Home - News ( United States | United Kingdom | Italy | Germany ) - Football scores

Showing content from https://pandas.pydata.org/pandas-docs/stable/reference/api/../api/../../user_guide/io.html below:

IO tools (text, CSV, HDF5, …) — pandas 2.3.1 documentation

The pandas I/O API is a set of top level reader functions accessed like pandas.read_csv() that generally return a pandas object. The corresponding writer functions are object methods that are accessed like DataFrame.to_csv(). Below is a table containing available readers and writers.

Here is an informal performance comparison for some of these IO methods.

Note

For examples that use the StringIO class, make sure you import it with from io import StringIO for Python 3.

The workhorse function for reading text files (a.k.a. flat files) is read_csv(). See the cookbook for some advanced strategies.

read_csv() accepts the following common arguments:

filepath_or_buffervarious

Either a path to a file (a str, pathlib.Path, or py:py._path.local.LocalPath), URL (including http, ftp, and S3 locations), or any object with a read() method (such as an open file or StringIO).

sepstr, defaults to ',' for read_csv(), \t for read_table()

Delimiter to use. If sep is None, the C engine cannot automatically detect the separator, but the Python parsing engine can, meaning the latter will be used and automatically detect the separator by Python’s builtin sniffer tool, csv.Sniffer. In addition, separators longer than 1 character and different from '\s+' will be interpreted as regular expressions and will also force the use of the Python parsing engine. Note that regex delimiters are prone to ignoring quoted data. Regex example: '\\r\\t'.

delimiterstr, default None

Alternative argument name for sep.

delim_whitespaceboolean, default False

Specifies whether or not whitespace (e.g. ' ' or '\t') will be used as the delimiter. Equivalent to setting sep='\s+'. If this option is set to True, nothing should be passed in for the delimiter parameter.

headerint or list of ints, default 'infer'

Row number(s) to use as the column names, and the start of the data. Default behavior is to infer the column names: if no names are passed the behavior is identical to header=0 and column names are inferred from the first line of the file, if column names are passed explicitly then the behavior is identical to header=None. Explicitly pass header=0 to be able to replace existing names.

The header can be a list of ints that specify row locations for a MultiIndex on the columns e.g. [0,1,3]. Intervening rows that are not specified will be skipped (e.g. 2 in this example is skipped). Note that this parameter ignores commented lines and empty lines if skip_blank_lines=True, so header=0 denotes the first line of data rather than the first line of the file.

namesarray-like, default None

List of column names to use. If file contains no header row, then you should explicitly pass header=None. Duplicates in this list are not allowed.

index_colint, str, sequence of int / str, or False, optional, default None

Column(s) to use as the row labels of the DataFrame, either given as string name or column index. If a sequence of int / str is given, a MultiIndex is used.

Note

index_col=False can be used to force pandas to not use the first column as the index, e.g. when you have a malformed file with delimiters at the end of each line.

The default value of None instructs pandas to guess. If the number of fields in the column header row is equal to the number of fields in the body of the data file, then a default index is used. If it is larger, then the first columns are used as index so that the remaining number of fields in the body are equal to the number of fields in the header.

The first row after the header is used to determine the number of columns, which will go into the index. If the subsequent rows contain less columns than the first row, they are filled with NaN.

This can be avoided through usecols. This ensures that the columns are taken as is and the trailing data are ignored.

usecolslist-like or callable, default None

Return a subset of the columns. If list-like, all elements must either be positional (i.e. integer indices into the document columns) or strings that correspond to column names provided either by the user in names or inferred from the document header row(s). If names are given, the document header row(s) are not taken into account. For example, a valid list-like usecols parameter would be [0, 1, 2] or ['foo', 'bar', 'baz'].

Element order is ignored, so usecols=[0, 1] is the same as [1, 0]. To instantiate a DataFrame from data with element order preserved use pd.read_csv(data, usecols=['foo', 'bar'])[['foo', 'bar']] for columns in ['foo', 'bar'] order or pd.read_csv(data, usecols=['foo', 'bar'])[['bar', 'foo']] for ['bar', 'foo'] order.

If callable, the callable function will be evaluated against the column names, returning names where the callable function evaluates to True:

In [1]: import pandas as pd

In [2]: from io import StringIO

In [3]: data = "col1,col2,col3\na,b,1\na,b,2\nc,d,3"

In [4]: pd.read_csv(StringIO(data))
Out[4]: 
  col1 col2  col3
0    a    b     1
1    a    b     2
2    c    d     3

In [5]: pd.read_csv(StringIO(data), usecols=lambda x: x.upper() in ["COL1", "COL3"])
Out[5]: 
  col1  col3
0    a     1
1    a     2
2    c     3

Using this parameter results in much faster parsing time and lower memory usage when using the c engine. The Python engine loads the data first before deciding which columns to drop.

dtypeType name or dict of column -> type, default None

Data type for data or columns. E.g. {'a': np.float64, 'b': np.int32, 'c': 'Int64'} Use str or object together with suitable na_values settings to preserve and not interpret dtype. If converters are specified, they will be applied INSTEAD of dtype conversion.

Added in version 1.5.0: Support for defaultdict was added. Specify a defaultdict as input where the default determines the dtype of the columns which are not explicitly listed.

dtype_backend{“numpy_nullable”, “pyarrow”}, defaults to NumPy backed DataFrames

Which dtype_backend to use, e.g. whether a DataFrame should have NumPy arrays, nullable dtypes are used for all dtypes that have a nullable implementation when “numpy_nullable” is set, pyarrow is used for all dtypes if “pyarrow” is set.

The dtype_backends are still experimential.

Added in version 2.0.

engine{'c', 'python', 'pyarrow'}

Parser engine to use. The C and pyarrow engines are faster, while the python engine is currently more feature-complete. Multithreading is currently only supported by the pyarrow engine.

Added in version 1.4.0: The “pyarrow” engine was added as an experimental engine, and some features are unsupported, or may not work correctly, with this engine.

convertersdict, default None

Dict of functions for converting values in certain columns. Keys can either be integers or column labels.

true_valueslist, default None

Values to consider as True.

false_valueslist, default None

Values to consider as False.

skipinitialspaceboolean, default False

Skip spaces after delimiter.

skiprowslist-like or integer, default None

Line numbers to skip (0-indexed) or number of lines to skip (int) at the start of the file.

If callable, the callable function will be evaluated against the row indices, returning True if the row should be skipped and False otherwise:

In [6]: data = "col1,col2,col3\na,b,1\na,b,2\nc,d,3"

In [7]: pd.read_csv(StringIO(data))
Out[7]: 
  col1 col2  col3
0    a    b     1
1    a    b     2
2    c    d     3

In [8]: pd.read_csv(StringIO(data), skiprows=lambda x: x % 2 != 0)
Out[8]: 
  col1 col2  col3
0    a    b     2

skipfooterint, default 0

Number of lines at bottom of file to skip (unsupported with engine=’c’).

nrowsint, default None

Number of rows of file to read. Useful for reading pieces of large files.

low_memoryboolean, default True

Internally process the file in chunks, resulting in lower memory use while parsing, but possibly mixed type inference. To ensure no mixed types either set False, or specify the type with the dtype parameter. Note that the entire file is read into a single DataFrame regardless, use the chunksize or iterator parameter to return the data in chunks. (Only valid with C parser)

memory_mapboolean, default False

If a filepath is provided for filepath_or_buffer, map the file object directly onto memory and access the data directly from there. Using this option can improve performance because there is no longer any I/O overhead.

na_valuesscalar, str, list-like, or dict, default None

Additional strings to recognize as NA/NaN. If dict passed, specific per-column NA values. See na values const below for a list of the values interpreted as NaN by default.

keep_default_naboolean, default True

Whether or not to include the default NaN values when parsing the data. Depending on whether na_values is passed in, the behavior is as follows:

• If keep_default_na is True, and na_values are specified, na_values is appended to the default NaN values used for parsing.

• If keep_default_na is True, and na_values are not specified, only the default NaN values are used for parsing.

• If keep_default_na is False, and na_values are specified, only the NaN values specified na_values are used for parsing.

• If keep_default_na is False, and na_values are not specified, no strings will be parsed as NaN.

Note that if na_filter is passed in as False, the keep_default_na and na_values parameters will be ignored.

na_filterboolean, default True

Detect missing value markers (empty strings and the value of na_values). In data without any NAs, passing na_filter=False can improve the performance of reading a large file.

verboseboolean, default False

Indicate number of NA values placed in non-numeric columns.

skip_blank_linesboolean, default True

If True, skip over blank lines rather than interpreting as NaN values.

parse_datesboolean or list of ints or names or list of lists or dict, default False.

• If True -> try parsing the index.

• If [1, 2, 3] -> try parsing columns 1, 2, 3 each as a separate date column.

• If [[1, 3]] -> combine columns 1 and 3 and parse as a single date column.

• If {'foo': [1, 3]} -> parse columns 1, 3 as date and call result ‘foo’.

Note

A fast-path exists for iso8601-formatted dates.

infer_datetime_formatboolean, default False

If True and parse_dates is enabled for a column, attempt to infer the datetime format to speed up the processing.

Deprecated since version 2.0.0: A strict version of this argument is now the default, passing it has no effect.

keep_date_colboolean, default False

If True and parse_dates specifies combining multiple columns then keep the original columns.

date_parserfunction, default None

Function to use for converting a sequence of string columns to an array of datetime instances. The default uses dateutil.parser.parser to do the conversion. pandas will try to call date_parser in three different ways, advancing to the next if an exception occurs: 1) Pass one or more arrays (as defined by parse_dates) as arguments; 2) concatenate (row-wise) the string values from the columns defined by parse_dates into a single array and pass that; and 3) call date_parser once for each row using one or more strings (corresponding to the columns defined by parse_dates) as arguments.

Deprecated since version 2.0.0: Use date_format instead, or read in as object and then apply to_datetime() as-needed.

date_formatstr or dict of column -> format, default None

If used in conjunction with parse_dates, will parse dates according to this format. For anything more complex, please read in as object and then apply to_datetime() as-needed.

Added in version 2.0.0.

dayfirstboolean, default False

DD/MM format dates, international and European format.

cache_datesboolean, default True

If True, use a cache of unique, converted dates to apply the datetime conversion. May produce significant speed-up when parsing duplicate date strings, especially ones with timezone offsets.

iteratorboolean, default False

Return TextFileReader object for iteration or getting chunks with get_chunk().

chunksizeint, default None

Return TextFileReader object for iteration. See iterating and chunking below.

compression{'infer', 'gzip', 'bz2', 'zip', 'xz', 'zstd', None, dict}, default 'infer'

For on-the-fly decompression of on-disk data. If ‘infer’, then use gzip, bz2, zip, xz, or zstandard if filepath_or_buffer is path-like ending in ‘.gz’, ‘.bz2’, ‘.zip’, ‘.xz’, ‘.zst’, respectively, and no decompression otherwise. If using ‘zip’, the ZIP file must contain only one data file to be read in. Set to None for no decompression. Can also be a dict with key 'method' set to one of {'zip', 'gzip', 'bz2', 'zstd'} and other key-value pairs are forwarded to zipfile.ZipFile, gzip.GzipFile, bz2.BZ2File, or zstandard.ZstdDecompressor. As an example, the following could be passed for faster compression and to create a reproducible gzip archive: compression={'method': 'gzip', 'compresslevel': 1, 'mtime': 1}.

