avsecz / kopt Goto Github PK

As far as I understand from expressions such as np.arange(min(n_train, train[1].shape[0])) in test_fn, targets must be list/tuple right now, although they can be dict in Keras to allow models with multiple outputs. So it would be nice to add dict target support, at some point. So I am adding this issue just as a TODO item.

Is there any possibility of using a generator in the data function for datasets that don't fit in memory? Keras' fit_generator method doesn't seem to be used anywhere.
Any suggestions on how this could be achieved?

As described here, custom_objects argument of keras.models.load_model must be used for loading models with custom loss/layer/activation function. Adding here as a TODO item.

I am adding the code here, which is adjusted to my needs. Probably it will not suit the needs of other users in the stage in which it is now, because the support for variable number of data units is needed. Also the early stopping should be optional. P.S. issues doesn't support .py extensions.
hyopt_copy.txt

The current version of kopt on PyPi is not compatible with pyyaml>=6.0 due to yaml/pyyaml#576. This has been solved for 15 months in 6a5c890, but this change is not on PyPi. Can you please publish a new release?