Error evaluating model log probability: Non-finite gradient.

[liangzulin]

I experience this error and failure. Is it the input data's problem or other? Actually, there may be some zero in time series Data Frame. How to avoid this happen?

Actually, I try to make Prophet integrate in Spark and running in cluster mode. each map job is a single Prophet forecast job. I install Prophet Version 0.1.1 to each Spark node. When my spark job failure, I collect logs from yarn and got these:

INFO:fbprophet.forecaster:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.
Initial log joint probability = -27264.6
    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes
      99       11.5285     0.0188875       100.973           1           1      148
    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes
     152       15.5503     0.0122178       115.865   9.111e-05       0.001      250  LS failed, Hessian reset
     199       16.4349   3.90987e-07       100.279      0.9131      0.9131      313
    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes
     219       16.4881   0.000391014       100.237   3.901e-06       0.001      381  LS failed, Hessian reset
     262       16.5172   8.72562e-09       100.227      0.2859           1      439
Optimization terminated normally:
  Convergence detected: absolute parameter change was below tolerance
('cap_value:', 1793)
INFO:fbprophet.forecaster:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.
Initial log joint probability = -27264.6
    Iter      log prob        ||dx||      ||grad||       alpha      alpha0  # evals  Notes
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.

Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.

      37      -7004.34      0.806687       277.106       0.001       0.001      658  LS failed, Hessian reset
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.
Error evaluating model log probability: Non-finite gradient.

Thanks

[bletham]

I'm guessing that something with this particular dataset is leading to poor model fit and numerical issues. Would you be able to post a dataset that produces this issue?

[LastAncientOne]

error: INFO:fbprophet.forecaster:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.

stanfit4anon_model_1988b9d517e3c16daf27a07f09e3de97_2863119867656750088.pyx in stanfit4anon_model_1988b9d517e3c16daf27a07f09e3de97_2863119867656750088.StanFit4Model.init()

TypeError: init() takes exactly 2 positional arguments (1 given)

I am running daily data and using stock prices. Can you tell me why is not working?

[bletham]

@LastAncientOne this looks like the issue from #324. In short, you need to force a re-install of fbprophet; see the commands at that issue.

[bletham]

I'm hoping this has been resolved, but if you're still running into this issue and can post data then please re-open.