Changed in version 1.2.0: Previous versions forwarded dict entries for ‘gzip’ to gzip.open.

thousandsstr, default None

Thousands separator.

decimalstr, default '.'

Character to recognize as decimal point. E.g. use ',' for European data.

float_precisionstring, default None

Specifies which converter the C engine should use for floating-point values. The options are None for the ordinary converter, high for the high-precision converter, and round_trip for the round-trip converter.

lineterminatorstr (length 1), default None

Character to break file into lines. Only valid with C parser.

quotecharstr (length 1)

The character used to denote the start and end of a quoted item. Quoted items can include the delimiter and it will be ignored.

quotingint or csv.QUOTE_* instance, default 0

Control field quoting behavior per csv.QUOTE_* constants. Use one of QUOTE_MINIMAL (0), QUOTE_ALL (1), QUOTE_NONNUMERIC (2) or QUOTE_NONE (3).

doublequoteboolean, default True

When quotechar is specified and quoting is not QUOTE_NONE, indicate whether or not to interpret two consecutive quotechar elements inside a field as a single quotechar element.

escapecharstr (length 1), default None

One-character string used to escape delimiter when quoting is QUOTE_NONE.

commentstr, default None

Indicates remainder of line should not be parsed. If found at the beginning of a line, the line will be ignored altogether. This parameter must be a single character. Like empty lines (as long as skip_blank_lines=True), fully commented lines are ignored by the parameter header but not by skiprows. For example, if comment='#', parsing ‘#empty\na,b,c\n1,2,3’ with header=0 will result in ‘a,b,c’ being treated as the header.

encodingstr, default None

Encoding to use for UTF when reading/writing (e.g. 'utf-8'). List of Python standard encodings.

dialectstr or csv.Dialect instance, default None

If provided, this parameter will override values (default or not) for the following parameters: delimiter, doublequote, escapechar, skipinitialspace, quotechar, and quoting. If it is necessary to override values, a ParserWarning will be issued. See csv.Dialect documentation for more details.

on_bad_lines(‘error’, ‘warn’, ‘skip’), default ‘error’

Specifies what to do upon encountering a bad line (a line with too many fields). Allowed values are :

• ‘error’, raise an ParserError when a bad line is encountered.

• ‘warn’, print a warning when a bad line is encountered and skip that line.

• ‘skip’, skip bad lines without raising or warning when they are encountered.

Added in version 1.3.0.

You can indicate the data type for the whole DataFrame or individual columns:

In [9]: import numpy as np

In [10]: data = "a,b,c,d\n1,2,3,4\n5,6,7,8\n9,10,11"

In [11]: print(data)
a,b,c,d
1,2,3,4
5,6,7,8
9,10,11

In [12]: df = pd.read_csv(StringIO(data), dtype=object)

In [13]: df
Out[13]: 
   a   b   c    d
0  1   2   3    4
1  5   6   7    8
2  9  10  11  NaN

In [14]: df["a"][0]
Out[14]: '1'

In [15]: df = pd.read_csv(StringIO(data), dtype={"b": object, "c": np.float64, "d": "Int64"})

In [16]: df.dtypes
Out[16]: 
a      int64
b     object
c    float64
d      Int64
dtype: object

Fortunately, pandas offers more than one way to ensure that your column(s) contain only one dtype. If you’re unfamiliar with these concepts, you can see here to learn more about dtypes, and here to learn more about object conversion in pandas.

For instance, you can use the converters argument of read_csv():

In [17]: data = "col_1\n1\n2\n'A'\n4.22"

In [18]: df = pd.read_csv(StringIO(data), converters={"col_1": str})

In [19]: df
Out[19]: 
  col_1
0     1
1     2
2   'A'
3  4.22

In [20]: df["col_1"].apply(type).value_counts()
Out[20]: 
col_1
<class 'str'>    4
Name: count, dtype: int64

Or you can use the to_numeric() function to coerce the dtypes after reading in the data,

In [21]: df2 = pd.read_csv(StringIO(data))

In [22]: df2["col_1"] = pd.to_numeric(df2["col_1"], errors="coerce")

In [23]: df2
Out[23]: 
   col_1
0   1.00
1   2.00
2    NaN
3   4.22

In [24]: df2["col_1"].apply(type).value_counts()
Out[24]: 
col_1
<class 'float'>    4
Name: count, dtype: int64

which will convert all valid parsing to floats, leaving the invalid parsing as NaN.

Ultimately, how you deal with reading in columns containing mixed dtypes depends on your specific needs. In the case above, if you wanted to NaN out the data anomalies, then to_numeric() is probably your best option. However, if you wanted for all the data to be coerced, no matter the type, then using the converters argument of read_csv() would certainly be worth trying.

Note

In some cases, reading in abnormal data with columns containing mixed dtypes will result in an inconsistent dataset. If you rely on pandas to infer the dtypes of your columns, the parsing engine will go and infer the dtypes for different chunks of the data, rather than the whole dataset at once. Consequently, you can end up with column(s) with mixed dtypes. For example,

In [25]: col_1 = list(range(500000)) + ["a", "b"] + list(range(500000))

In [26]: df = pd.DataFrame({"col_1": col_1})

In [27]: df.to_csv("foo.csv")

In [28]: mixed_df = pd.read_csv("foo.csv")

In [29]: mixed_df["col_1"].apply(type).value_counts()
Out[29]: 
col_1
<class 'int'>    737858
<class 'str'>    262144
Name: count, dtype: int64

In [30]: mixed_df["col_1"].dtype
Out[30]: dtype('O')

will result with mixed_df containing an int dtype for certain chunks of the column, and str for others due to the mixed dtypes from the data that was read in. It is important to note that the overall column will be marked with a dtype of object, which is used for columns with mixed dtypes.

Setting dtype_backend="numpy_nullable" will result in nullable dtypes for every column.

In [31]: data = """a,b,c,d,e,f,g,h,i,j
   ....: 1,2.5,True,a,,,,,12-31-2019,
   ....: 3,4.5,False,b,6,7.5,True,a,12-31-2019,
   ....: """
   ....: 

In [32]: df = pd.read_csv(StringIO(data), dtype_backend="numpy_nullable", parse_dates=["i"])

In [33]: df
Out[33]: 
   a    b      c  d     e     f     g     h          i     j
0  1  2.5   True  a  <NA>  <NA>  <NA>  <NA> 2019-12-31  <NA>
1  3  4.5  False  b     6   7.5  True     a 2019-12-31  <NA>

In [34]: df.dtypes
Out[34]: 
a             Int64
b           Float64
c           boolean
d    string[python]
e             Int64
f           Float64
g           boolean
h    string[python]
i    datetime64[ns]
j             Int64
dtype: object

Categorical columns can be parsed directly by specifying dtype='category' or dtype=CategoricalDtype(categories, ordered).

In [35]: data = "col1,col2,col3\na,b,1\na,b,2\nc,d,3"

In [36]: pd.read_csv(StringIO(data))
Out[36]: 
  col1 col2  col3
0    a    b     1
1    a    b     2
2    c    d     3

In [37]: pd.read_csv(StringIO(data)).dtypes
Out[37]: 
col1    object
col2    object
col3     int64
dtype: object

In [38]: pd.read_csv(StringIO(data), dtype="category").dtypes
Out[38]: 
col1    category
col2    category
col3    category
dtype: object

Individual columns can be parsed as a Categorical using a dict specification:

In [39]: pd.read_csv(StringIO(data), dtype={"col1": "category"}).dtypes
Out[39]: 
col1    category
col2      object
col3       int64
dtype: object

Specifying dtype='category' will result in an unordered Categorical whose categories are the unique values observed in the data. For more control on the categories and order, create a CategoricalDtype ahead of time, and pass that for that column’s dtype.

In [40]: from pandas.api.types import CategoricalDtype

In [41]: dtype = CategoricalDtype(["d", "c", "b", "a"], ordered=True)

In [42]: pd.read_csv(StringIO(data), dtype={"col1": dtype}).dtypes
Out[42]: 
col1    category
col2      object
col3       int64
dtype: object

When using dtype=CategoricalDtype, “unexpected” values outside of dtype.categories are treated as missing values.

In [43]: dtype = CategoricalDtype(["a", "b", "d"])  # No 'c'

In [44]: pd.read_csv(StringIO(data), dtype={"col1": dtype}).col1
Out[44]: 
0      a
1      a
2    NaN
Name: col1, dtype: category
Categories (3, object): ['a', 'b', 'd']

This matches the behavior of Categorical.set_categories().

Note

With dtype='category', the resulting categories will always be parsed as strings (object dtype). If the categories are numeric they can be converted using the to_numeric() function, or as appropriate, another converter such as to_datetime().

When dtype is a CategoricalDtype with homogeneous categories ( all numeric, all datetimes, etc.), the conversion is done automatically.

In [45]: df = pd.read_csv(StringIO(data), dtype="category")

In [46]: df.dtypes
Out[46]: 
col1    category
col2    category
col3    category
dtype: object

In [47]: df["col3"]
Out[47]: 
0    1
1    2
2    3
Name: col3, dtype: category
Categories (3, object): ['1', '2', '3']

In [48]: new_categories = pd.to_numeric(df["col3"].cat.categories)

In [49]: df["col3"] = df["col3"].cat.rename_categories(new_categories)

In [50]: df["col3"]
Out[50]: 
0    1
1    2
2    3
Name: col3, dtype: category
Categories (3, int64): [1, 2, 3]

A file may or may not have a header row. pandas assumes the first row should be used as the column names:

In [51]: data = "a,b,c\n1,2,3\n4,5,6\n7,8,9"

In [52]: print(data)
a,b,c
1,2,3
4,5,6
7,8,9

In [53]: pd.read_csv(StringIO(data))
Out[53]: 
   a  b  c
0  1  2  3
1  4  5  6
2  7  8  9

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 [54]: print(data)
a,b,c
1,2,3
4,5,6
7,8,9

In [55]: pd.read_csv(StringIO(data), names=["foo", "bar", "baz"], header=0)
Out[55]: 
   foo  bar  baz
0    1    2    3
1    4    5    6
2    7    8    9

In [56]: pd.read_csv(StringIO(data), names=["foo", "bar", "baz"], header=None)
Out[56]: 
  foo bar baz
0   a   b   c
1   1   2   3
2   4   5   6
3   7   8   9

If the header is in a row other than the first, pass the row number to header. This will skip the preceding rows:

In [57]: data = "skip this skip it\na,b,c\n1,2,3\n4,5,6\n7,8,9"

In [58]: pd.read_csv(StringIO(data), header=1)
Out[58]: 
   a  b  c
0  1  2  3
1  4  5  6
2  7  8  9

Note

Default behavior is to infer the column names: if no names are passed the behavior is identical to header=0 and column names are inferred from the first non-blank line of the file, if column names are passed explicitly then the behavior is identical to header=None.

