Pandas - automatically detect date columns **at run time**

Question

I was wondering if pandas is capable of automatically detecting which columns are datetime objects and read those columns in as dates instead of strings?

I am looking at the api and related stack overflow posts but I can't seem to figure it out.

This is a black-box system that takes in arbitrary csv schema on production so I do not what the column names will be.

This seems like it would work but you have to know which columns are date fields:

import pandas as pd

#creating the test data
df = pd.DataFrame({'0': ['a', 'b', 'c'], '1': ['2015-12-27','2015-12-28', '2015-12-29'], '2': [11,12,13]})
df.to_csv('test.csv', index=False)

#loading the test data
df = pd.read_csv('test.csv', parse_dates=True)
print df.dtypes 
# prints (object, object, int64) instead of (object,datetime, int64)

I am thinking if it cannot do this, then I can write something that:

1. Finds columns with string type.
2. Grab a few unique values and try to parse them.
3. If successful then try to parse the whole column.

Edit. I wrote a simple method convertDateColumns that will do this:

import pandas as pd
from dateutil import parser

def convertDateColumns(self, df):
    object_cols = df.columns.values[df.dtypes.values == 'object']
    date_cols = [c for c in object_cols if testIfColumnIsDate(df[c], num_tries=3)]

    for col in date_cols:
        try:
            df[col] = pd.to_datetime(df[col], coerce=True, infer_datetime_format=True)
        except ValueError:
            pass

    return df

def testIfColumnIsDate(series, num_tries=4):
""" Test if a column contains date values.
    This can try a few times for the scenerio where a date column may have
    a couple of null or missing values but we still want to parse when
    possible (and convert those null/missing to NaD values)
"""
    if series.dtype != 'object':
        return False

    vals = set()
    for val in series:
        vals.add(val)
        if len(vals) > num_tries:
            break

    for val in list(vals):
        try:
            if type(val) is int:
                continue

            parser.parse(val)
            return True
        except ValueError:
            pass

    return False

Answer 1

I would use pd.to_datetime, and catch exceptions on columns that don't work. For example:

import pandas as pd

df = pd.read_csv('test.csv')

for col in df.columns:
    if df[col].dtype == 'object':
        try:
            df[col] = pd.to_datetime(df[col])
        except ValueError:
            pass

df.dtypes
# (object, datetime64[ns], int64)

I believe this is as close to "automatic" as you can get for this application.

Answer 2

You can get rid of the for loop by using the parameter errors='ignore' to avoid modifying unwanted values. In the code below we apply a to_datetime transformation (ignoring errors) on all object columns -- other columns are returned as is.

If ignore, then invalid parsing will return the input

df = df.apply(lambda col: pd.to_datetime(col, errors='ignore') 
              if col.dtypes == object 
              else col, 
              axis=0)

df.dtypes

# 0            object
# 1    datetime64[ns]
# 2             int64