I believe the 'Reset Count' button in the DeepLearning extension does not work correctly. If the best result was at the very beginning of the network training, but auto-saving of the best results started later, the button only resets the counter without resetting the starting point of the best result. Consequently, when new best results appear, the counter does not reset because it is still affected by the result from the very beginning of the training.
Rename
Another wish for improving DeepLearning. Please make it possible to save default network parameters so that there is no need to change network parameters every time.
I am currently using the following network parameters. With these parameters there are no big jumps in the learning curve with my input layers.
With these parameters, the learning process proceeds perfectly. However, it's a shame that auto-stop cannot be applied, since the 'best' result at the very beginning of training prevents the counter from resetting when current best results are achieved.
It's working as designed. Resetting the count is intended to do only that, reset the count for when checking for another best result will occur. It's not intended to reset the best result.
Is it possible to manually set hidden layer size (neurons) in Torch engine to a specific value ?
Actually Found an answer in setting TorchSharp LSTM model
Actually Found an answer in setting TorchSharp LSTM model
Your Response
Post
Edit Post
Login is required