If the file or header contains duplicate names, pandas will by default distinguish between them so as to prevent overwriting data:

In [59]: data = "a,b,a\n0,1,2\n3,4,5"

In [60]: pd.read_csv(StringIO(data))
Out[60]: 
   a  b  a.1
0  0  1    2
1  3  4    5

There is no more duplicate data because duplicate columns ‘X’, …, ‘X’ become ‘X’, ‘X.1’, …, ‘X.N’.

The usecols argument allows you to select any subset of the columns in a file, either using the column names, position numbers or a callable:

In [61]: data = "a,b,c,d\n1,2,3,foo\n4,5,6,bar\n7,8,9,baz"

In [62]: pd.read_csv(StringIO(data))
Out[62]: 
   a  b  c    d
0  1  2  3  foo
1  4  5  6  bar
2  7  8  9  baz

In [63]: pd.read_csv(StringIO(data), usecols=["b", "d"])
Out[63]: 
   b    d
0  2  foo
1  5  bar
2  8  baz

In [64]: pd.read_csv(StringIO(data), usecols=[0, 2, 3])
Out[64]: 
   a  c    d
0  1  3  foo
1  4  6  bar
2  7  9  baz

In [65]: pd.read_csv(StringIO(data), usecols=lambda x: x.upper() in ["A", "C"])
Out[65]: 
   a  c
0  1  3
1  4  6
2  7  9

The usecols argument can also be used to specify which columns not to use in the final result:

In [66]: pd.read_csv(StringIO(data), usecols=lambda x: x not in ["a", "c"])
Out[66]: 
   b    d
0  2  foo
1  5  bar
2  8  baz

In this case, the callable is specifying that we exclude the “a” and “c” columns from the output.

If the comment parameter is specified, then completely commented lines will be ignored. By default, completely blank lines will be ignored as well.

In [67]: data = "\na,b,c\n  \n# commented line\n1,2,3\n\n4,5,6"

In [68]: print(data)

a,b,c
  
# commented line
1,2,3

4,5,6

In [69]: pd.read_csv(StringIO(data), comment="#")
Out[69]: 
   a  b  c
0  1  2  3
1  4  5  6

If skip_blank_lines=False, then read_csv will not ignore blank lines:

In [70]: data = "a,b,c\n\n1,2,3\n\n\n4,5,6"

In [71]: pd.read_csv(StringIO(data), skip_blank_lines=False)
Out[71]: 
     a    b    c
0  NaN  NaN  NaN
1  1.0  2.0  3.0
2  NaN  NaN  NaN
3  NaN  NaN  NaN
4  4.0  5.0  6.0

Warning

The presence of ignored lines might create ambiguities involving line numbers; the parameter header uses row numbers (ignoring commented/empty lines), while skiprows uses line numbers (including commented/empty lines):

In [72]: data = "#comment\na,b,c\nA,B,C\n1,2,3"

In [73]: pd.read_csv(StringIO(data), comment="#", header=1)
Out[73]: 
   A  B  C
0  1  2  3

In [74]: data = "A,B,C\n#comment\na,b,c\n1,2,3"

In [75]: pd.read_csv(StringIO(data), comment="#", skiprows=2)
Out[75]: 
   a  b  c
0  1  2  3

If both header and skiprows are specified, header will be relative to the end of skiprows. For example:

In [76]: data = (
   ....:     "# empty\n"
   ....:     "# second empty line\n"
   ....:     "# third emptyline\n"
   ....:     "X,Y,Z\n"
   ....:     "1,2,3\n"
   ....:     "A,B,C\n"
   ....:     "1,2.,4.\n"
   ....:     "5.,NaN,10.0\n"
   ....: )
   ....: 

In [77]: print(data)
# empty
# second empty line
# third emptyline
X,Y,Z
1,2,3
A,B,C
1,2.,4.
5.,NaN,10.0


In [78]: pd.read_csv(StringIO(data), comment="#", skiprows=4, header=1)
Out[78]: 
     A    B     C
0  1.0  2.0   4.0
1  5.0  NaN  10.0

The encoding argument should be used for encoded unicode data, which will result in byte strings being decoded to unicode in the result:

In [86]: from io import BytesIO

In [87]: data = b"word,length\n" b"Tr\xc3\xa4umen,7\n" b"Gr\xc3\xbc\xc3\x9fe,5"

In [88]: data = data.decode("utf8").encode("latin-1")

In [89]: df = pd.read_csv(BytesIO(data), encoding="latin-1")

In [90]: df
Out[90]: 
      word  length
0  Träumen       7
1    Grüße       5

In [91]: df["word"][1]
Out[91]: 'Grüße'

Some formats which encode all characters as multiple bytes, like UTF-16, won’t parse correctly at all without specifying the encoding. Full list of Python standard encodings.

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 [92]: data = "a,b,c\n4,apple,bat,5.7\n8,orange,cow,10"

In [93]: pd.read_csv(StringIO(data))
Out[93]: 
        a    b     c
4   apple  bat   5.7
8  orange  cow  10.0

In [94]: data = "index,a,b,c\n4,apple,bat,5.7\n8,orange,cow,10"

In [95]: pd.read_csv(StringIO(data), index_col=0)
Out[95]: 
            a    b     c
index                   
4       apple  bat   5.7
8      orange  cow  10.0

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 [96]: data = "a,b,c\n4,apple,bat,\n8,orange,cow,"

In [97]: print(data)
a,b,c
4,apple,bat,
8,orange,cow,

In [98]: pd.read_csv(StringIO(data))
Out[98]: 
        a    b   c
4   apple  bat NaN
8  orange  cow NaN

In [99]: pd.read_csv(StringIO(data), index_col=False)
Out[99]: 
   a       b    c
0  4   apple  bat
1  8  orange  cow

If a subset of data is being parsed using the usecols option, the index_col specification is based on that subset, not the original data.

In [100]: data = "a,b,c\n4,apple,bat,\n8,orange,cow,"

In [101]: print(data)
a,b,c
4,apple,bat,
8,orange,cow,

In [102]: pd.read_csv(StringIO(data), usecols=["b", "c"])
Out[102]: 
     b   c
4  bat NaN
8  cow NaN

In [103]: pd.read_csv(StringIO(data), usecols=["b", "c"], index_col=0)
Out[103]: 
     b   c
4  bat NaN
8  cow NaN

To better facilitate working with datetime data, read_csv() uses the keyword arguments parse_dates and date_format 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:

In [104]: with open("foo.csv", mode="w") as f:
   .....:     f.write("date,A,B,C\n20090101,a,1,2\n20090102,b,3,4\n20090103,c,4,5")
   .....: 

# Use a column as an index, and parse it as dates.
In [105]: df = pd.read_csv("foo.csv", index_col=0, parse_dates=True)

In [106]: df
Out[106]: 
            A  B  C
date               
2009-01-01  a  1  2
2009-01-02  b  3  4
2009-01-03  c  4  5

# These are Python datetime objects
In [107]: df.index
Out[107]: DatetimeIndex(['2009-01-01', '2009-01-02', '2009-01-03'], dtype='datetime64[ns]', name='date', freq=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 [108]: data = (
   .....:     "KORD,19990127, 19:00:00, 18:56:00, 0.8100\n"
   .....:     "KORD,19990127, 20:00:00, 19:56:00, 0.0100\n"
   .....:     "KORD,19990127, 21:00:00, 20:56:00, -0.5900\n"
   .....:     "KORD,19990127, 21:00:00, 21:18:00, -0.9900\n"
   .....:     "KORD,19990127, 22:00:00, 21:56:00, -0.5900\n"
   .....:     "KORD,19990127, 23:00:00, 22:56:00, -0.5900"
   .....: )
   .....: 

In [109]: with open("tmp.csv", "w") as fh:
   .....:     fh.write(data)
   .....: 

In [110]: df = pd.read_csv("tmp.csv", header=None, parse_dates=[[1, 2], [1, 3]])

In [111]: df
Out[111]: 
                  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

By default the parser removes the component date columns, but you can choose to retain them via the keep_date_col keyword:

In [112]: df = pd.read_csv(
   .....:     "tmp.csv", header=None, parse_dates=[[1, 2], [1, 3]], keep_date_col=True
   .....: )
   .....: 

In [113]: df
Out[113]: 
                  1_2                 1_3     0  ...          2          3     4
0 1999-01-27 19:00:00 1999-01-27 18:56:00  KORD  ...   19:00:00   18:56:00  0.81
1 1999-01-27 20:00:00 1999-01-27 19:56:00  KORD  ...   20:00:00   19:56:00  0.01
2 1999-01-27 21:00:00 1999-01-27 20:56:00  KORD  ...   21:00:00   20:56:00 -0.59
3 1999-01-27 21:00:00 1999-01-27 21:18:00  KORD  ...   21:00:00   21:18:00 -0.99
4 1999-01-27 22:00:00 1999-01-27 21:56:00  KORD  ...   22:00:00   21:56:00 -0.59
5 1999-01-27 23:00:00 1999-01-27 22:56:00  KORD  ...   23:00:00   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 [114]: date_spec = {"nominal": [1, 2], "actual": [1, 3]}

In [115]: df = pd.read_csv("tmp.csv", header=None, parse_dates=date_spec)

In [116]: df
Out[116]: 
              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

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 [117]: date_spec = {"nominal": [1, 2], "actual": [1, 3]}

In [118]: df = pd.read_csv(
   .....:     "tmp.csv", header=None, parse_dates=date_spec, index_col=0
   .....: )  # index is the nominal column
   .....: 

In [119]: df
Out[119]: 
                                 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

Note

If a column or index contains an unparsable date, the entire column or index will be returned unaltered as an object data type. For non-standard datetime parsing, use to_datetime() after pd.read_csv.

Note

read_csv has a fast_path for parsing datetime strings in iso8601 format, e.g “2000-01-01T00:01:02+00:00” and similar variations. If you can arrange for your data to store datetimes in this format, load times will be significantly faster, ~20x has been observed.

Deprecated since version 2.2.0: Combining date columns inside read_csv is deprecated. Use pd.to_datetime on the relevant result columns instead.

Finally, the parser allows you to specify a custom date_format. Performance-wise, you should try these methods of parsing dates in order:

1. If you know the format, use date_format, e.g.: date_format="%d/%m/%Y" or date_format={column_name: "%d/%m/%Y"}.

2. If you different formats for different columns, or want to pass any extra options (such as utc) to to_datetime, then you should read in your data as object dtype, and then use to_datetime.

pandas cannot natively represent a column or index with mixed timezones. If your CSV file contains columns with a mixture of timezones, the default result will be an object-dtype column with strings, even with parse_dates. To parse the mixed-timezone values as a datetime column, read in as object dtype and then call to_datetime() with utc=True.

