DataFrame.replace only replaces the first occurrence of replacement pattern

[fonnesbeck]

This is best explained by a screenshot:

I'm running a pretty recent build of Pandas ('0.13.1-213-gc174c3d') on Python 2.7.5 on OS X 0.9.2.

[dsm054]

Feels like a changing-dtype issue, or maybe a mixup of values and keys (True == 1 and False == 0):

>>> df = pd.DataFrame({"a": [True, False, True]})
>>> df
       a
0   True
1  False
2   True

[3 rows x 1 columns]
>>> df.replace({True: "Y", False: "N"})
   a
0  Y
1  N
2  Y

[3 rows x 1 columns]
>>> df.replace({"a": {True: "Y", False: "N"}})
      a
0     N
1     Y
2  True

[3 rows x 1 columns]
>>> df.astype(object).replace({"a": {True: "Y", False: "N"}})
   a
0  Y
1  N
2  Y

[3 rows x 1 columns]

but I've never fully understood the intended semantics of replace.

[dsm054]

Yeah, it looks like the keys of the inner dict are being interpreted as indices to match, not values, which explains the Y/N, N/Y swap above. For example:

>>> df = pd.DataFrame({"a": [True, False, True]})
>>> df.replace({"a": {0: "zero", 1: "one", 2: "two"}})
      a
0  zero
1   one
2   two

[3 rows x 1 columns]

[cpcloud]

Are you guys running master? I fixed a similar bug somewhat recently.

[dsm054]

@cpcloud: I am, at least.

[cpcloud]

Okay thanks I'll take a look. Looks like a dtype issue. @fonnesbeck thanks for the report.

[cpcloud]

@jreback Is there a way to select and set a block by name? Something like df._data['a'] = df._data['a'].some_method()?

[cpcloud]

i want to operate on a block inplace or out of place but see the changes in the whole block manager

[cpcloud]

Nevermind.