[feature] separate panel uncertainty plot

[LeonieFreisinger]

Referencing issue #710
Problem
The plot() function already includes plotting the uncertainty. However, the plot_components() function does not include plotting the uncertainty of yhat as a component.

Solution
Modify the plot_components() function to plot the uncertainty of yhat as a component in a separate panel. Therefore, the matplotlib and plotly version will be modified. The uncertainty will be plotted with the x-axis = 0.5 quantile as reference.

[codecov-commenter]

Codecov Report

Merging #779 (80dc5cf) into main (aa99674) will increase coverage by 0.05%.
The diff coverage is 92.85%.

@@            Coverage Diff             @@
##             main     #779      +/-   ##
==========================================
+ Coverage   87.00%   87.06%   +0.05%     
==========================================
  Files          16       16              
  Lines        4580     4622      +42     
==========================================
+ Hits         3985     4024      +39     
- Misses        595      598       +3     

Impacted Files	Coverage Δ
neuralprophet/forecaster.py	86.26% <62.50%> (-0.22%)	⬇️
neuralprophet/plot_forecast.py	75.11% <95.83%> (+2.20%)	⬆️
neuralprophet/plot_forecast_plotly.py	87.55% <100.00%> (+1.00%)	⬆️

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

[Kevin-Chen0]

Thanks for all your work @LeonieFreisinger! I'll review this shortly and will give you feedback if any.

[LeonieFreisinger]

@Kevin-Chen0 Thanks a lot!! :) I just added an additional test in test_plotting.py to improve code cov.

[ourownstory]

Some improvement ideas:

• Use same color schema as in main uncertainty plotting (as in plotly)
• require highlight_nth_step_ahead when n_forecasts > 1 and quantiles in use

[Kevin-Chen0]

Hey Leonie, maybe you can also modify the uncertainty_estimation_peyton_manning.ipynb example notebook so the m.plot_components(forecast) are so plotting the uncertainties. This notebook is a good way to illustrate your changes around uncertainty/quantiles.

[Kevin-Chen0]

Also, should the m.plot_parameters() include uncertainty as well or just m.plot_components(forecast)?

[LeonieFreisinger]

@Kevin-Chen0 @ourownstory thanks a lot!
I will modify the notebook. I think the uncertainties shouldn´t be in the m.plot_parameters()

[LeonieFreisinger]

@ourownstory code check fails because of test_regularization_lagged_regressor. This should be fixed when merging #794 into main and rebasing this branch.

[ourownstory]

Also, should the m.plot_parameters() include uncertainty as well or just m.plot_components(forecast)?

we could allow the user to decide to plot a specific percentile, instead of hard-coding the mean.

[ourownstory]

In all your string formatings, please use f-strings.

[ourownstory]

Please also remove warnings printouts from the tutorial notebooks.

[LeonieFreisinger]

Also, should the m.plot_parameters() include uncertainty as well or just m.plot_components(forecast)?

we could allow the user to decide to plot a specific percentile, instead of hard-coding the mean.

okay. I would suggest following up on that feature in a new PR. Is that okay? @ourownstory

[ourownstory]

LGTM