Conversation
The storage requirement will instead be controlled by the driver (hdf5).
…unsupervised learning.
… datasets folders.
__iter__() and __getitem__ are false methods. They actually are class methods, which makes their definition on the fly quite tricky. The old solution would reassing the correct definition in the __init__ method. However, since it is a class method, it also affect other object (that might have a need of the other iterator or getter). Thus, the best way to make it work is the naive way that checks each time is dataset.target is None.
There was a problem hiding this comment.
Support for Python3.4 is not release yet for Travis. See travis-ci/travis-ci#1989
There was a problem hiding this comment.
By the time you'll get around to fix CI, support should be available.
|
Folder I know we discussed it, but are we planning on supporting both Python2.7.+ and Python3.+ ? |
There was a problem hiding this comment.
This line failed when $HOME/.mldataConfig does not already exist.
There was a problem hiding this comment.
Good catch, will be fixed.
|
Why is there a capitalized D in |
|
There is no folder |
|
Oups, you're right about Regarding the uppercase D, I think it should be put in lowercase because it is refering to the file you are testing: that is |
There was a problem hiding this comment.
If the dataset is not found in the config file, _load_from_file will fail on the os.join with None. Maybe we should display a better error message.
We can now either gave the ``splits`` be given in the form (nb_train, ..., nb_test) or (nb_train, ..., nb_train + ... + nb_valid)
Creates a tuple, each containing a generator over a part of the dataset, following the given splits.
Promote the use of itertools.cycle(iter) instead.
There was a problem hiding this comment.
If lazy==False do we want data to be a ndarray? Right now it is a HDF5 dataset but still supports iteration and indexing as numpy.
2 participants
Adds :