Julius Kunze

You can already use`random_key()` within `@parametrized`: ```python @parametrized def dropout(inputs): keep_rate = 1 - rate keep = random.bernoulli(random_key(), keep_rate, inputs.shape) return np.where(keep, inputs / keep_rate, 0) ``` An independent `seed`...

Hey! You can download it from https://clarin.phonetik.uni-muenchen.de/BASRepository/, but (if I remember correctly) you have to register/login to the website first.

Hi @sruteesh, thank you! ASG is not supported, but an implementation can be plugged in at https://github.com/JuliusKunze/speechless/blob/master/speechless/net.py#L397.

@jakevdp I hadn't yet (probably should have), thanks for doing this, very interesting! How would you explain this behavior, i. e. is branching optimized out during compilation or just dirt-cheap?...

Hi, thanks for reporting this! It looks like mp4 is not supported by Tensorflow IO on Windows yet: https://github.com/tensorflow/io/issues/1064. Tensorflow datasets don't depend on TF IO. However, the project description...

The default configuration uses original parameters from the paper, listed in `config.py`. There are no other settings. `num_examples=100` and `num_samples=1`, perhaps you are confusing the two?

`eval.py` and https://github.com/juliuskunze/cwvae-jax/blob/main/tools.py#L50 should have the details you are looking for. On the day of release it was up to date, so it must have been (roughly) [jax 0.2.20](https://github.com/google/jax/releases/tag/jax-v0.2.20) and...

To me that looks like it has no impact on accuracy.

You're welcome, I hope it helped!

Since the JAX version seems to be consistent with the original code, it's probably best if you directly ask the paper authors, i. e. @vaibhavsaxena11. (As you have done already...