DataFrame.replace() doesn't work correctly when passing a nested dict

[estevopaz]

In [47]: import pandas as pd

In [48]: df = pd.DataFrame({'col': [1,2,3,4]})

In [49]: df.replace({'col': {1: 'a', 4: 'b'}})
Out[49]: 
  col
0   a
1   2
2   3
3   b

In [50]: df.replace({'col': {-1: '-', 1: 'a', 4: 'b'}})
Out[50]: 
  col
0   a
1   2
2   3
3   a
^^^^ Value 4 must be mapped to "b" instead of "a"

In [51]: pd.__version__
Out[51]: '0.12.0'

[cpcloud]

Can you reproduce this on master ie the latest commit

[estevopaz]

Yes, same problem. And not only with -1 value, also when there are no lineal sequence in IDs, for example {10: 'a', 1: 'b', 3: 'c'}.

Thanks a lot for your attention.

[cpcloud]

@estevopaz I'm not sure what you mean by

when there are no lineal sequence in IDs

The key value pairs in a Python dict are arbitrarily ordered, so the dicts {1: 'a', 2: 'b'} and {2: 'b', 1: 'a'} are equivalent.