Improved memory management for DataFrame and Series data
Initial support for pandas.read_sql()
pandas.read_csv() reads a directory of csv files
pandas.read_csv() reads from S3, and Hadoop Distributed File
System (HDFS)
pandas.read_parquet() now reads all integer types (like int16) and
gets nullable information for integer columns from pandas metadata
pandas.read_parquet() now supports reading columns of list of
string elements
avoid type error for unselected columns in Parquet files
support pandas.RangeIndex when reading a non-partitioned parquet
dataset
pandas.Dataframe.to_parquet() to Hadoop Distributed File System
(HDFS)
pandas.Dataframe.to_parquet() always writes pandas.RangeIndex to
Parquet metadata
support pandas.Dataframe.to_parquet() writing datetime64 (default
in Pandas) and datatime.date types to Parquet files
support decimal.Decimal type in dataframes and Parquet I/O
Support for &, |, and pandas.Series.dt in
pandas.Dataframe.query()
Support added for groupby last operation
min, max, and sum support in groupby() for string columns
non-constant list of column names as argument support for functions
like groupby()
MultiIndex support for groupby(...).agg(as_index=False)
pandas.Dataframe.merge() one dataframe on index, and the other on
a column
sorting compilation time improvement
supports for integer, float, string, string list, datetime.date,
datetime.datetime, and datetime.timedelta types in
pandas.Series.cummin(), pandas.DataFrame.cummin(),
pandas.Series.cummax(), and pandas.DataFrame.cummax()
NAs in datetime.date array
better datetime.timedelta support
Support for min and max in pandas.Timestamp and
datetime.date
pandas.DataFrame.all() for boolean series
pandas.Series.astype() to float, int, str
Convert string columns to float using astype()
NA support for Series.str.split()
refactored and improved Dataframe indexing: pandas.loc(),
pandas.Dataframe.iloc(), and pandas.Dataframe.iat()
better support for pandas.Series.shift(),
pandas.Series.pct_change(), pandas.Dataframe.drop()
groupby(...).agg() when passing a dictionary of functions: support
mix of multi-function lists and single functions
Fixed Numpy slicing error in a corner case when the slice is
equivalent to array and array size is a constant
proper construction of dataframe from slicing Numpy 2D array
pandas.read_csv reads a dataframe containing only datetime like
columns
When using pandas.merge() and pandas.join() integer columns
which can have a missing value NA are returned as
nullable integer array (as opposed to 0 and
-1 before)