[lightning] Speed-up

[karl-richter]

work in progress, resolves #972

Ready from my side, follow up changes are in #999

[github-actions]

d6e4da3

Model Benchmark

Benchmark	Metric	main	current	diff
YosemiteTemps	MAE_val	1.72948	1.72949	0.0%	✅
YosemiteTemps	RMSE_val	2.27386	2.27386	0.0%	✅
YosemiteTemps	Loss_val	0.00096	0.00096	0.0%	✅
YosemiteTemps	RegLoss_val	0	0	0.0%	✅
YosemiteTemps	epoch	84	84	0.0%	✅
YosemiteTemps	MAE	1.45189	1.45189	0.0%	✅
YosemiteTemps	RMSE	2.16631	2.16631	0.0%	✅
YosemiteTemps	Loss	0.00066	0.00066	0.0%	✅
YosemiteTemps	RegLoss	0	0	0.0%	✅
YosemiteTemps	time	153.24	100.07	-34.7%	✅
PeytonManning	MAE_val	0.64636	0.64636	0.0%	✅
PeytonManning	RMSE_val	0.79276	0.79276	0.0%	✅
PeytonManning	Loss_val	0.01494	0.01494	0.0%	✅
PeytonManning	RegLoss_val	0	0	0.0%	✅
PeytonManning	epoch	37	37	0.0%	✅
PeytonManning	MAE	0.42701	0.42701	0.0%	✅
PeytonManning	RMSE	0.57032	0.57032	0.0%	✅
PeytonManning	Loss	0.00635	0.00635	0.0%	✅
PeytonManning	RegLoss	0	0	0.0%	✅
PeytonManning	time	18.88	12.49	-33.85%	✅
AirPassengers	MAE_val	15.2698	15.2698	0.0%	✅
AirPassengers	RMSE_val	19.4209	19.4209	0.0%	✅
AirPassengers	Loss_val	0.00195	0.00195	0.0%	✅
AirPassengers	RegLoss_val	0	0	0.0%	✅
AirPassengers	epoch	89	89	0.0%	✅
AirPassengers	MAE	9.82902	9.82902	0.0%	✅
AirPassengers	RMSE	11.7005	11.7005	0.0%	✅
AirPassengers	Loss	0.00056	0.00056	0.0%	✅
AirPassengers	RegLoss	0	0	0.0%	✅
AirPassengers	time	7.2	4.78	-33.61%	✅

Model training plots

Model Training

PeytonManning

YosemiteTemps

AirPassengers

[codecov-commenter]

Codecov Report

Merging #973 (d9d49a3) into main (9bf731a) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #973      +/-   ##
==========================================
+ Coverage   89.81%   89.83%   +0.02%     
==========================================
  Files          18       18              
  Lines        4632     4645      +13     
==========================================
+ Hits         4160     4173      +13     
  Misses        472      472              

Impacted Files	Coverage Δ
neuralprophet/forecaster.py	88.01% <100.00%> (ø)
neuralprophet/logger.py	100.00% <100.00%> (ø)
neuralprophet/time_net.py	89.85% <100.00%> (-0.17%)	⬇️
neuralprophet/utils.py	80.79% <100.00%> (+0.73%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[karl-richter]

@Kevin-Chen0 / everyone, lets please not update this branch with the main until we have the problem fixed - otherwise it becomes impossible for me to trace back what causes the performance overhead. Thanks!

[karl-richter]

Times
92.64, 104.6
11.79, 12.61
4.49, 4.97

[noxan]

Like the idea to check if regularization is enabled only once (instead of calling the regularization function with each iteration and figuring out it was not active) - yet we should only have one way to check if regularization is enabled or not - not multiple ones.

[noxan]

Looks great, especially the performance improvements. There are some parts on the internals config flags of lightning I trust you on. Also one follow up issue would be to reduce our regularization checks to a single implementation again.