In [120]: content = """\
   .....: a
   .....: 2000-01-01T00:00:00+05:00
   .....: 2000-01-01T00:00:00+06:00"""
   .....: 

In [121]: df = pd.read_csv(StringIO(content))

In [122]: df["a"] = pd.to_datetime(df["a"], utc=True)

In [123]: df["a"]
Out[123]: 
0   1999-12-31 19:00:00+00:00
1   1999-12-31 18:00:00+00:00
Name: a, dtype: datetime64[ns, UTC]

Here are some examples of datetime strings that can be guessed (all representing December 30th, 2011 at 00:00:00):

• “20111230”

• “2011/12/30”

• “20111230 00:00:00”

• “12/30/2011 00:00:00”

• “30/Dec/2011 00:00:00”

• “30/December/2011 00:00:00”

Note that format inference is sensitive to dayfirst. With dayfirst=True, it will guess “01/12/2011” to be December 1st. With dayfirst=False (default) it will guess “01/12/2011” to be January 12th.

If you try to parse a column of date strings, pandas will attempt to guess the format from the first non-NaN element, and will then parse the rest of the column with that format. If pandas fails to guess the format (for example if your first string is '01 December US/Pacific 2000'), then a warning will be raised and each row will be parsed individually by dateutil.parser.parse. The safest way to parse dates is to explicitly set format=.

In [124]: df = pd.read_csv(
   .....:     "foo.csv",
   .....:     index_col=0,
   .....:     parse_dates=True,
   .....: )
   .....: 

In [125]: df
Out[125]: 
            A  B  C
date               
2009-01-01  a  1  2
2009-01-02  b  3  4
2009-01-03  c  4  5

In the case that you have mixed datetime formats within the same column, you can pass format='mixed'

In [126]: data = StringIO("date\n12 Jan 2000\n2000-01-13\n")

In [127]: df = pd.read_csv(data)

In [128]: df['date'] = pd.to_datetime(df['date'], format='mixed')

In [129]: df
Out[129]: 
        date
0 2000-01-12
1 2000-01-13

or, if your datetime formats are all ISO8601 (possibly not identically-formatted):

In [130]: data = StringIO("date\n2020-01-01\n2020-01-01 03:00\n")

In [131]: df = pd.read_csv(data)

In [132]: df['date'] = pd.to_datetime(df['date'], format='ISO8601')

In [133]: df
Out[133]: 
                 date
0 2020-01-01 00:00:00
1 2020-01-01 03:00:00

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 [134]: data = "date,value,cat\n1/6/2000,5,a\n2/6/2000,10,b\n3/6/2000,15,c"

In [135]: print(data)
date,value,cat
1/6/2000,5,a
2/6/2000,10,b
3/6/2000,15,c

In [136]: with open("tmp.csv", "w") as fh:
   .....:     fh.write(data)
   .....: 

In [137]: pd.read_csv("tmp.csv", parse_dates=[0])
Out[137]: 
        date  value cat
0 2000-01-06      5   a
1 2000-02-06     10   b
2 2000-03-06     15   c

In [138]: pd.read_csv("tmp.csv", dayfirst=True, parse_dates=[0])
Out[138]: 
        date  value cat
0 2000-06-01      5   a
1 2000-06-02     10   b
2 2000-06-03     15   c

Added in version 1.2.0.

df.to_csv(..., mode="wb") allows writing a CSV to a file object opened binary mode. In most cases, it is not necessary to specify mode as Pandas will auto-detect whether the file object is opened in text or binary mode.

In [139]: import io

In [140]: data = pd.DataFrame([0, 1, 2])

In [141]: buffer = io.BytesIO()

In [142]: data.to_csv(buffer, encoding="utf-8", compression="gzip")

The parameter float_precision can be specified in order to use a specific floating-point converter during parsing with the C engine. The options are the ordinary converter, the high-precision converter, and the round-trip converter (which is guaranteed to round-trip values after writing to a file). For example:

In [143]: val = "0.3066101993807095471566981359501369297504425048828125"

In [144]: data = "a,b,c\n1,2,{0}".format(val)

In [145]: abs(
   .....:     pd.read_csv(
   .....:         StringIO(data),
   .....:         engine="c",
   .....:         float_precision=None,
   .....:     )["c"][0] - float(val)
   .....: )
   .....: 
Out[145]: 5.551115123125783e-17

In [146]: abs(
   .....:     pd.read_csv(
   .....:         StringIO(data),
   .....:         engine="c",
   .....:         float_precision="high",
   .....:     )["c"][0] - float(val)
   .....: )
   .....: 
Out[146]: 5.551115123125783e-17

In [147]: abs(
   .....:     pd.read_csv(StringIO(data), engine="c", float_precision="round_trip")["c"][0]
   .....:     - float(val)
   .....: )
   .....: 
Out[147]: 0.0

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 [148]: data = (
   .....:     "ID|level|category\n"
   .....:     "Patient1|123,000|x\n"
   .....:     "Patient2|23,000|y\n"
   .....:     "Patient3|1,234,018|z"
   .....: )
   .....: 

In [149]: with open("tmp.csv", "w") as fh:
   .....:     fh.write(data)
   .....: 

In [150]: df = pd.read_csv("tmp.csv", sep="|")

In [151]: df
Out[151]: 
         ID      level category
0  Patient1    123,000        x
1  Patient2     23,000        y
2  Patient3  1,234,018        z

In [152]: df.level.dtype
Out[152]: dtype('O')

The thousands keyword allows integers to be parsed correctly:

In [153]: df = pd.read_csv("tmp.csv", sep="|", thousands=",")

In [154]: df
Out[154]: 
         ID    level category
0  Patient1   123000        x
1  Patient2    23000        y
2  Patient3  1234018        z

In [155]: df.level.dtype
Out[155]: dtype('int64')

To control which values are parsed as missing values (which are signified by NaN), specify a string in na_values. If you specify a list of strings, then all values in it are considered to be missing 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', '#N/A', 'N/A', 'n/a', 'NA', '<NA>', '#NA', 'NULL', 'null', 'NaN', '-NaN', 'nan', '-nan', 'None', ''].

Let us consider some examples:

pd.read_csv("path_to_file.csv", na_values=[5])

In the example above 5 and 5.0 will be recognized as NaN, in addition to the defaults. A string will first be interpreted as a numerical 5, then as a NaN.

pd.read_csv("path_to_file.csv", keep_default_na=False, na_values=[""])

Above, only an empty field will be recognized as NaN.

pd.read_csv("path_to_file.csv", keep_default_na=False, na_values=["NA", "0"])

Above, both NA and 0 as strings are NaN.

pd.read_csv("path_to_file.csv", na_values=["Nope"])

The default values, in addition to the string "Nope" are recognized as NaN.

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.

The common values True, False, TRUE, and FALSE are all recognized as boolean. Occasionally you might want to recognize other values as being boolean. To do this, use the true_values and false_values options as follows:

In [156]: data = "a,b,c\n1,Yes,2\n3,No,4"

In [157]: print(data)
a,b,c
1,Yes,2
3,No,4

In [158]: pd.read_csv(StringIO(data))
Out[158]: 
   a    b  c
0  1  Yes  2
1  3   No  4

In [159]: pd.read_csv(StringIO(data), true_values=["Yes"], false_values=["No"])
Out[159]: 
   a      b  c
0  1   True  2
1  3  False  4

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 fields will raise an error by default:

In [160]: data = "a,b,c\n1,2,3\n4,5,6,7\n8,9,10"

In [161]: pd.read_csv(StringIO(data))
---------------------------------------------------------------------------
ParserError                               Traceback (most recent call last)
Cell In[161], line 1
----> 1 pd.read_csv(StringIO(data))

File ~/work/pandas/pandas/pandas/io/parsers/readers.py:1026, in read_csv(filepath_or_buffer, sep, delimiter, header, names, index_col, usecols, dtype, engine, converters, true_values, false_values, skipinitialspace, skiprows, skipfooter, nrows, na_values, keep_default_na, na_filter, verbose, skip_blank_lines, parse_dates, infer_datetime_format, keep_date_col, date_parser, date_format, dayfirst, cache_dates, iterator, chunksize, compression, thousands, decimal, lineterminator, quotechar, quoting, doublequote, escapechar, comment, encoding, encoding_errors, dialect, on_bad_lines, delim_whitespace, low_memory, memory_map, float_precision, storage_options, dtype_backend)
   1013 kwds_defaults = _refine_defaults_read(
   1014     dialect,
   1015     delimiter,
   (...)
   1022     dtype_backend=dtype_backend,
   1023 )
   1024 kwds.update(kwds_defaults)
-> 1026 return _read(filepath_or_buffer, kwds)

File ~/work/pandas/pandas/pandas/io/parsers/readers.py:626, in _read(filepath_or_buffer, kwds)
    623     return parser
    625 with parser:
--> 626     return parser.read(nrows)

File ~/work/pandas/pandas/pandas/io/parsers/readers.py:1923, in TextFileReader.read(self, nrows)
   1916 nrows = validate_integer("nrows", nrows)
   1917 try:
   1918     # error: "ParserBase" has no attribute "read"
   1919     (
   1920         index,
   1921         columns,
   1922         col_dict,
-> 1923     ) = self._engine.read(  # type: ignore[attr-defined]
   1924         nrows
   1925     )
   1926 except Exception:
   1927     self.close()

File ~/work/pandas/pandas/pandas/io/parsers/c_parser_wrapper.py:234, in CParserWrapper.read(self, nrows)
    232 try:
    233     if self.low_memory:
--> 234         chunks = self._reader.read_low_memory(nrows)
    235         # destructive to chunks
    236         data = _concatenate_chunks(chunks)

File ~/work/pandas/pandas/pandas/_libs/parsers.pyx:838, in pandas._libs.parsers.TextReader.read_low_memory()

File ~/work/pandas/pandas/pandas/_libs/parsers.pyx:905, in pandas._libs.parsers.TextReader._read_rows()

File ~/work/pandas/pandas/pandas/_libs/parsers.pyx:874, in pandas._libs.parsers.TextReader._tokenize_rows()

File ~/work/pandas/pandas/pandas/_libs/parsers.pyx:891, in pandas._libs.parsers.TextReader._check_tokenize_status()

File ~/work/pandas/pandas/pandas/_libs/parsers.pyx:2061, in pandas._libs.parsers.raise_parser_error()

ParserError: Error tokenizing data. C error: Expected 3 fields in line 3, saw 4

You can elect to skip bad lines:

In [162]: data = "a,b,c\n1,2,3\n4,5,6,7\n8,9,10"

In [163]: pd.read_csv(StringIO(data), on_bad_lines="skip")
Out[163]: 
   a  b   c
0  1  2   3
1  8  9  10

Added in version 1.4.0.

Or pass a callable function to handle the bad line if engine="python". The bad line will be a list of strings that was split by the sep:

In [164]: external_list = []

In [165]: def bad_lines_func(line):
   .....:     external_list.append(line)
   .....:     return line[-3:]
   .....: 

In [166]: external_list
Out[166]: []

Note

The callable function will handle only a line with too many fields. Bad lines caused by other errors will be silently skipped.

