-
It looks like the initial hidden layer in both LSTM Generator and RelGAN_G are chosen so they are the same every time. Wouldn't that reduce the diversity of generated samples? What makes sure that the…
-
All the GS wrappers other than Gumbel Softmax are broken if temperature updating is enabled.
```
assert hasattr(agent, 'gs_layer') and hasattr(agent.gs_layer, 'temperature'), 'Agent must have a `tem…
-
Can you talk about the details of "samples one sub-graph at one training iteration"?
As far as I know, the result of Gumbel Softmax may not be a one hot vector. It may be a vector like [0.96, 0.01,…
-
Hi, Thank you for your excellent work!
While searching for width, I wonder whether the operations with different channel numbers share the weight from the one-shot network? If yes, how does the weigh…
-
Hi, I have some problems with your implementation of IRGAN. It seems that your implementation doesn't utilize a pre-trained model to initialize the weights of the generator or discriminator as the off…
-
Hi, I testing training model on new versions of repo, and I have some troubles with 0.0.14 and 0.0.15.
On 0.0.14, model always return nan on forward pass, version 0.0.15 lead to CUDA error:
```pyt…
-
I am using `Keras.Backend.armax()` in a gamma layer. The model compiles fine but throws an error during fit().
ValueError: An operation has `None` for gradient. Please make sure that all of you…
-
First of all, thanks for your great work!
However, I have some questions about training algorithm while searching. Since in back-propagation, soft max function is applied, will the gradients genera…
-
We have introduced `testing.FunctionTestCase` (#3499).
This tool enables writing a function test for various backends (including ChainerX) in much simpler way.
This tool also provides an option to a…
-
```python
helper_2 = tx.modules.GumbelSoftmaxEmbeddingHelper(
# embedding=self.encoder.word_embedder,
start_tokens=torch.tensor([101]*2).long(),
…