Dzmitry Bahdanau
Dzmitry Bahdanau
The usecase that you refer to is valid. I didn't know that there is some support in the core for deletion of children, @bartvm wrote that part. Would like to...
@dwf, looks like I approved this PR more than a month ago. Do you wish to proceed and merge it or should I close it?
We have `AbstractDataStream` mostly because we have `DataStream` and `Transformer` and they need a common base class. There are also arguments for having `Dataset` instead of `AbstractDataset` just because it...
I am not sure if it makes sense to speak of a generic implementation of dataset. A dataset deals with a data format, and there is no default data format...
Actually when you work with recurrent network, `get_dim` argument is usually not a role. It accepts a name to distinguish between different inputs and output, of which you sometimes have...
Regarding `input_dim` and `output_dim`: not all the bricks that people work with are feedforward. Only feedforward bricks have those properties. For other bricks, `get_dim` makes a lot of sense. I...
> Pickle isn't really the right kind of serialization format for this, I feel. Can I ask you to elaborate a bit?
I also like the idea of robust unpickling, though my current knowledge how pickling works is insufficient to see if it is possible.
For the record: I read https://docs.python.org/2/library/pickle.html#subclassing-unpicklers and was moderately inspired by the perspective to work with it. Also I am a bit worried about the performance impact that having a...