Permalink
Commits on Apr 13, 2015
  1. BUG: Fix error when reading postgres table with timezone #7139

    danbirken authored and jorisvandenbossche committed Mar 24, 2015
    `read_sql_table()` will break if it reads a table with a `timestamp
    with time zone` column if individual rows within that column have
    different time zones. This is very common due to daylight savings time.
    
    Pandas right now does not have good support for a Series containing
    datetimes with different time zones (hence this bug).  So this change
    simply converts a `timestamp with time zone` column into UTC during
    import, which pandas has great support for.
Commits on Jan 25, 2014
  1. DOC/BUG: Fix documentation for `infer_datetime_format` #6073

    danbirken committed Jan 25, 2014
    Fix formatting typos and ensure the "foo.csv" ipython processing works.
Commits on Jan 24, 2014
  1. PERF: Add infer_datetime_format to read_csv() #5490

    danbirken committed Jan 24, 2014
    This allows read_csv() to attempt to infer the datetime format for any
    columns where parse_dates is enabled.  In cases where the datetime
    format can be inferred, this should speed up processing datetimes
    by ~10x.
    
    Additionally add documentation and benchmarks for read_csv().
  2. PERF: Speed up pd.to_datetime() by optionally inferring dt format #5490

    danbirken committed Jan 24, 2014
    Given an array of strings that represent datetimes, infer_format=True
    will attempt to guess the format of the datetimes, and if it can infer
    the format, it will use a faster function to convert/import the
    datetimes.  In cases where this speed-up can be used, the function
    should be about 10x faster.
Commits on Jan 19, 2014
  1. BUG: Fix to_datetime to properly deal with tz offsets #3944

    danbirken committed Jan 15, 2014
    Currently for certain formats of datetime strings, the tz offset will
    just be ignored.
Commits on Oct 21, 2013
  1. TST: Fix edge cases in assert_almost_equal() + tests #4398

    danbirken committed Oct 21, 2013
    Many of the edge cases were related to ordering of the items, but in
    some cases there were also issues with type checking.  This fixes both
    of those issues and massively expands the testing for this function.
  2. TST/PERF: Re-write assert_almost_equal() in cython #4398

    danbirken committed Oct 21, 2013
    Add a testing.pyx cython file, and port assert_almost_equal() from
    python to cython.
    
    On my machine this brings a modest gain to the suite of "not slow" tests
    (160s -> 140s), but on assert_almost_equal() heavy tests, like
    test_expressions.py, it shows a large improvement (14s -> 4s).
Commits on Oct 8, 2013
Commits on Oct 4, 2013
  1. BUG: Fix bound checking for Timestamp() with dt64 #4065

    danbirken committed Oct 4, 2013
    To fix the bug, this change adds bounds checking to
    _get_datetime64_nanos() for numpy datetimes that aren't already in [ns]
    units.
    
    Additionally, it updates _check_dts_bounds() to do the bound check just
    based off the pandas_datetimestruct, by comparing to the minimum and
    maximum valid pandas_datetimestructs for datetime64[ns].  It is simpler
    and more accurate than the previous system.
    
    Also includes a number of small refactors/fixes to deal with new error
    cases that didn't exist when invalid datetime64s were just silently
    coerced into the valid datetime64[ns] range.
Commits on Sep 20, 2013
  1. BUG: Constrain date parsing from strings a little bit more #4601

    danbirken committed Sep 20, 2013
    Currently dateutil will parse almost any string into a datetime.  This
    change adds a filter in front of dateutil that will prevent it from
    parsing certain strings that don't look like datetimes:
    
    1) Strings that parse to float values that are less than 1000
    2) Certain special one character strings (this was already in there,
       this just moves that code)
    
    Additionally, this filters out datetimes that are out of range for the
    datetime64[ns] type.  Currently any out-of-range datetimes will just
    overflow and be mapped to some random time within the bounds of
    datetime64[ns].
Commits on Aug 5, 2013
Commits on Jun 29, 2013
Commits on Jun 25, 2013
  1. Add release notes for #3598

    danbirken committed Jun 25, 2013
Commits on Apr 1, 2013
  1. ENH: Declare a BoolBlock as a NumericBlock

    danbirken committed Apr 1, 2013
    BUG: GH2641 fixes "df.decribe() with boolean column"
    
    This change will make all numeric operations on boolean data work, by
    just transparently treating them as integers values 1 and 0.  This is
    not pandas specific behavior, this is the default operations of both
    numpy and python.
Commits on Mar 24, 2013
Commits on Mar 15, 2012