In [167]: bad_lines_func = lambda line: print(line)

In [168]: data = 'name,type\nname a,a is of type a\nname b,"b\" is of type b"'

In [169]: data
Out[169]: 'name,type\nname a,a is of type a\nname b,"b" is of type b"'

In [170]: pd.read_csv(StringIO(data), on_bad_lines=bad_lines_func, engine="python")
Out[170]: 
     name            type
0  name a  a is of type a

The line was not processed in this case, as a “bad line” here is caused by an escape character.

You can also use the usecols parameter to eliminate extraneous column data that appear in some lines but not others:

In [171]: pd.read_csv(StringIO(data), usecols=[0, 1, 2])
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[171], line 1
----> 1 pd.read_csv(StringIO(data), usecols=[0, 1, 2])

File ~/work/pandas/pandas/pandas/io/parsers/readers.py:1026, in read_csv(filepath_or_buffer, sep, delimiter, header, names, index_col, usecols, dtype, engine, converters, true_values, false_values, skipinitialspace, skiprows, skipfooter, nrows, na_values, keep_default_na, na_filter, verbose, skip_blank_lines, parse_dates, infer_datetime_format, keep_date_col, date_parser, date_format, dayfirst, cache_dates, iterator, chunksize, compression, thousands, decimal, lineterminator, quotechar, quoting, doublequote, escapechar, comment, encoding, encoding_errors, dialect, on_bad_lines, delim_whitespace, low_memory, memory_map, float_precision, storage_options, dtype_backend)
   1013 kwds_defaults = _refine_defaults_read(
   1014     dialect,
   1015     delimiter,
   (...)
   1022     dtype_backend=dtype_backend,
   1023 )
   1024 kwds.update(kwds_defaults)
-> 1026 return _read(filepath_or_buffer, kwds)

File ~/work/pandas/pandas/pandas/io/parsers/readers.py:620, in _read(filepath_or_buffer, kwds)
    617 _validate_names(kwds.get("names", None))
    619 # Create the parser.
--> 620 parser = TextFileReader(filepath_or_buffer, **kwds)
    622 if chunksize or iterator:
    623     return parser

File ~/work/pandas/pandas/pandas/io/parsers/readers.py:1620, in TextFileReader.__init__(self, f, engine, **kwds)
   1617     self.options["has_index_names"] = kwds["has_index_names"]
   1619 self.handles: IOHandles | None = None
-> 1620 self._engine = self._make_engine(f, self.engine)

File ~/work/pandas/pandas/pandas/io/parsers/readers.py:1898, in TextFileReader._make_engine(self, f, engine)
   1895     raise ValueError(msg)
   1897 try:
-> 1898     return mapping[engine](f, **self.options)
   1899 except Exception:
   1900     if self.handles is not None:

File ~/work/pandas/pandas/pandas/io/parsers/c_parser_wrapper.py:155, in CParserWrapper.__init__(self, src, **kwds)
    152     # error: Cannot determine type of 'names'
    153     if len(self.names) < len(usecols):  # type: ignore[has-type]
    154         # error: Cannot determine type of 'names'
--> 155         self._validate_usecols_names(
    156             usecols,
    157             self.names,  # type: ignore[has-type]
    158         )
    160 # error: Cannot determine type of 'names'
    161 self._validate_parse_dates_presence(self.names)  # type: ignore[has-type]

File ~/work/pandas/pandas/pandas/io/parsers/base_parser.py:988, in ParserBase._validate_usecols_names(self, usecols, names)
    986 missing = [c for c in usecols if c not in names]
    987 if len(missing) > 0:
--> 988     raise ValueError(
    989         f"Usecols do not match columns, columns expected but not found: "
    990         f"{missing}"
    991     )
    993 return usecols

ValueError: Usecols do not match columns, columns expected but not found: [0, 1, 2]

In case you want to keep all data including the lines with too many fields, you can specify a sufficient number of names. This ensures that lines with not enough fields are filled with NaN.

In [172]: pd.read_csv(StringIO(data), names=['a', 'b', 'c', 'd'])
Out[172]: 
        a                b   c   d
0    name             type NaN NaN
1  name a   a is of type a NaN NaN
2  name b  b is of type b" NaN NaN

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 [173]: data = "label1,label2,label3\n" 'index1,"a,c,e\n' "index2,b,d,f"

In [174]: 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 [175]: import csv

In [176]: dia = csv.excel()

In [177]: dia.quoting = csv.QUOTE_NONE

In [178]: pd.read_csv(StringIO(data), dialect=dia)
Out[178]: 
       label1 label2 label3
index1     "a      c      e
index2      b      d      f

All of the dialect options can be specified separately by keyword arguments:

In [179]: data = "a,b,c~1,2,3~4,5,6"

In [180]: pd.read_csv(StringIO(data), lineterminator="~")
Out[180]: 
   a  b  c
0  1  2  3
1  4  5  6

Another common dialect option is skipinitialspace, to skip any whitespace after a delimiter:

In [181]: data = "a, b, c\n1, 2, 3\n4, 5, 6"

In [182]: print(data)
a, b, c
1, 2, 3
4, 5, 6

In [183]: pd.read_csv(StringIO(data), skipinitialspace=True)
Out[183]: 
   a  b  c
0  1  2  3
1  4  5  6

The parsers make every attempt to “do the right thing” and not be fragile. Type inference is a pretty big deal. If a column can be coerced to integer dtype without altering the contents, the parser will do so. Any non-numeric columns will come through as object dtype as with the rest of pandas objects.

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 [184]: data = 'a,b\n"hello, \\"Bob\\", nice to see you",5'

In [185]: print(data)
a,b
"hello, \"Bob\", nice to see you",5

In [186]: pd.read_csv(StringIO(data), escapechar="\\")
Out[186]: 
                               a  b
0  hello, "Bob", nice to see you  5

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, and a different usage of the delimiter parameter:

• 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 behavior, if not specified, is to infer.

• widths: A list of field widths which can be used instead of ‘colspecs’ if the intervals are contiguous.

• delimiter: Characters to consider as filler characters in the fixed-width file. Can be used to specify the filler character of the fields if it is not spaces (e.g., ‘~’).

Consider a typical fixed-width data file:

In [187]: data1 = (
   .....:     "id8141    360.242940   149.910199   11950.7\n"
   .....:     "id1594    444.953632   166.985655   11788.4\n"
   .....:     "id1849    364.136849   183.628767   11806.2\n"
   .....:     "id1230    413.836124   184.375703   11916.8\n"
   .....:     "id1948    502.953953   173.237159   12468.3"
   .....: )
   .....: 

In [188]: with open("bar.csv", "w") as f:
   .....:     f.write(data1)
   .....: 

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 [189]: colspecs = [(0, 6), (8, 20), (21, 33), (34, 43)]

In [190]: df = pd.read_fwf("bar.csv", colspecs=colspecs, header=None, index_col=0)

In [191]: df
Out[191]: 
                 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

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 [192]: widths = [6, 14, 13, 10]

In [193]: df = pd.read_fwf("bar.csv", widths=widths, header=None)

In [194]: df
Out[194]: 
        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

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.

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 [195]: df = pd.read_fwf("bar.csv", header=None, index_col=0)

In [196]: df
Out[196]: 
                 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

read_fwf supports the dtype parameter for specifying the types of parsed columns to be different from the inferred type.

In [197]: pd.read_fwf("bar.csv", header=None, index_col=0).dtypes
Out[197]: 
1    float64
2    float64
3    float64
dtype: object

In [198]: pd.read_fwf("bar.csv", header=None, dtype={2: "object"}).dtypes
Out[198]: 
0     object
1    float64
2     object
3    float64
dtype: object

Consider a file with one less entry in the header than the number of data column:

In [199]: data = "A,B,C\n20090101,a,1,2\n20090102,b,3,4\n20090103,c,4,5"

In [200]: print(data)
A,B,C
20090101,a,1,2
20090102,b,3,4
20090103,c,4,5

In [201]: with open("foo.csv", "w") as f:
   .....:     f.write(data)
   .....: 

In this special case, read_csv assumes that the first column is to be used as the index of the DataFrame:

In [202]: pd.read_csv("foo.csv")
Out[202]: 
          A  B  C
20090101  a  1  2
20090102  b  3  4
20090103  c  4  5

Note that the dates weren’t automatically parsed. In that case you would need to do as before:

In [203]: df = pd.read_csv("foo.csv", parse_dates=True)

In [204]: df.index
Out[204]: DatetimeIndex(['2009-01-01', '2009-01-02', '2009-01-03'], dtype='datetime64[ns]', freq=None)

Suppose you have data indexed by two columns:

In [205]: data = 'year,indiv,zit,xit\n1977,"A",1.2,.6\n1977,"B",1.5,.5'

In [206]: print(data)
year,indiv,zit,xit
1977,"A",1.2,.6
1977,"B",1.5,.5

In [207]: with open("mindex_ex.csv", mode="w") as f:
   .....:     f.write(data)
   .....: 

The index_col argument to read_csv can take a list of column numbers to turn multiple columns into a MultiIndex for the index of the returned object:

In [208]: df = pd.read_csv("mindex_ex.csv", index_col=[0, 1])

In [209]: df
Out[209]: 
            zit  xit
year indiv          
1977 A      1.2  0.6
     B      1.5  0.5

In [210]: df.loc[1977]
Out[210]: 
       zit  xit
indiv          
A      1.2  0.6
B      1.5  0.5

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 intervening rows.

In [211]: mi_idx = pd.MultiIndex.from_arrays([[1, 2, 3, 4], list("abcd")], names=list("ab"))

In [212]: mi_col = pd.MultiIndex.from_arrays([[1, 2], list("ab")], names=list("cd"))

In [213]: df = pd.DataFrame(np.ones((4, 2)), index=mi_idx, columns=mi_col)

In [214]: df.to_csv("mi.csv")

In [215]: print(open("mi.csv").read())
c,,1,2
d,,a,b
a,b,,
1,a,1.0,1.0
2,b,1.0,1.0
3,c,1.0,1.0
4,d,1.0,1.0


In [216]: pd.read_csv("mi.csv", header=[0, 1, 2, 3], index_col=[0, 1])
Out[216]: 
c                    1                  2
d                    a                  b
a   Unnamed: 2_level_2 Unnamed: 3_level_2
1                  1.0                1.0
2 b                1.0                1.0
3 c                1.0                1.0
4 d                1.0                1.0

read_csv is also able to interpret a more common format of multi-columns indices.

In [217]: data = ",a,a,a,b,c,c\n,q,r,s,t,u,v\none,1,2,3,4,5,6\ntwo,7,8,9,10,11,12"

In [218]: print(data)
,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 [219]: with open("mi2.csv", "w") as fh:
   .....:     fh.write(data)
   .....: 

In [220]: pd.read_csv("mi2.csv", header=[0, 1], index_col=0)
Out[220]: 
     a         b   c    
     q  r  s   t   u   v
one  1  2  3   4   5   6
two  7  8  9  10  11  12

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.

read_csv is capable of inferring delimited (not necessarily comma-separated) files, as pandas uses the csv.Sniffer class of the csv module. For this, you have to specify sep=None.

