read_csv interprets index column as dates

[takluyver]

I loaded a dataset with an index column going from 1 to 304. Then I try to get the repr of the dataframe, but it raises ValueError: year=100 is before 1900; the datetime strftime() methods require year >= 1900. Looking at df.index shows that the numeric indices have been transformed into datetimes (the first 31 to days of this month, 32-60 as years 20xx, 61-99 as years 19xx, and higher numbers as a raw year).

[wesm]

Hm, this is caused by a bit of legacy cruft. Perhaps it's time to add a parse_dates=True option to read_csv/read_table and have it default to False. Currently it attempts to parse the index column expecting time series data but it's apparently a bit too aggressive.

[wesm]

Alright, I'm just going to do this. I also am going to change the default for index_col to None-- I don't like breaking APIs but in retrospect index_col=0 was the wrong default. In certain cases (there the number of names in the header is 1 less than the number of columns of data) it should be able to auto-infer that the index_col=0, however (there is a test for this, I think)

[takluyver]

Oh, you're already addressing the next issue I filed. Kudos for responding so quickly!

[wesm]

Again, tragic to break APIs but changing to parse_dates=False and index_col=None by default makes the most sense. I'll make sure to sound the alarm at next release time so that users review their usage of read_csv / read_table. Could use more unit testing on this-- also need to update the sphinx docs to reflect the new defaults and to document usage with MultiIndex (e.g. index_col=[0, 1])-- if you get some energy feel free to do so. I'll create a separate ticket about updating the docs

[takluyver]

Sure, I'll see what I can do.

Presumably this makes the next release 0.5, rather than 0.4.4, since APIs are changing.

[wesm]

That seems like a good idea. If any urgent bug fixes come along I can create a 0.4.x maintenance branch for cherry-picking, otherwise, I suppose I can target end of month or early November for a 0.5 release. Lot of stuff I want to do before then.