Jan Schlüter
This changes the bit of the code which collects hyperparameters from command line flags into a named tuple. Since Nov 6 (commit 2652704), TensorFlow uses `absl.flags`, which means `FLAGS.__flags` does...
This issue shall serve as a place to announce and discuss new results, to avoid the discussion being spread over several pull requests that just happened to be there (#11,...
When passing a two-dimensional input layer to LSTMLayer, it will break with an uninterpretable error message: ``` >>> lasagne.layers.LSTMLayer((10, 20), 30) [...] TypeError: 'numpy.float64' object cannot be interpreted as an...
Upscale1DLayer, Upscale2DLayer, Upscale3DLayer share a lot of code (and tests). They could be refactored to be derived from a common UpscaleLayer that supports any number of spatial dimensions, or maybe...
Would be great to implement the initialization scheme proposed in https://arxiv.org/abs/1707.02937 and recommend it in the `TransposedConv2DLayer`. Could be a "meta-initializer" that can be used to wrap any of the...
Dilated convolution is currently implemented as a convolution backward pass because Theano did not have any other support for this. For performance reasons (avoiding a dimshuffle -> gpu_contiguous), the weights...
The [module docstring for `lasagne.layers.recurrent`](http://lasagne.readthedocs.io/en/latest/modules/layers/recurrent.html) has an example that uses reshapes to put a dense layer on top of a recurrent layer -- this should be updated to use the...
This was asked for on the mailing list: https://groups.google.com/forum/#!topic/lasagne-users/vYIzqT1Wn7g For the implementation, the constructor would need to create new shared variables for all params of the layer to be inverted,...
The guide at http://lasagne.readthedocs.io/en/latest/user/custom_layers.html should be extended to cover at least two more topics: * A layer with multiple inputs (derive from MergeLayer) * A layer with dynamic output shape...
This is a proposal to resolve the long-standing discussions on multiple-output layers (#337) and merge layers accepting dictionaries (#537), both required for recurrent containers (#629). It was originally formulated in...