API: read_csv inconsistent with from_csv -- parses ints as dates

[darindillon]

Using pandas 0.10.1.
I read the docs, but didn't see any explanation of why this would be true. pandas.read_csv() works exactly as you'd expect, but pandas.DataFrame.from_csv() is different. Looks like the latter method assumes you're probably dealing with time series data, so it sets defaults parameters to automatically convert integers to dates. I disagree that this is desired, but even if it is, why would it be true for the later method but not the former? Why shouldn't both methods assume the same default assumptions?

Create a CSV like this:
a,b
1,4
2,3

Now this does exactly what you'd expect:
p = pandas.read_csv(your_csv_file)

But this converts the first column into a data. Almost certainly not what you'd expect:
p = pandas.DataFrame.from_csv(your_csv_file)

There is an optional parameter on the second method "parse_dates" which is default False. If you add that flag, then the second method works just like the first. But why the inconsistency? I'd expect this method to default to acting just like the other one.

[jtratner]

resolution on this? should we just change the default kwarg to parse_dates=False? (granted, it's weird that even with parse_dates=True it interprets that column as dates...

[jreback]

I think we discusses this before

but would just make DataFrame.from_csv just be a direct pass thru call to read_csv

I think the different defaults are confusing

only issue is there is no easy way to warn the user (aside from release notes) that the API has changed (to be read_csv)

not sure what is actually different though

(also if u fix this I think there is an issue about deprecating DataFrame.from_csv. which could be closed)

[jtratner]

I have zero idea if it matters to change the from_csv API. The differences
seem strange enough that I'm not sure we have to be concerned that many
people are using it now.

[jreback]

closing in favor of #4916