yangsenius / learning-to-learn-by-pytorch Goto Github PK

@yangsenius Thanks for your helpful implementation. There seems a minor bug during runtime.
The line for CUDA configuration is redundant to avoid potential GPU utilization.

In the process of program training, GPU memory leaks and GPU memory keeps increasing until out of memory. Can you figure this bug ?

Hello,
It seems in the learn function, the variable name should be optimizer instead of optimizee as LSTM is an optimizer and not optimizee. Can you clarify this?

I have some confusion here.

Thanks.
Rahul

Why is preprocess=False in the code? If it is changed to Ture instead, NAN appears.

when i run:
RuntimeError: can't retain_grad on Tensor that has requires_grad=False

As I read the original paper and the repo from deepmind, it seems to me that the LSTM optimizer should only take 1 variable as input to optimize and save the LSTM state for each variables. In other words, with an arbitrary number of parameters, it just updates one after another. While in this implementation, the dimension of the optimizer is fixed as the number of the optimizee.