etienne87

It is a bit vague for me so I will try to summarize in order to be corrected : we need a fully connected layer outputting 2 values, add 1...

Here is an example: ``` class GaussianPolicyOutput(PolicyOutput): def __init__(self, logits_mu, logits_var): self.logits_mu = logits_mu self.logits_var = logits_var #print("self.logits_mu.data: ", self.logits_mu.data) @cached_property def action_indices(self): # the function has same name as...

i was thinking to use torch.autograd.grad with only_inputs=True, but i never used this feature and it seems you need to call it N times otherwise RuntimeError: grad can be implicitly...

i did a first version just with numerical gradient (not the complicated autodiff stuff yet) https://gist.github.com/etienne87/0519972704e711c20706f56d55b63890 ``` grad = [None,None] for r in [-1,1]: for dim in [0,1]: bx2 =...

~~but you learn directly the gradient, or make so that the gradient of your network match the gradient of the ground truth?~~ ok so you also learn with a "numerical...

I think you might want to normalize the pixels in [0,1]. See for example (https://github.com/mikesj-public/convolutional_autoencoder/blob/master/mnist_conv_autoencode.py).

i also struggle with this, i coded my c++ rountine with std::vector and i use py::array object.cast and this is abysmally slow (around 3s for a volume of 300x300x300)... what...

in case of LSTM, shouldn't the batch be organized in (N, T, C, H, W) format?

@mbz, thanks for pointing to this. Now i am super confused with this part of the code! can you take a look at #6 ? I don't see how these...

LSTM would require reset_state func to address a specific row from the batch right? ``` class NetworkVP(): [...] def reset_state(self, idx): #todo... self.lstm_state_c[idx,...] = 0 self.lstm_state_h[idx,...] = 0 ``` sorry...