 2000-01-01 00:00:00 to 2000-01-02 00:00:00
Minor_axis axis: A to D
Please note that HDF5 DOES NOT RECLAIM SPACE in the h5 files automatically. Thus, repeatedly deleting (or removing nodes) and adding again WILL TEND TO INCREASE THE FILE SIZE. To clean the file, use ptrepack (see below).
Compression¶
PyTables allows the stored data to be compressed. Tthis applies to all kinds of stores, not just tables.
- Pass complevel=int for a compression level (1-9, with 0 being no compression, and the default)
- Pass complib=lib where lib is any of zlib, bzip2, lzo, blosc for whichever compression library you prefer.
HDFStore will use the file based compression scheme if no overriding complib or complevel options are provided. blosc offers very fast compression, and is my most used. Note that lzo and bzip2 may not be installed (by Python) by default.
Compression for all objects within the file
- store_compressed = HDFStore('store_compressed.h5', complevel=9, complib='blosc')
Or on-the-fly compression (this only applies to tables). You can turn off file compression for a specific table by passing complevel=0
- store.append('df', df, complib='zlib', complevel=5)
ptrepack
PyTables offers better write performance when tables are compressed after they are written, as opposed to turning on compression at the very beginning. You can use the supplied PyTables utility ptrepack. In addition, ptrepack can change compression levels after the fact.
- ptrepack --chunkshape=auto --propindexes --complevel=9 --complib=blosc in.h5 out.h5
Furthermore ptrepack in.h5 out.h5 will repack the file to allow you to reuse previously deleted space. Aalternatively, one can simply remove the file and write again, or use the copy method.
Notes & Caveats¶
- Once a table is created its items (Panel) / columns (DataFrame) are fixed; only exactly the same columns can be appended
- If a row has np.nan for EVERY COLUMN (having a nan in a string, or a NaT in a datetime-like column counts as having a value), then those rows WILL BE DROPPED IMPLICITLY. This limitation may be addressed in the future.
- HDFStore is not-threadsafe for writing. The underlying PyTables only supports concurrent reads (via threading or processes). If you need reading and writing at the same time, you need to serialize these operations in a single thread in a single process. You will corrupt your data otherwise. See the issue (:2397) for more information.
- If you use locks to manage write access between multiple processes, you may want to use fsync() before releasing write locks. For convenience you can use store.flush(fsync=True) to do this for you.
- PyTables only supports fixed-width string columns in tables. The sizes of a string based indexing column (e.g. columns or minor_axis) are determined as the maximum size of the elements in that axis or by passing the parameter
Warning
PyTables will show a NaturalNameWarning if a column name cannot be used as an attribute selector. Generally identifiers that have spaces, start with numbers, or _, or have - embedded are not considered natural. These types of identifiers cannot be used in a where clause and are generally a bad idea.
DataTypes¶
HDFStore will map an object dtype to the PyTables underlying dtype. This means the following types are known to work:
- floating : float64, float32, float16 (using np.nan to represent invalid values)
- integer : int64, int32, int8, uint64, uint32, uint8
- bool
- datetime64[ns] (using NaT to represent invalid values)
- object : strings (using np.nan to represent invalid values)
Currently, unicode and datetime columns (represented with a dtype of object), WILL FAIL. In addition, even though a column may look like a datetime64[ns], if it contains np.nan, this WILL FAIL. You can try to convert datetimelike columns to proper datetime64[ns] columns, that possibily contain NaT to represent invalid values. (Some of these issues have been addressed and these conversion may not be necessary in future versions of pandas)
In [339]: import datetime In [340]: df = DataFrame(dict(datelike=Series([datetime.datetime(2001, 1, 1), .....: datetime.datetime(2001, 1, 2), np.nan]))) .....: In [341]: df datelike 0 2001-01-01 00:00:00 1 2001-01-02 00:00:00 2 NaN [3 rows x 1 columns] In [342]: df.dtypes datelike object dtype: object # to convert In [343]: df['datelike'] = Series(df['datelike'].values, dtype='M8[ns]') In [344]: df datelike 0 2001-01-01 00:00:00 1 2001-01-02 00:00:00 2 NaT [3 rows x 1 columns] In [345]: df.dtypes datelike datetime64[ns] dtype: object
String Columns¶
min_itemsize
The underlying implementation of HDFStore uses a fixed column width (itemsize) for string columns. A string column itemsize is calculated as the maximum of the length of data (for that column) that is passed to the HDFStore, in the first append. Subsequent appends, may introduce a string for a column larger than the column can hold, an Exception will be raised (otherwise you could have a silent truncation of these columns, leading to loss of information). In the future we may relax this and allow a user-specified truncation to occur.
