Core and memory problem

Hi,


I am currently trying to create a LSTM model to make stock price predictions.

The problem I am faced with is that it seems we only have 4000MB available, which prevents me from doing a complex enough model. I also struggle to launch simple models (2 layers of 25 to 50 neurons) because the core crashes in the middle of the fit (with only one epoch).

Is there something we can do to avoid both issues?


Thank you in advance for your time,


Julien

Hi Julien

The kernel will crash each time you overflow your memory. You should make sure you manage resources accordingly.

What BTPP window do you use? You can reduce it if too long to have a lighter model. 2 layers of 25 to 50 neurons should be quite light. Make sure you train it by batch. Decrease your batch size too, that should help.


Let us know if that helps already

Thanks