output

Timing to_datetime():
Datetime format: %m-%d-%Y
---------------
          Without infer_format: cad7e6333f1cd2ebe272d7fc7553cd27 - 3.082s
             With infer_format: cad7e6333f1cd2ebe272d7fc7553cd27 - 0.268s (11.51x baseline)
            Passing the format: cad7e6333f1cd2ebe272d7fc7553cd27 - 0.267s (11.54x baseline)

Datetime format: %m/%d/%Y %H:%M:%S.%f
---------------
          Without infer_format: ec2ef27adb2e95e2c5386d8d5d8513a1 - 5.162s
             With infer_format: ec2ef27adb2e95e2c5386d8d5d8513a1 - 0.501s (10.31x baseline)
            Passing the format: ec2ef27adb2e95e2c5386d8d5d8513a1 - 0.509s (10.14x baseline)

Datetime format: %Y-%m-%dT%H:%M:%S.%f
---------------
          Without infer_format: ec2ef27adb2e95e2c5386d8d5d8513a1 - 0.013s
             With infer_format: ec2ef27adb2e95e2c5386d8d5d8513a1 - 0.012s (1.06x baseline)
            Passing the format: ec2ef27adb2e95e2c5386d8d5d8513a1 - 0.491s (0.03x baseline)


Testing reading CSV:
Datetime format: %m-%d-%Y
---------------
          Without infer_format: e8bde3ae42c769a7509cb94e9f36ca5e - 3.085s
             With infer_format: e8bde3ae42c769a7509cb94e9f36ca5e - 0.298s (10.37x baseline)
     With strptime date_parser: e8bde3ae42c769a7509cb94e9f36ca5e - 0.906s (3.41x baseline)

Datetime format: %m/%d/%Y %H:%M:%S.%f
---------------
          Without infer_format: 2f7019b7f7795146b4de4e83fbc0ebb5 - 5.230s
             With infer_format: 2f7019b7f7795146b4de4e83fbc0ebb5 - 0.555s (9.42x baseline)
     With strptime date_parser: 2f7019b7f7795146b4de4e83fbc0ebb5 - 1.245s (4.20x baseline)

Datetime format: %Y-%m-%dT%H:%M:%S.%f
---------------
          Without infer_format: f0028f10cf5d2fc66900e1a57d5bbf9a - 0.073s
             With infer_format: f0028f10cf5d2fc66900e1a57d5bbf9a - 0.062s (1.19x baseline)
     With strptime date_parser: f0028f10cf5d2fc66900e1a57d5bbf9a - 1.249s (0.06x baseline)

test_infer_speed.py

import datetime
import hashlib
import time

import pandas as pd

def time_with_hash(desc, f, baseline=None):
    start = time.time()
    output = f()
    timing = time.time() - start

    if baseline is None:
        print '%30s: %s - %.3fs' % (
            desc,
            hashlib.md5(output.to_json()).hexdigest(),
            timing
        )
    else:
        print '%30s: %s - %.3fs (%.2fx baseline)' % (
            desc,
            hashlib.md5(output.to_json()).hexdigest(),
            timing,
            baseline / timing
        )

    return timing

test_formats = [
    '%m-%d-%Y',
    '%m/%d/%Y %H:%M:%S.%f',
    '%Y-%m-%dT%H:%M:%S.%f',
]

print 'Timing to_datetime():'

for test_format in test_formats:
    s = (
            pd
            .Series(pd.date_range('20000101', periods=50000, freq='H'))
            .apply(lambda x: x.strftime(test_format))
    )

    print 'Datetime format:', test_format
    print '---------------'
    baseline = time_with_hash('Without infer_format', lambda: pd.to_datetime(s))
    time_with_hash(
        'With infer_format',
        lambda: pd.to_datetime(s, infer_format=True),
        baseline=baseline
    )
    time_with_hash(
        'Passing the format',
        lambda: pd.to_datetime(s, format=test_format),
        baseline=baseline
    )
    print

print
print 'Testing reading CSV:'

for test_format in test_formats:
    s = (
            pd
            .Series(pd.date_range('20000101', periods=50000, freq='H'))
            .apply(lambda x: x.strftime(test_format))
    )

    s.to_csv('/tmp/test.csv')

    def date_parser(s):
        return datetime.datetime.strptime(s, test_format)

    print 'Datetime format:', test_format
    print '---------------'
    baseline = time_with_hash('Without infer_format',
        lambda: pd.read_csv(
            '/tmp/test.csv',
            parse_dates=[1,],
        )
    )
    time_with_hash('With infer_format',
        lambda: pd.read_csv(
            '/tmp/test.csv',
            parse_dates=[1,],
            infer_datetime_format=True
        ),
        baseline=baseline
    )
    time_with_hash('With strptime date_parser',
        lambda: pd.read_csv(
            '/tmp/test.csv',
            parse_dates=[1,],
            date_parser=date_parser,
        ),
        baseline=baseline
    )
    print