Pass min_itemsize on the first table creation to a-priori specifiy the minimum length of a particular string column. min_itemsize can be an integer, or a dict mapping a column name to an integer. You can pass values as a key to allow all indexables or data_columns to have this min_itemsize.
Starting in 0.11.0, passing a min_itemsize dict will cause all passed columns to be created as data_columns automatically.
Note
If you are not passing any data_columns, then the min_itemsize will be the maximum of the length of any string passed
In [346]: dfs = DataFrame(dict(A = 'foo', B = 'bar'),index=list(range(5)))
In [347]: dfs
A B
0 foo bar
1 foo bar
2 foo bar
3 foo bar
4 foo bar
[5 rows x 2 columns]
# A and B have a size of 30
In [348]: store.append('dfs', dfs, min_itemsize = 30)
In [349]: store.get_storer('dfs').table
/dfs/table (Table(5,)) ''
description := {
"index": Int64Col(shape=(), dflt=0, pos=0),
"values_block_0": StringCol(itemsize=30, shape=(2,), dflt='', pos=1)}
byteorder := 'little'
chunkshape := (963,)
autoindex := True
colindexes := {
"index": Index(6, medium, shuffle, zlib(1)).is_csi=False}
# A is created as a data_column with a size of 30
# B is size is calculated
In [350]: store.append('dfs2', dfs, min_itemsize = { 'A' : 30 })
In [351]: store.get_storer('dfs2').table
/dfs2/table (Table(5,)) ''
description := {
"index": Int64Col(shape=(), dflt=0, pos=0),
"values_block_0": StringCol(itemsize=3, shape=(1,), dflt='', pos=1),
"A": StringCol(itemsize=30, shape=(), dflt='', pos=2)}
byteorder := 'little'
chunkshape := (1598,)
autoindex := True
colindexes := {
"A": Index(6, medium, shuffle, zlib(1)).is_csi=False,
"index": Index(6, medium, shuffle, zlib(1)).is_csi=False}
nan_rep
String columns will serialize a np.nan (a missing value) with the nan_rep string representation. This defaults to the string value nan. You could inadvertently turn an actual nan value into a missing value.
In [352]: dfss = DataFrame(dict(A = ['foo','bar','nan']))
In [353]: dfss
A
0 foo
1 bar
2 nan
[3 rows x 1 columns]
In [354]: store.append('dfss', dfss)
In [355]: store.select('dfss')
A
0 foo
1 bar
2 NaN
[3 rows x 1 columns]
# here you need to specify a different nan rep
In [356]: store.append('dfss2', dfss, nan_rep='_nan_')
In [357]: store.select('dfss2')
A
0 foo
1 bar
2 nan
[3 rows x 1 columns]
External Compatibility¶
HDFStore write table format objects in specific formats suitable for producing loss-less roundtrips to pandas objects. For external compatibility, HDFStore can read native PyTables format tables. It is possible to write an HDFStore object that can easily be imported into R using the rhdf5 library. Create a table format store like this:
In [358]: store_export = HDFStore('export.h5') In [359]: store_export.append('df_dc', df_dc, data_columns=df_dc.columns) In [360]: store_export <class 'pandas.io.pytables.HDFStore'> File path: export.h5 /df_dc frame_table (typ->appendable,nrows->8,ncols->5,indexers->[index],dc->[A,B,C,string,string2])
Backwards Compatibility¶
0.10.1 of HDFStore can read tables created in a prior version of pandas, however query terms using the prior (undocumented) methodology are unsupported. HDFStore will issue a warning if you try to use a legacy-format file. You must read in the entire file and write it out using the new format, using the method copy to take advantage of the updates. The group attribute pandas_version contains the version information. copy takes a number of options, please see the docstring.
