Changes in .ix behavior that break backwards compat

[twiecki]

It seems like there was major refactoring of the .ix[] indexing in 0.13 that breaks backwards compatibility in some ways.

E.g. .ix[row][col_name] = x used to work but doesn't. Sometimes .ix[row, col_name] = x also doesn't work and I have to use .loc[row, col_name] = x instead although the docs state that .ix[] just falls back to either .loc or .iloc.

I'm changing most of my code but I guess I'm just wondering whether that was made explicit as I can't find it in the release notes or if unittests failed to catch it?

[jreback]

see here and related: #6058

fixed in 0.13.1

your options for 0.13 are to upgrade numpy to 1.8
or use ix

you are doing chained indexing assignment. not guaranteed to work. (it will work in 0.13.1, but is in general a bad idea)

ix will always work, if you have a specific case that it doesn't, put up as a new issue

[jreback]

@twiecki appreciate the bug reports though.....

believe me wish had caught this before 0.13...but that's why we have 0.13.1...had lots of moving parts

keep reporting!

[twiecki]

@jreback Appreciate the hard work you guys are doing. I suppose the success of pandas is also a burden in the way that if something breaks many people start complaining ;).

[twiecki]

There's another thing I noticed for which I'm happy to open up an issue if you don't know about it.

I noticed that that numpy-style access breaks sometimes under 0.13. While I haven't been able to pin-point the issue, calls like pylab.hist(-df.ix[row, col_name]) fail with some x[0] index error and I have to use pylab.hist(-df.ix[row, col_name]).values. Only sometimes though and I haven't been able to create a reproducible case.

[jreback]

hmm...ok...

do a try except around those statemetns and if it breaks, then either save a pickle/csv of the frame if you can
so can take a look (and dump whatever info you can)....

its the reproduce that is important as you know :)

[twiecki]

OK, issue opened at #6127.