SARIMAX prediction LinAlgError: Non-positive definite forecast about rltrader HOT 15 CLOSED

I found this issue that seems to be about the same issue (I also get it).

Possibly adding: simple_differencing=True where SARIMAX is called at this line here will solve this.

Will try it myself and report if it prevents this error or not.

from rltrader.

This fix seems to be working for me - upgrading to statsmodel 0.10.0rc2

from rltrader.

This is "normal", as in I have also run into this. However, it causes trials to fail unnecessarily, which is not ideal. I have spent a bit of time debugging this specific error, and have not been able to find a solution. Has anyone else had any success?

from rltrader.

The error seems to be caused by a divide-by-zero (ZeroDivisionError: float division) within the SARIMAX.fit() function, though I am unsure how to prevent it currently.

from rltrader.

Great, I'd love to hear your results.

from rltrader.

Think that works. I ran optimize.py for a while and didn't get that error again.

from rltrader.

Great, I will add the change to the repo and get it pushed.

from rltrader.

I was about to submit a PR. Will and hope you see this.

from rltrader.

@JohnAllen thanks, I will wait for your PR!

from rltrader.

After running optimize.py for a couple hours with simple_differencing=True, I have run into the issue again. Unfortunately, this does not seem to have fixed the issue.

from rltrader.

The thread above mentions upgrading the statsmodels package to version 0.10.0rc2 to fix the issue. I will try this.

from rltrader.

Confirm, its gone now...

from rltrader.

The thread above mentions upgrading the statsmodels package to version 0.10.0rc2 to fix the issue. I will try this.

Run for a couple of hours now, seem to be working

from rltrader.

Just ran into this as a new user. I recommend you change requirements.txt to

statsmodels==0.10.0rc2

from rltrader.

Glad to hear this fixed the issue for everyone. Pushed the fix

from rltrader.