# a legacy store In [361]: legacy_store = HDFStore(legacy_file_path,'r') In [362]: legacy_store <class 'pandas.io.pytables.HDFStore'> File path: /home/docbuild/CI/pandas/doc/source/_static/legacy_0.10.h5 /a series (shape->[30]) /b frame (shape->[30,4]) /df1_mixed frame_table [0.10.0] (typ->appendable,nrows->30,ncols->11,indexers->[index]) /p1_mixed wide_table [0.10.0] (typ->appendable,nrows->120,ncols->9,indexers->[major_axis,minor_axis]) /p4d_mixed ndim_table [0.10.0] (typ->appendable,nrows->360,ncols->9,indexers->[items,major_axis,minor_axis]) /foo/bar wide (shape->[3,30,4]) # copy (and return the new handle) In [363]: new_store = legacy_store.copy('store_new.h5') In [364]: new_store <class 'pandas.io.pytables.HDFStore'> File path: store_new.h5 /a series (shape->[30]) /b frame (shape->[30,4]) /df1_mixed frame_table (typ->appendable,nrows->30,ncols->11,indexers->[index]) /p1_mixed wide_table (typ->appendable,nrows->120,ncols->9,indexers->[major_axis,minor_axis]) /p4d_mixed wide_table (typ->appendable,nrows->360,ncols->9,indexers->[items,major_axis,minor_axis]) /foo/bar wide (shape->[3,30,4]) In [365]: new_store.close()
Performance¶
- Tables come with a writing performance penalty as compared to regular stores. The benefit is the ability to append/delete and query (potentially very large amounts of data). Write times are generally longer as compared with regular stores. Query times can be quite fast, especially on an indexed axis.
- You can pass chunksize=<int> to append, specifying the write chunksize (default is 50000). This will signficantly lower your memory usage on writing.
- You can pass expectedrows=<int> to the first append, to set the TOTAL number of expected rows that PyTables will expected. This will optimize read/write performance.
- Duplicate rows can be written to tables, but are filtered out in selection (with the last items being selected; thus a table is unique on major, minor pairs)
- A PerformanceWarning will be raised if you are attempting to store types that will be pickled by PyTables (rather than stored as endemic types). See Here for more information and some solutions.
Experimental¶
HDFStore supports Panel4D storage.
In [366]: p4d = Panel4D({ 'l1' : wp })
In [367]: p4d
<class 'pandas.core.panelnd.Panel4D'>
Dimensions: 1 (labels) x 2 (items) x 5 (major_axis) x 4 (minor_axis)
Labels axis: l1 to l1
Items axis: Item1 to Item2
Major_axis axis: 2000-01-01 00:00:00 to 2000-01-05 00:00:00
Minor_axis axis: A to D
In [368]: store.append('p4d', p4d)
In [369]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/df1_mt frame_table (typ->appendable,nrows->7,ncols->2,indexers->[index],dc->[A,B])
/df2_mt frame_table (typ->appendable,nrows->7,ncols->5,indexers->[index])
/df_coord frame_table (typ->appendable,nrows->1000,ncols->2,indexers->[index])
/df_dc frame_table (typ->appendable,nrows->8,ncols->5,indexers->[index],dc->[B,C,string,string2])
/df_mask frame_table (typ->appendable,nrows->1000,ncols->2,indexers->[index])
/df_mi frame_table (typ->appendable_multi,nrows->10,ncols->5,indexers->[index],dc->[bar,foo])
/df_mixed frame_table (typ->appendable,nrows->8,ncols->7,indexers->[index])
/dfq frame_table (typ->appendable,nrows->10,ncols->4,indexers->[index],dc->[A,B,C,D])
/dfs frame_table (typ->appendable,nrows->5,ncols->2,indexers->[index])
/dfs2 frame_table (typ->appendable,nrows->5,ncols->2,indexers->[index],dc->[A])
/dfss frame_table (typ->appendable,nrows->3,ncols->1,indexers->[index])
/dfss2 frame_table (typ->appendable,nrows->3,ncols->1,indexers->[index])
/dftd frame_table (typ->appendable,nrows->10,ncols->3,indexers->[index],dc->[A,B,C])
/p4d wide_table (typ->appendable,nrows->40,ncols->1,indexers->[items,major_axis,minor_axis])
/wp wide_table (typ->appendable,nrows->8,ncols->2,indexers->[major_axis,minor_axis])
/foo/bar/bah frame (shape->[8,3])
These, by default, index the three axes items, major_axis, minor_axis. On an AppendableTable it is possible to setup with the first append a different indexing scheme, depending on how you want to store your data. Pass the axes keyword with a list of dimensions (currently must by exactly 1 less than the total dimensions of the object). This cannot be changed after table creation.
