neo-alex

@wochinge Fine thanks, sorry for the delay - I will try to do required changes today and will let you know in case I need some help :)

Alright, I tried to answer all your comments above with associated micro-commits. Let me know what you think when you get a chance to review again @wochinge Cheers

I am looking forward to this DrQ implementation! Just wondering if it wouldn't be better to rather reproduce DrQv2 directly? It should be the same amount of work but the...

Hello, The ususal assumption in RL is that you are **not** necessarily able to initialize the environment from a given state (like in real life) - only an argument-free reset()...

I'm not sure either about the "scene reset", however I definitely see value in the ability to toggle rendering on/off at any time, especially when training (e.g. with RL): it...

Thank you for your reply, but I am not sure to get your point about "gradient recording issues". In my understanding, jit-compilation is the new default in Keras 3 (at...

In the meantime, I also tried with "torch" backend and everything works fine, like with "tensorflow" (so the issue mentioned above seems specific to JAX with JIT compilation)

My bad, I think the issue is solved if I change the call function of my MaskedGlobalAveragePooling1D to: ```python def call(self, inputs): mask = self.masking.compute_mask(inputs) return self.pooling(inputs, mask=mask) ``` Still,...