In [221]: df = pd.DataFrame(np.random.randn(10, 4))

In [222]: df.to_csv("tmp2.csv", sep=":", index=False)

In [223]: pd.read_csv("tmp2.csv", sep=None, engine="python")
Out[223]: 
          0         1         2         3
0  0.469112 -0.282863 -1.509059 -1.135632
1  1.212112 -0.173215  0.119209 -1.044236
2 -0.861849 -2.104569 -0.494929  1.071804
3  0.721555 -0.706771 -1.039575  0.271860
4 -0.424972  0.567020  0.276232 -1.087401
5 -0.673690  0.113648 -1.478427  0.524988
6  0.404705  0.577046 -1.715002 -1.039268
7 -0.370647 -1.157892 -1.344312  0.844885
8  1.075770 -0.109050  1.643563 -1.469388
9  0.357021 -0.674600 -1.776904 -0.968914

It’s best to use concat() to combine multiple files. See the cookbook for an example.

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 [224]: df = pd.DataFrame(np.random.randn(10, 4))

In [225]: df.to_csv("tmp.csv", index=False)

In [226]: table = pd.read_csv("tmp.csv")

In [227]: table
Out[227]: 
          0         1         2         3
0 -1.294524  0.413738  0.276662 -0.472035
1 -0.013960 -0.362543 -0.006154 -0.923061
2  0.895717  0.805244 -1.206412  2.565646
3  1.431256  1.340309 -1.170299 -0.226169
4  0.410835  0.813850  0.132003 -0.827317
5 -0.076467 -1.187678  1.130127 -1.436737
6 -1.413681  1.607920  1.024180  0.569605
7  0.875906 -2.211372  0.974466 -2.006747
8 -0.410001 -0.078638  0.545952 -1.219217
9 -1.226825  0.769804 -1.281247 -0.727707

By specifying a chunksize to read_csv, the return value will be an iterable object of type TextFileReader:

In [228]: with pd.read_csv("tmp.csv", chunksize=4) as reader:
   .....:     print(reader)
   .....:     for chunk in reader:
   .....:         print(chunk)
   .....: 
<pandas.io.parsers.readers.TextFileReader object at 0x7f10516496c0>
          0         1         2         3
0 -1.294524  0.413738  0.276662 -0.472035
1 -0.013960 -0.362543 -0.006154 -0.923061
2  0.895717  0.805244 -1.206412  2.565646
3  1.431256  1.340309 -1.170299 -0.226169
          0         1         2         3
4  0.410835  0.813850  0.132003 -0.827317
5 -0.076467 -1.187678  1.130127 -1.436737
6 -1.413681  1.607920  1.024180  0.569605
7  0.875906 -2.211372  0.974466 -2.006747
          0         1         2         3
8 -0.410001 -0.078638  0.545952 -1.219217
9 -1.226825  0.769804 -1.281247 -0.727707

Changed in version 1.2: read_csv/json/sas return a context-manager when iterating through a file.

Specifying iterator=True will also return the TextFileReader object:

In [229]: with pd.read_csv("tmp.csv", iterator=True) as reader:
   .....:     print(reader.get_chunk(5))
   .....: 
          0         1         2         3
0 -1.294524  0.413738  0.276662 -0.472035
1 -0.013960 -0.362543 -0.006154 -0.923061
2  0.895717  0.805244 -1.206412  2.565646
3  1.431256  1.340309 -1.170299 -0.226169
4  0.410835  0.813850  0.132003 -0.827317

Pandas currently supports three engines, the C engine, the python engine, and an experimental pyarrow engine (requires the pyarrow package). In general, the pyarrow engine is fastest on larger workloads and is equivalent in speed to the C engine on most other workloads. The python engine tends to be slower than the pyarrow and C engines on most workloads. However, the pyarrow engine is much less robust than the C engine, which lacks a few features compared to the Python engine.

Where possible, pandas uses the C parser (specified as engine='c'), but it may fall back to Python if C-unsupported options are specified.

Currently, options unsupported by the C and pyarrow engines include:

• sep other than a single character (e.g. regex separators)

• skipfooter

• sep=None with delim_whitespace=False

Specifying any of the above options will produce a ParserWarning unless the python engine is selected explicitly using engine='python'.

Options that are unsupported by the pyarrow engine which are not covered by the list above include:

• float_precision

• chunksize

• comment

• nrows

• thousands

• memory_map

• dialect

• on_bad_lines

• delim_whitespace

• quoting

• lineterminator

• converters

• decimal

• iterator

• dayfirst

• infer_datetime_format

• verbose

• skipinitialspace

• low_memory

Specifying these options with engine='pyarrow' will raise a ValueError.

You can pass in a URL to read or write remote files to many of pandas’ IO functions - the following example shows reading a CSV file:

df = pd.read_csv("https://download.bls.gov/pub/time.series/cu/cu.item", sep="\t")

Added in version 1.3.0.

A custom header can be sent alongside HTTP(s) requests by passing a dictionary of header key value mappings to the storage_options keyword argument as shown below:

headers = {"User-Agent": "pandas"}
df = pd.read_csv(
    "https://download.bls.gov/pub/time.series/cu/cu.item",
    sep="\t",
    storage_options=headers
)

All URLs which are not local files or HTTP(s) are handled by fsspec, if installed, and its various filesystem implementations (including Amazon S3, Google Cloud, SSH, FTP, webHDFS…). Some of these implementations will require additional packages to be installed, for example S3 URLs require the s3fs library:

df = pd.read_json("s3://pandas-test/adatafile.json")

When dealing with remote storage systems, you might need extra configuration with environment variables or config files in special locations. For example, to access data in your S3 bucket, you will need to define credentials in one of the several ways listed in the S3Fs documentation. The same is true for several of the storage backends, and you should follow the links at fsimpl1 for implementations built into fsspec and fsimpl2 for those not included in the main fsspec distribution.

You can also pass parameters directly to the backend driver. Since fsspec does not utilize the AWS_S3_HOST environment variable, we can directly define a dictionary containing the endpoint_url and pass the object into the storage option parameter:

storage_options = {"client_kwargs": {"endpoint_url": "http://127.0.0.1:5555"}}}
df = pd.read_json("s3://pandas-test/test-1", storage_options=storage_options)

More sample configurations and documentation can be found at S3Fs documentation.

If you do not have S3 credentials, you can still access public data by specifying an anonymous connection, such as

Added in version 1.2.0.

pd.read_csv(
    "s3://ncei-wcsd-archive/data/processed/SH1305/18kHz/SaKe2013"
    "-D20130523-T080854_to_SaKe2013-D20130523-T085643.csv",
    storage_options={"anon": True},
)

fsspec also allows complex URLs, for accessing data in compressed archives, local caching of files, and more. To locally cache the above example, you would modify the call to

pd.read_csv(
    "simplecache::s3://ncei-wcsd-archive/data/processed/SH1305/18kHz/"
    "SaKe2013-D20130523-T080854_to_SaKe2013-D20130523-T085643.csv",
    storage_options={"s3": {"anon": True}},
)

where we specify that the “anon” parameter is meant for the “s3” part of the implementation, not to the caching implementation. Note that this caches to a temporary directory for the duration of the session only, but you can also specify a permanent store.

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_or_buf: A string path to the file to write or a file object. If a file object it must be opened with newline=''

• sep : Field delimiter for the output file (default “,”)

• na_rep: A string representation of a missing value (default ‘’)

• float_format: Format string for floating point numbers

• columns: 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’

• encoding: a string representing the encoding to use if the contents are non-ASCII, for Python versions prior to 3

• lineterminator: Character sequence denoting line end (default os.linesep)

• quoting: Set quoting rules as in csv module (default csv.QUOTE_MINIMAL). Note that if you have set a float_format then floats are converted to strings and csv.QUOTE_NONNUMERIC will treat them as non-numeric

• quotechar: Character used to quote fields (default ‘”’)

• doublequote: Control quoting of quotechar in fields (default True)

• escapechar: Character used to escape sep and quotechar when appropriate (default None)

• chunksize: Number of rows to write at a time

• date_format: Format string for datetime objects

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.

Read and write JSON format files and strings.

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, table}

  The format of the JSON string

  split

  dict like {index -> [index]; columns -> [columns]; data -> [values]}

  records

  list like [{column -> value}; … ]

  index

  dict like {index -> {column -> value}}

  columns

  dict like {column -> {index -> value}}

  values

  just the values array

  table

  adhering to the JSON Table Schema

• 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 serializable object.

• lines : If records orient, then will write each record per line as json.

• mode : string, writer mode when writing to path. ‘w’ for write, ‘a’ for append. Default ‘w’

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 [230]: dfj = pd.DataFrame(np.random.randn(5, 2), columns=list("AB"))

In [231]: json = dfj.to_json()

In [232]: json
Out[232]: '{"A":{"0":-0.1213062281,"1":0.6957746499,"2":0.9597255933,"3":-0.6199759194,"4":-0.7323393705},"B":{"0":-0.0978826728,"1":0.3417343559,"2":-1.1103361029,"3":0.1497483186,"4":0.6877383895}}'

There are a number of different options for the format of the resulting JSON file / string. Consider the following DataFrame and Series:

In [233]: dfjo = pd.DataFrame(
   .....:     dict(A=range(1, 4), B=range(4, 7), C=range(7, 10)),
   .....:     columns=list("ABC"),
   .....:     index=list("xyz"),
   .....: )
   .....: 

In [234]: dfjo
Out[234]: 
   A  B  C
x  1  4  7
y  2  5  8
z  3  6  9

In [235]: sjo = pd.Series(dict(x=15, y=16, z=17), name="D")

In [236]: sjo
Out[236]: 
x    15
y    16
z    17
Name: D, dtype: int64

Column oriented (the default for DataFrame) serializes the data as nested JSON objects with column labels acting as the primary index:

In [237]: dfjo.to_json(orient="columns")
Out[237]: '{"A":{"x":1,"y":2,"z":3},"B":{"x":4,"y":5,"z":6},"C":{"x":7,"y":8,"z":9}}'

# Not available for Series

Index oriented (the default for Series) similar to column oriented but the index labels are now primary:

In [238]: dfjo.to_json(orient="index")
Out[238]: '{"x":{"A":1,"B":4,"C":7},"y":{"A":2,"B":5,"C":8},"z":{"A":3,"B":6,"C":9}}'

In [239]: sjo.to_json(orient="index")
Out[239]: '{"x":15,"y":16,"z":17}'

Record oriented serializes 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 [240]: dfjo.to_json(orient="records")
Out[240]: '[{"A":1,"B":4,"C":7},{"A":2,"B":5,"C":8},{"A":3,"B":6,"C":9}]'

In [241]: sjo.to_json(orient="records")
Out[241]: '[15,16,17]'

Value oriented is a bare-bones option which serializes to nested JSON arrays of values only, column and index labels are not included:

In [242]: dfjo.to_json(orient="values")
Out[242]: '[[1,4,7],[2,5,8],[3,6,9]]'

# Not available for Series

Split oriented serializes to a JSON object containing separate entries for values, index and columns. Name is also included for Series:

In [243]: dfjo.to_json(orient="split")
Out[243]: '{"columns":["A","B","C"],"index":["x","y","z"],"data":[[1,4,7],[2,5,8],[3,6,9]]}'

In [244]: sjo.to_json(orient="split")
Out[244]: '{"name":"D","index":["x","y","z"],"data":[15,16,17]}'

Table oriented serializes to the JSON Table Schema, allowing for the preservation of metadata including but not limited to dtypes and index names.