In [370]: store.append('p4d2', p4d, axes=['labels', 'major_axis', 'minor_axis'])
In [371]: store
<class 'pandas.io.pytables.HDFStore'>
File path: store.h5
/df frame_table (typ->appendable,nrows->8,ncols->3,indexers->[index])
/df1_mt frame_table (typ->appendable,nrows->7,ncols->2,indexers->[index],dc->[A,B])
/df2_mt frame_table (typ->appendable,nrows->7,ncols->5,indexers->[index])
/df_coord frame_table (typ->appendable,nrows->1000,ncols->2,indexers->[index])
/df_dc frame_table (typ->appendable,nrows->8,ncols->5,indexers->[index],dc->[B,C,string,string2])
/df_mask frame_table (typ->appendable,nrows->1000,ncols->2,indexers->[index])
/df_mi frame_table (typ->appendable_multi,nrows->10,ncols->5,indexers->[index],dc->[bar,foo])
/df_mixed frame_table (typ->appendable,nrows->8,ncols->7,indexers->[index])
/dfq frame_table (typ->appendable,nrows->10,ncols->4,indexers->[index],dc->[A,B,C,D])
/dfs frame_table (typ->appendable,nrows->5,ncols->2,indexers->[index])
/dfs2 frame_table (typ->appendable,nrows->5,ncols->2,indexers->[index],dc->[A])
/dfss frame_table (typ->appendable,nrows->3,ncols->1,indexers->[index])
/dfss2 frame_table (typ->appendable,nrows->3,ncols->1,indexers->[index])
/dftd frame_table (typ->appendable,nrows->10,ncols->3,indexers->[index],dc->[A,B,C])
/p4d wide_table (typ->appendable,nrows->40,ncols->1,indexers->[items,major_axis,minor_axis])
/p4d2 wide_table (typ->appendable,nrows->20,ncols->2,indexers->[labels,major_axis,minor_axis])
/wp wide_table (typ->appendable,nrows->8,ncols->2,indexers->[major_axis,minor_axis])
/foo/bar/bah frame (shape->[8,3])
In [372]: store.select('p4d2', [ Term('labels=l1'), Term('items=Item1'), Term('minor_axis=A_big_strings') ])
<class 'pandas.core.panelnd.Panel4D'>
Dimensions: 0 (labels) x 1 (items) x 0 (major_axis) x 0 (minor_axis)
Labels axis: None
Items axis: Item1 to Item1
Major_axis axis: None
Minor_axis axis: None
SQL Queries¶
The pandas.io.sql module provides a collection of query wrappers to both facilitate data retrieval and to reduce dependency on DB-specific API. These wrappers only support the Python database adapters which respect the Python DB-API. See some cookbook examples for some advanced strategies
For example, suppose you want to query some data with different types from a table such as:
| id | Date | Col_1 | Col_2 | Col_3 |
|---|---|---|---|---|
| 26 | 2012-10-18 | X | 25.7 | True |
| 42 | 2012-10-19 | Y | -12.4 | False |
| 63 | 2012-10-20 | Z | 5.73 | True |
Functions from pandas.io.sql can extract some data into a DataFrame. In the following example, we use the SQlite SQL database engine. You can use a temporary SQLite database where data are stored in “memory”. Just do:
import sqlite3
from pandas.io import sql
# Create your connection.
cnx = sqlite3.connect(':memory:')
Let data be the name of your SQL table. With a query and your database connection, just use the read_frame() function to get the query results into a DataFrame:
In [373]: sql.read_frame("SELECT * FROM data;", cnx)
id date Col_1 Col_2 Col_3
0 26 2010-10-18 00:00:00 X 27.50 1
1 42 2010-10-19 00:00:00 Y -12.50 0
2 63 2010-10-20 00:00:00 Z 5.73 1
[3 rows x 5 columns]
You can also specify the name of the column as the DataFrame index:
In [374]: sql.read_frame("SELECT * FROM data;", cnx, index_col='id')
date Col_1 Col_2 Col_3
id
26 2010-10-18 00:00:00 X 27.50 1
42 2010-10-19 00:00:00 Y -12.50 0
63 2010-10-20 00:00:00 Z 5.73 1
[3 rows x 4 columns]
In [375]: sql.read_frame("SELECT * FROM data;", cnx, index_col='date')
id Col_1 Col_2 Col_3
date
2010-10-18 00:00:00 26 X 27.50 1
2010-10-19 00:00:00 42 Y -12.50 0
2010-10-20 00:00:00 63 Z 5.73 1
[3 rows x 4 columns]
Of course, you can specify a more “complex” query.
In [376]: sql.read_frame("SELECT id, Col_1, Col_2 FROM data WHERE id = 42;", cnx)
id Col_1 Col_2
0 42 Y -12.5
[1 rows x 3 columns]
There are a few other available functions:
- tquery returns a list of tuples corresponding to each row.
