search

huseinzol05 / Stock-Prediction-Models

Gathers machine learning and deep learning models for Stock forecasting including trading bots and simulations
Apache License 2.0
7.96k stars 2.81k forks source link

Stuck in evolution-strategy agent when the price is small #44

Closed soulaw-mkii closed 5 years ago

soulaw-mkii commented 5 years ago

Hi Husein,

I am trying to swap the data source from stock to Forex price. When I train the model, the reward keep zeros.

iter 10. reward: 0.000000 iter 20. reward: 0.000000 iter 30. reward: 0.000000 :

Then I did some experiment: 1 magnify FX close price by 10000, the training resumes normal

  1. use TSLA close price divided by 1000 and the reward stuck to zero

It looks to me if the price is small, training will stuck. Do you have idea why it is?

Regards, Steve

huseinzol05 commented 5 years ago

the changes are too small during feed-forward, so the neural network not able to learn anything. Scaling is very important.

soulaw-mkii commented 5 years ago

Thanks for the prompt reply. = D