ML Flatlined

Unity Engine
,
1

So, I was watching the progress on tensor flow for a day or two and the progress kinda flatlined and went erratic around 80M

I tried the .NN file and it was basically useless.

I know the training is working fine as I periodically stop and test the NN file, no issues, however, not sure why it did this… and can I go back to a better result?

Is this a typical issue if the ML decides to go bonkers after training so long?

6139071--669852--upload_2020-7-27_19-18-35.png
6139071--669852--upload_2020-7-27_19-18-35.png1418×479 43.8 KB

2

It would be nice to save the NN file based on the highest degree of reward or other factors as it trains…

3

This is an old post, but the basic method still works with current ml-agents:
https://github.com/Unity-Technologies/ml-agents/issues/1047#issuecomment-409660869