- uquery does the same thing as tquery, but instead of returning results it returns the number of related rows.
- write_frame writes records stored in a DataFrame into the SQL table.
- has_table checks if a given SQLite table exists.
Note
For now, writing your DataFrame into a database works only with SQLite. Moreover, the index will currently be dropped.
Google BigQuery (Experimental)¶
New in version 0.13.0.
The pandas.io.gbq module provides a wrapper for Google’s BigQuery analytics web service to simplify retrieving results from BigQuery tables using SQL-like queries. Result sets are parsed into a pandas DataFrame with a shape derived from the source table. Additionally, DataFrames can be uploaded into BigQuery datasets as tables if the source datatypes are compatible with BigQuery ones.
For specifics on the service itself, see here
As an example, suppose you want to load all data from an existing table : test_dataset.test_table into BigQuery and pull it into a DataFrame.
from pandas.io import gbq
# Insert your BigQuery Project ID Here
# Can be found in the web console, or
# using the command line tool `bq ls`
projectid = "xxxxxxxx"
data_frame = gbq.read_gbq('SELECT * FROM test_dataset.test_table', project_id = projectid)
The user will then be authenticated by the bq command line client - this usually involves the default browser opening to a login page, though the process can be done entirely from command line if necessary. Datasets and additional parameters can be either configured with bq, passed in as options to read_gbq, or set using Google’s gflags (this is not officially supported by this module, though care was taken to ensure that they should be followed regardless of how you call the method).
Additionally, you can define which column to use as an index as well as a preferred column order as follows:
data_frame = gbq.read_gbq('SELECT * FROM test_dataset.test_table',
index_col='index_column_name',
col_order='[col1, col2, col3,...]', project_id = projectid)
Finally, if you would like to create a BigQuery table, my_dataset.my_table, from the rows of DataFrame, df:
df = pandas.DataFrame({'string_col_name' : ['hello'],
'integer_col_name' : [1],
'boolean_col_name' : [True]})
schema = ['STRING', 'INTEGER', 'BOOLEAN']
data_frame = gbq.to_gbq(df, 'my_dataset.my_table',
if_exists='fail', schema = schema, project_id = projectid)
To add more rows to this, simply:
df2 = pandas.DataFrame({'string_col_name' : ['hello2'],
'integer_col_name' : [2],
'boolean_col_name' : [False]})
data_frame = gbq.to_gbq(df2, 'my_dataset.my_table', if_exists='append', project_id = projectid)
Note
A default project id can be set using the command line: bq init.
There is a hard cap on BigQuery result sets, at 128MB compressed. Also, the BigQuery SQL query language has some oddities, see here
You can access the management console to determine project id’s by: <https://code.google.com/apis/console/b/0/?noredirect>
Warning
To use this module, you will need a BigQuery account. See <https://cloud.google.com/products/big-query> for details.
As of 10/10/13, there is a bug in Google’s API preventing result sets from being larger than 100,000 rows. A patch is scheduled for the week of 10/14/13.
STATA Format¶
New in version 0.12.0.
Writing to STATA format¶
The method to_stata() will write a DataFrame into a .dta file. The format version of this file is always 115 (Stata 12).
In [377]: df = DataFrame(randn(10, 2), columns=list('AB'))
In [378]: df.to_stata('stata.dta')
Reading from STATA format¶
The top-level function read_stata will read a dta format file and return a DataFrame: The class StataReader will read the header of the given dta file at initialization. Its method data() will read the observations, converting them to a DataFrame which is returned:
In [379]: pd.read_stata('stata.dta')
index A B
0 0 0.811031 -0.356817
1 1 1.047085 0.664705
2 2 -0.086919 0.416905
3 3 -0.764381 -0.287229
4 4 -0.089351 -1.035115
5 5 0.489131 -0.253340
6 6 -1.948100 -0.116556
7 7 0.800597 -0.796154
8 8 -0.382952 -0.397373
9 9 -0.717627 0.156995
[10 rows x 3 columns]
Currently the index is retrieved as a column on read back.
The parameter convert_categoricals indicates wheter value labels should be read and used to create a Categorical variable from them. Value labels can also be retrieved by the function variable_labels, which requires data to be called before (see pandas.io.stata.StataReader).
The StataReader supports .dta Formats 104, 105, 108, 113-115 and 117. Alternatively, the function read_stata() can be used