Note

Any orient option that encodes to a JSON object will not preserve the ordering of index and column labels during round-trip serialization. If you wish to preserve label ordering use the split option as it uses ordered containers.

Writing in ISO date format:

In [245]: dfd = pd.DataFrame(np.random.randn(5, 2), columns=list("AB"))

In [246]: dfd["date"] = pd.Timestamp("20130101")

In [247]: dfd = dfd.sort_index(axis=1, ascending=False)

In [248]: json = dfd.to_json(date_format="iso")

In [249]: json
Out[249]: '{"date":{"0":"2013-01-01T00:00:00.000","1":"2013-01-01T00:00:00.000","2":"2013-01-01T00:00:00.000","3":"2013-01-01T00:00:00.000","4":"2013-01-01T00:00:00.000"},"B":{"0":0.403309524,"1":0.3016244523,"2":-1.3698493577,"3":1.4626960492,"4":-0.8265909164},"A":{"0":0.1764443426,"1":-0.1549507744,"2":-2.1798606054,"3":-0.9542078401,"4":-1.7431609117}}'

Writing in ISO date format, with microseconds:

In [250]: json = dfd.to_json(date_format="iso", date_unit="us")

In [251]: json
Out[251]: '{"date":{"0":"2013-01-01T00:00:00.000000","1":"2013-01-01T00:00:00.000000","2":"2013-01-01T00:00:00.000000","3":"2013-01-01T00:00:00.000000","4":"2013-01-01T00:00:00.000000"},"B":{"0":0.403309524,"1":0.3016244523,"2":-1.3698493577,"3":1.4626960492,"4":-0.8265909164},"A":{"0":0.1764443426,"1":-0.1549507744,"2":-2.1798606054,"3":-0.9542078401,"4":-1.7431609117}}'

Epoch timestamps, in seconds:

In [252]: json = dfd.to_json(date_format="epoch", date_unit="s")

In [253]: json
Out[253]: '{"date":{"0":1,"1":1,"2":1,"3":1,"4":1},"B":{"0":0.403309524,"1":0.3016244523,"2":-1.3698493577,"3":1.4626960492,"4":-0.8265909164},"A":{"0":0.1764443426,"1":-0.1549507744,"2":-2.1798606054,"3":-0.9542078401,"4":-1.7431609117}}'

Writing to a file, with a date index and a date column:

In [254]: dfj2 = dfj.copy()

In [255]: dfj2["date"] = pd.Timestamp("20130101")

In [256]: dfj2["ints"] = list(range(5))

In [257]: dfj2["bools"] = True

In [258]: dfj2.index = pd.date_range("20130101", periods=5)

In [259]: dfj2.to_json("test.json")

In [260]: with open("test.json") as fh:
   .....:     print(fh.read())
   .....: 
{"A":{"1356998400000":-0.1213062281,"1357084800000":0.6957746499,"1357171200000":0.9597255933,"1357257600000":-0.6199759194,"1357344000000":-0.7323393705},"B":{"1356998400000":-0.0978826728,"1357084800000":0.3417343559,"1357171200000":-1.1103361029,"1357257600000":0.1497483186,"1357344000000":0.6877383895},"date":{"1356998400000":1356,"1357084800000":1356,"1357171200000":1356,"1357257600000":1356,"1357344000000":1356},"ints":{"1356998400000":0,"1357084800000":1,"1357171200000":2,"1357257600000":3,"1357344000000":4},"bools":{"1356998400000":true,"1357084800000":true,"1357171200000":true,"1357257600000":true,"1357344000000":true}}

If the JSON serializer cannot handle the container contents directly it will fall back in the following manner:

• if the dtype is unsupported (e.g. np.complex_) then the default_handler, if provided, will be called for each value, otherwise an exception is raised.

• if an object is unsupported it will attempt the following:

  • check if the object has defined a toDict method and call it. A toDict method should return a dict which will then be JSON serialized.

  • invoke the default_handler if one was provided.

  • convert the object to a dict by traversing its contents. However this will often fail with an OverflowError or give unexpected results.

In general the best approach for unsupported objects or dtypes is to provide a default_handler. For example:

>>> DataFrame([1.0, 2.0, complex(1.0, 2.0)]).to_json()  # raises
RuntimeError: Unhandled numpy dtype 15

can be dealt with by specifying a simple default_handler:

In [261]: pd.DataFrame([1.0, 2.0, complex(1.0, 2.0)]).to_json(default_handler=str)
Out[261]: '{"0":{"0":"(1+0j)","1":"(2+0j)","2":"(1+2j)"}}'

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 explicitly 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, table}

  The format of the JSON string

  split

  dict like {index -> [index]; columns -> [columns]; data -> [values]}

  records

  list like [{column -> value} …]

  index

  dict like {index -> {column -> value}}

  columns

  dict like {column -> {index -> value}}

  values

  just the values array

  table

  adhering to the JSON Table Schema

• 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 date-like columns, default is True.

• keep_default_dates : boolean, default True. If parsing dates, then parse the default date-like columns.

• 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.

• lines : reads file as one json object per line.

• encoding : The encoding to use to decode py3 bytes.

• chunksize : when used in combination with lines=True, return a pandas.api.typing.JsonReader which reads in chunksize lines per iteration.

• engine: Either "ujson", the built-in JSON parser, or "pyarrow" which dispatches to pyarrow’s pyarrow.json.read_json. The "pyarrow" is only available when lines=True

The parser will raise one of ValueError/TypeError/AssertionError if the JSON is not parseable.

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.

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. ‘date-like’ means that the column label meets one of the following criteria:

• it ends with '_at'

• it ends with '_time'

• it begins with 'timestamp'

• it is 'modified'

• it is 'date'

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 [262]: from io import StringIO

In [263]: pd.read_json(StringIO(json))
Out[263]: 
   date         B         A
0     1  0.403310  0.176444
1     1  0.301624 -0.154951
2     1 -1.369849 -2.179861
3     1  1.462696 -0.954208
4     1 -0.826591 -1.743161

Reading from a file:

In [264]: pd.read_json("test.json")
Out[264]: 
                   A         B  date  ints  bools
2013-01-01 -0.121306 -0.097883  1356     0   True
2013-01-02  0.695775  0.341734  1356     1   True
2013-01-03  0.959726 -1.110336  1356     2   True
2013-01-04 -0.619976  0.149748  1356     3   True
2013-01-05 -0.732339  0.687738  1356     4   True

Don’t convert any data (but still convert axes and dates):

In [265]: pd.read_json("test.json", dtype=object).dtypes
Out[265]: 
A        object
B        object
date     object
ints     object
bools    object
dtype: object

Specify dtypes for conversion:

In [266]: pd.read_json("test.json", dtype={"A": "float32", "bools": "int8"}).dtypes
Out[266]: 
A        float32
B        float64
date       int64
ints       int64
bools       int8
dtype: object

Preserve string indices:

In [267]: from io import StringIO

In [268]: si = pd.DataFrame(
   .....:     np.zeros((4, 4)), columns=list(range(4)), index=[str(i) for i in range(4)]
   .....: )
   .....: 

In [269]: si
Out[269]: 
     0    1    2    3
0  0.0  0.0  0.0  0.0
1  0.0  0.0  0.0  0.0
2  0.0  0.0  0.0  0.0
3  0.0  0.0  0.0  0.0

In [270]: si.index
Out[270]: Index(['0', '1', '2', '3'], dtype='object')

In [271]: si.columns
Out[271]: Index([0, 1, 2, 3], dtype='int64')

In [272]: json = si.to_json()

In [273]: sij = pd.read_json(StringIO(json), convert_axes=False)

In [274]: sij
Out[274]: 
   0  1  2  3
0  0  0  0  0
1  0  0  0  0
2  0  0  0  0
3  0  0  0  0

In [275]: sij.index
Out[275]: Index(['0', '1', '2', '3'], dtype='object')

In [276]: sij.columns
Out[276]: Index(['0', '1', '2', '3'], dtype='object')

Dates written in nanoseconds need to be read back in nanoseconds:

In [277]: from io import StringIO

In [278]: json = dfj2.to_json(date_unit="ns")

# Try to parse timestamps as milliseconds -> Won't Work
In [279]: dfju = pd.read_json(StringIO(json), date_unit="ms")

In [280]: dfju
Out[280]: 
                            A         B        date  ints  bools
1356998400000000000 -0.121306 -0.097883  1356998400     0   True
1357084800000000000  0.695775  0.341734  1356998400     1   True
1357171200000000000  0.959726 -1.110336  1356998400     2   True
1357257600000000000 -0.619976  0.149748  1356998400     3   True
1357344000000000000 -0.732339  0.687738  1356998400     4   True

# Let pandas detect the correct precision
In [281]: dfju = pd.read_json(StringIO(json))

In [282]: dfju
Out[282]: 
                   A         B       date  ints  bools
2013-01-01 -0.121306 -0.097883 2013-01-01     0   True
2013-01-02  0.695775  0.341734 2013-01-01     1   True
2013-01-03  0.959726 -1.110336 2013-01-01     2   True
2013-01-04 -0.619976  0.149748 2013-01-01     3   True
2013-01-05 -0.732339  0.687738 2013-01-01     4   True

# Or specify that all timestamps are in nanoseconds
In [283]: dfju = pd.read_json(StringIO(json), date_unit="ns")

In [284]: dfju
Out[284]: 
                   A         B        date  ints  bools
2013-01-01 -0.121306 -0.097883  1356998400     0   True
2013-01-02  0.695775  0.341734  1356998400     1   True
2013-01-03  0.959726 -1.110336  1356998400     2   True
2013-01-04 -0.619976  0.149748  1356998400     3   True
2013-01-05 -0.732339  0.687738  1356998400     4   True

By setting the dtype_backend argument you can control the default dtypes used for the resulting DataFrame.

In [285]: data = (
   .....:  '{"a":{"0":1,"1":3},"b":{"0":2.5,"1":4.5},"c":{"0":true,"1":false},"d":{"0":"a","1":"b"},'
   .....:  '"e":{"0":null,"1":6.0},"f":{"0":null,"1":7.5},"g":{"0":null,"1":true},"h":{"0":null,"1":"a"},'
   .....:  '"i":{"0":"12-31-2019","1":"12-31-2019"},"j":{"0":null,"1":null}}'
   .....: )
   .....: 

In [286]: df = pd.read_json(StringIO(data), dtype_backend="pyarrow")

In [287]: df
Out[287]: 
   a    b      c  d     e     f     g     h           i     j
0  1  2.5   True  a  <NA>  <NA>  <NA>  <NA>  12-31-2019  None
1  3  4.5  False  b     6   7.5  True     a  12-31-2019  None

In [288]: df.dtypes
Out[288]: 
a     int64[pyarrow]
b    double[pyarrow]
c      bool[pyarrow]
d    string[pyarrow]
e     int64[pyarrow]
f    double[pyarrow]
g      bool[pyarrow]
h    string[pyarrow]
i    string[pyarrow]
j      null[pyarrow]
dtype: object

pandas provides a utility function to take a dict or list of dicts and normalize this semi-structured data into a flat table.

