API: Can't override type sniffing in df.from_csv()?

[ghost]

In [10]: ix = [-352.737091, 183.575577]
    ...: df=pd.DataFrame([0,1],index=ix)
    ...: df.to_csv("/tmp/1.csv")
    ...: df2=pd.DataFrame.from_csv("/tmp/1.csv")
    ...: print df
    ...: print df2
             0
-352.737091  0
 183.575577  1
                            0
2105-11-21 22:43:41.128654  0
1936-11-21 22:43:41.128654  1

and if you try:

In [13]: df2.index.astype('f')
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-13-8d8276604fe4> in <module>()
----> 1 df2.index.astype('f')

/home/user1/src/pandas/pandas/tseries/index.pyc in astype(self, dtype)
    624             return self.asi8.copy()
    625         else:  # pragma: no cover
--> 626             raise ValueError('Cannot cast DatetimeIndex to dtype %s' % dtype)
    627 
    628     @property

ValueError: Cannot cast DatetimeIndex to dtype float32

related SO

[stephenwlin]

Wow, what a coincidence, I just happened to get this same thing when trying to process the results I just posted at #3146 and was going to post something:

(.env0)stephen@anacreon:~$ head -5 take.csv
count,naive,memmove,contig,sse
2,29.94,62.24,34.41,22.34
3,27.41,46.79,23.07,17.95
4,23.94,40.03,17.38,13.02
5,22.30,35.95,18.39,14.99
(.env0)stephen@anacreon:~$ ipython
In [1]: import pandas as p; p.DataFrame.from_csv("take.csv").head(4)
Out[1]: 
            naive  memmove  contig    sse
count                                    
2013-03-02  29.94    62.24   34.41  22.34
2013-03-03  27.41    46.79   23.07  17.95
2013-03-04  23.94    40.03   17.38  13.02
2013-03-05  22.30    35.95   18.39  14.99

[ghost]

That means users are hitting it too. needs to get fixed.

[jreback]

this is prob being forced because parse_date=True and index_col=0 by default and I guess the index somehow looks datelike....

[ghost]

saved by jeff... again.

I guess -352.737091 does look a little like october if you squint.

[jreback]

but that still shouldn't be forcibly parsing into a datelike

[jreback]

here's the correct parsing

but as I said above, the parser shouldn't interpret those floats as dates anyhow

In [12]: df3=pd.DataFrame.from_csv('/tmp/1.csv',parse_dates=False)

In [13]: df3
Out[13]: 
             0
-352.737091  0
 183.575577  1

In [14]: df3.index
Out[14]: Index([-352.737091,  183.575577], dtype=object)

[ghost]

yeah, I already used it in my test to get things working, will leave this issue open,
for a look at the type-detection code in the future.

[ghost]

moved type inferencing bug into it's own issue #4601