Closed AVPokrovsky closed 1 week ago
@FernandoCarazoMelo
Hi @AVPokrovsky ,
Given that this issue is specific to the RNN forecaster and not others, it might be related to the way RNNs handle memory and state across iterations.
RNNs can be memory-intensive, and repeated training without proper memory management can lead to memory leaks or exhaustion. Since you are using TensorFlow, ensure that you are freeing up memory explicitly if possible
Try it. If that doesn't work, to help better diagnose and resolve the problem effectively, I need a bit more information from you:
Environment Details: Operating System and version Python version TensorFlow version Batch Size
Minimal code snippet that reproduces the error would be extremely helpful. This allows me to understand the context better and try to replicate the issue on my end.
Hi, I added in the loop gc.collect() and have an impression that it was resolved, but not sure because tried on different data and different rnn parameters. Anyway, it passed 2661 iterations. More than in first case.
That's great. Let us know if you have any other issues with it. @JoaquinAmatRodrigo , we can close this issue.
Hi, unfortunately I was wrong. The issue still exists..
Environment Details: Operating System and version Windows 11 Home edition Python version 3.10.4 TensorFlow version 2.15.1 Batch Size 128
Hi, my program executes forecaster.fit(series=data_m) in a loop for ForecasterRnn. After 1908 iterations error appears. It seems kind of Windows fatal error. Listing is attached. The error does not appear for other forecasters, only for Rnn. I guessed it is because of recursion and added sys.setrecursionlimit(10000), but it does not help.
LSTM.txt