In [289]: data = [
   .....:     {"id": 1, "name": {"first": "Coleen", "last": "Volk"}},
   .....:     {"name": {"given": "Mark", "family": "Regner"}},
   .....:     {"id": 2, "name": "Faye Raker"},
   .....: ]
   .....: 

In [290]: pd.json_normalize(data)
Out[290]: 
    id name.first name.last name.given name.family        name
0  1.0     Coleen      Volk        NaN         NaN         NaN
1  NaN        NaN       NaN       Mark      Regner         NaN
2  2.0        NaN       NaN        NaN         NaN  Faye Raker

In [291]: data = [
   .....:     {
   .....:         "state": "Florida",
   .....:         "shortname": "FL",
   .....:         "info": {"governor": "Rick Scott"},
   .....:         "county": [
   .....:             {"name": "Dade", "population": 12345},
   .....:             {"name": "Broward", "population": 40000},
   .....:             {"name": "Palm Beach", "population": 60000},
   .....:         ],
   .....:     },
   .....:     {
   .....:         "state": "Ohio",
   .....:         "shortname": "OH",
   .....:         "info": {"governor": "John Kasich"},
   .....:         "county": [
   .....:             {"name": "Summit", "population": 1234},
   .....:             {"name": "Cuyahoga", "population": 1337},
   .....:         ],
   .....:     },
   .....: ]
   .....: 

In [292]: pd.json_normalize(data, "county", ["state", "shortname", ["info", "governor"]])
Out[292]: 
         name  population    state shortname info.governor
0        Dade       12345  Florida        FL    Rick Scott
1     Broward       40000  Florida        FL    Rick Scott
2  Palm Beach       60000  Florida        FL    Rick Scott
3      Summit        1234     Ohio        OH   John Kasich
4    Cuyahoga        1337     Ohio        OH   John Kasich

The max_level parameter provides more control over which level to end normalization. With max_level=1 the following snippet normalizes until 1st nesting level of the provided dict.

In [293]: data = [
   .....:     {
   .....:         "CreatedBy": {"Name": "User001"},
   .....:         "Lookup": {
   .....:             "TextField": "Some text",
   .....:             "UserField": {"Id": "ID001", "Name": "Name001"},
   .....:         },
   .....:         "Image": {"a": "b"},
   .....:     }
   .....: ]
   .....: 

In [294]: pd.json_normalize(data, max_level=1)
Out[294]: 
  CreatedBy.Name Lookup.TextField                    Lookup.UserField Image.a
0        User001        Some text  {'Id': 'ID001', 'Name': 'Name001'}       b

pandas is able to read and write line-delimited json files that are common in data processing pipelines using Hadoop or Spark.

For line-delimited json files, pandas can also return an iterator which reads in chunksize lines at a time. This can be useful for large files or to read from a stream.

In [295]: from io import StringIO

In [296]: jsonl = """
   .....:     {"a": 1, "b": 2}
   .....:     {"a": 3, "b": 4}
   .....: """
   .....: 

In [297]: df = pd.read_json(StringIO(jsonl), lines=True)

In [298]: df
Out[298]: 
   a  b
0  1  2
1  3  4

In [299]: df.to_json(orient="records", lines=True)
Out[299]: '{"a":1,"b":2}\n{"a":3,"b":4}\n'

# reader is an iterator that returns ``chunksize`` lines each iteration
In [300]: with pd.read_json(StringIO(jsonl), lines=True, chunksize=1) as reader:
   .....:     reader
   .....:     for chunk in reader:
   .....:         print(chunk)
   .....: 
Empty DataFrame
Columns: []
Index: []
   a  b
0  1  2
   a  b
1  3  4

Line-limited json can also be read using the pyarrow reader by specifying engine="pyarrow".

In [301]: from io import BytesIO

In [302]: df = pd.read_json(BytesIO(jsonl.encode()), lines=True, engine="pyarrow")

In [303]: df
Out[303]: 
   a  b
0  1  2
1  3  4

Added in version 2.0.0.

Table Schema is a spec for describing tabular datasets as a JSON object. The JSON includes information on the field names, types, and other attributes. You can use the orient table to build a JSON string with two fields, schema and data.

In [304]: df = pd.DataFrame(
   .....:     {
   .....:         "A": [1, 2, 3],
   .....:         "B": ["a", "b", "c"],
   .....:         "C": pd.date_range("2016-01-01", freq="d", periods=3),
   .....:     },
   .....:     index=pd.Index(range(3), name="idx"),
   .....: )
   .....: 

In [305]: df
Out[305]: 
     A  B          C
idx                 
0    1  a 2016-01-01
1    2  b 2016-01-02
2    3  c 2016-01-03

In [306]: df.to_json(orient="table", date_format="iso")
Out[306]: '{"schema":{"fields":[{"name":"idx","type":"integer"},{"name":"A","type":"integer"},{"name":"B","type":"string"},{"name":"C","type":"datetime"}],"primaryKey":["idx"],"pandas_version":"1.4.0"},"data":[{"idx":0,"A":1,"B":"a","C":"2016-01-01T00:00:00.000"},{"idx":1,"A":2,"B":"b","C":"2016-01-02T00:00:00.000"},{"idx":2,"A":3,"B":"c","C":"2016-01-03T00:00:00.000"}]}'

The schema field contains the fields key, which itself contains a list of column name to type pairs, including the Index or MultiIndex (see below for a list of types). The schema field also contains a primaryKey field if the (Multi)index is unique.

The second field, data, contains the serialized data with the records orient. The index is included, and any datetimes are ISO 8601 formatted, as required by the Table Schema spec.

The full list of types supported are described in the Table Schema spec. This table shows the mapping from pandas types:

pandas type

Table Schema type

int64

integer

float64

number

bool

boolean

datetime64[ns]

datetime

timedelta64[ns]

duration

categorical

any

object

str

A few notes on the generated table schema:

• The schema object contains a pandas_version field. This contains the version of pandas’ dialect of the schema, and will be incremented with each revision.

• All dates are converted to UTC when serializing. Even timezone naive values, which are treated as UTC with an offset of 0.

  In [307]: from pandas.io.json import build_table_schema
  
  In [308]: s = pd.Series(pd.date_range("2016", periods=4))
  
  In [309]: build_table_schema(s)
  Out[309]: 
  {'fields': [{'name': 'index', 'type': 'integer'},
    {'name': 'values', 'type': 'datetime'}],
   'primaryKey': ['index'],
   'pandas_version': '1.4.0'}
  

• datetimes with a timezone (before serializing), include an additional field tz with the time zone name (e.g. 'US/Central').

  In [310]: s_tz = pd.Series(pd.date_range("2016", periods=12, tz="US/Central"))
  
  In [311]: build_table_schema(s_tz)
  Out[311]: 
  {'fields': [{'name': 'index', 'type': 'integer'},
    {'name': 'values', 'type': 'datetime', 'tz': 'US/Central'}],
   'primaryKey': ['index'],
   'pandas_version': '1.4.0'}
  

• Periods are converted to timestamps before serialization, and so have the same behavior of being converted to UTC. In addition, periods will contain and additional field freq with the period’s frequency, e.g. 'A-DEC'.

  In [312]: s_per = pd.Series(1, index=pd.period_range("2016", freq="Y-DEC", periods=4))
  
  In [313]: build_table_schema(s_per)
  Out[313]: 
  {'fields': [{'name': 'index', 'type': 'datetime', 'freq': 'YE-DEC'},
    {'name': 'values', 'type': 'integer'}],
   'primaryKey': ['index'],
   'pandas_version': '1.4.0'}
  

• Categoricals use the any type and an enum constraint listing the set of possible values. Additionally, an ordered field is included:

  In [314]: s_cat = pd.Series(pd.Categorical(["a", "b", "a"]))
  
  In [315]: build_table_schema(s_cat)
  Out[315]: 
  {'fields': [{'name': 'index', 'type': 'integer'},
    {'name': 'values',
     'type': 'any',
     'constraints': {'enum': ['a', 'b']},
     'ordered': False}],
   'primaryKey': ['index'],
   'pandas_version': '1.4.0'}
  

• A primaryKey field, containing an array of labels, is included if the index is unique:

  In [316]: s_dupe = pd.Series([1, 2], index=[1, 1])
  
  In [317]: build_table_schema(s_dupe)
  Out[317]: 
  {'fields': [{'name': 'index', 'type': 'integer'},
    {'name': 'values', 'type': 'integer'}],
   'pandas_version': '1.4.0'}
  

• The primaryKey behavior is the same with MultiIndexes, but in this case the primaryKey is an array:

  In [318]: s_multi = pd.Series(1, index=pd.MultiIndex.from_product([("a", "b"), (0, 1)]))
  
  In [319]: build_table_schema(s_multi)
  Out[319]: 
  {'fields': [{'name': 'level_0', 'type': 'string'},
    {'name': 'level_1', 'type': 'integer'},
    {'name': 'values', 'type': 'integer'}],
   'primaryKey': FrozenList(['level_0', 'level_1']),
   'pandas_version': '1.4.0'}
  

• The default naming roughly follows these rules:

  • For series, the object.name is used. If that’s none, then the name is values

  • For DataFrames, the stringified version of the column name is used

  • For Index (not MultiIndex), index.name is used, with a fallback to index if that is None.

  • For MultiIndex, mi.names is used. If any level has no name, then level_<i> is used.

read_json also accepts orient='table' as an argument. This allows for the preservation of metadata such as dtypes and index names in a round-trippable manner.

In [320]: df = pd.DataFrame(
   .....:     {
   .....:         "foo": [1, 2, 3, 4],
   .....:         "bar": ["a", "b", "c", "d"],
   .....:         "baz": pd.date_range("2018-01-01", freq="d", periods=4),
   .....:         "qux": pd.Categorical(["a", "b", "c", "c"]),
   .....:     },
   .....:     index=pd.Index(range(4), name="idx"),
   .....: )
   .....: 

In [321]: df
Out[321]: 
     foo bar        baz qux
idx                        
0      1   a 2018-01-01   a
1      2   b 2018-01-02   b
2      3   c 2018-01-03   c
3      4   d 2018-01-04   c

In [322]: df.dtypes
Out[322]: 
foo             int64
bar            object
baz    datetime64[ns]
qux          category
dtype: object

In [323]: df.to_json("test.json", orient="table")

In [324]: new_df = pd.read_json("test.json", orient="table")

In [325]: new_df
Out[325]: 
     foo bar        baz qux
idx                        
0      1   a 2018-01-01   a
1      2   b 2018-01-02   b
2      3   c 2018-01-03   c
3      4   d 2018-01-04   c

In [326]: new_df.dtypes
Out[326]: 
foo             int64
bar            object
baz    datetime64[ns]
qux          category
dtype: object