Make pd.read_hdf('data.h5') work when pandas object stored contained categorical columns #13359
Conversation
…ork when storing a dataframe that contains categorical data.
Current coverage is 84.23%@@ master #13359 diff @@
==========================================
Files 138 138
Lines 50724 50736 +12
Methods 0 0
Messages 0 0
Branches 0 0
==========================================
+ Hits 42726 42737 +11
- Misses 7998 7999 +1
Partials 0 0
|
| @@ -4877,13 +4877,26 @@ def test_read_nokey(self): | |||
| df = DataFrame(np.random.rand(4, 5), | |||
| index=list('abcd'), | |||
| columns=list('ABCDE')) | |||
| # Categorical dtype not supported for "fixed" format. So no need | |||
|
looks good. I think there is a failing case that needs to be tested. |
| key = keys[0] | ||
| groups = store.groups() | ||
| if len(groups) == 0: | ||
| raise ValueError('No dataset in HDF file.') |
There was a problem hiding this comment.
Not sure that ValueError is the right exception here, but at least it's the same type as the exception raised by 0.18.1. (Although the message for 0.18.1 in this case is "key must be provided when HDF file contains multiple datasets.", which is a bit confusing.) And by the way, the exception raised when trying to do pd.read_hdf('empty.h5', 'some_key') is (sensibly) "KeyError: 'No object named some_key in the file'". But raising KeyError for the case where key=None and we are trying to automatically figure out the (single, valid) key in the file seems wrong to me. Let me know if you can think of a better exception (or set of exceptions) for this.
There was a problem hiding this comment.
I think this is ok. The user is explicity trying to do something which doesn't work. Its the same type of error when you have multiple keys, so its consistent. (and ok)
|
thanks @chrish42 |
git diff upstream/master | flake8 --diff