[Minor] Make tests deterministic

[MaiBe-ctrl]

• Add flag to make tests deterministic to better track the metrics and performance of the fitted models.

[github-actions]

Model Benchmark

Benchmark	Metric	main	current	diff
PeytonManning	MAE_val	0.35511	0.35511	0.0%	✅
PeytonManning	RMSE_val	0.50375	0.50375	0.0%	✅
PeytonManning	Loss_val	0.01801	0.01801	0.0%	✅
PeytonManning	train_loss	0.01463	0.01463	0.0%	✅
PeytonManning	reg_loss	0	0	0.0%	✅
PeytonManning	MAE	0.34734	0.34734	0.0%	✅
PeytonManning	RMSE	0.49386	0.49386	0.0%	✅
PeytonManning	Loss	0.01463	0.01463	0.0%	✅
PeytonManning	time	16.2464	14.25	-12.29%	🎉
AirPassengers	MAE_val	30.8786	30.8786	0.0%	✅
AirPassengers	RMSE_val	31.8365	31.8365	0.0%	✅
AirPassengers	Loss_val	0.01303	0.01303	0.0%	✅
AirPassengers	train_loss	0.00071	0.00071	0.0%	✅
AirPassengers	reg_loss	0	0	0.0%	✅
AirPassengers	MAE	6.85845	6.85845	0.0%	✅
AirPassengers	RMSE	8.90745	8.90745	0.0%	✅
AirPassengers	Loss	0.00076	0.00076	0.0%	✅
AirPassengers	time	10.2769	8.59	-16.41%	🎉
EnergyPriceDaily	MAE_val	5.64156	5.64156	0.0%	✅
EnergyPriceDaily	RMSE_val	7.19554	7.19554	0.0%	✅
EnergyPriceDaily	Loss_val	0.0289	0.0289	0.0%	✅
EnergyPriceDaily	train_loss	0.02962	0.02962	0.0%	✅
EnergyPriceDaily	reg_loss	0	0	0.0%	✅
EnergyPriceDaily	MAE	6.40067	6.40067	0.0%	✅
EnergyPriceDaily	RMSE	8.56962	8.56962	0.0%	✅
EnergyPriceDaily	Loss	0.02941	0.02941	0.0%	✅
EnergyPriceDaily	time	19.9239	17.51	-12.12%	🎉
YosemiteTemps	MAE_val	0.48186	0.48186	0.0%	✅
YosemiteTemps	RMSE_val	0.71525	0.71525	0.0%	✅
YosemiteTemps	Loss_val	0.0003	0.0003	0.0%	✅
YosemiteTemps	train_loss	0.00103	0.00103	0.0%	✅
YosemiteTemps	reg_loss	0	0	0.0%	✅
YosemiteTemps	MAE	0.85851	0.85851	0.0%	✅
YosemiteTemps	RMSE	1.54684	1.54684	0.0%	✅
YosemiteTemps	Loss	0.00103	0.00103	0.0%	✅
YosemiteTemps	time	45.3692	39.31	-13.36%	🎉

\n

Model training plots

\n ## Model Training ### PeytonManning  ### YosemiteTemps  ### AirPassengers  ### EnergyPriceDaily  \n

[ourownstory]

Great to see the tests be deterministic now!

I looked through the code and have 3 thoughts:

• What do you think about moving the deterministic flag from the NeuralProphet class init to the train method? AFAIK it only affects training?
• Can you please update the tutorial on reproducibility to include this new flag?
• Looks like many of the 'glocal' seasonality tests are removed - is this intentional?

[MaiBe-ctrl]

• Yes, actually it makes more sense to have the flag in the train rather than the model itself. I will update it.
• For some reason, some of the tests in glocal are duplicated. That's also the reason for the linting issues.

[ourownstory]

Thank you!

[ourownstory]

change to >=2.0.0