Bugfix: return correctly scaled multiplicative seasonality component

[denyslazarenko]

Resolve issue: https://github.com/ourownstory/neural_prophet/issues/429 Used hint: trend = trend(scaled)*y_scale + y_shift season_x = season_x(scaled) *y_scale season_x_m = season_x_m(scaled)*trend(scaled)y_scale season_x_m = season_x_m(scaled)(trend - y_shift)

[ourownstory]

based on our conversation's new logic: for all components (except trend)

• multiply with scale Then, for all multiplicative terms:
• multiply with (raw/normalized) trend Finally
• multiply trend with scale, add shift

Now, multiplicative terms are no longer in 0-1 scale -> need make sure plotting does not get messed up

[rbeldagarcia]

Hello @ourownstory and @denyslazarenko,

Thanks in deep for your clarification. However, it is still not clear for me how to transform the data. In particular, I have the following questions:

• trend(scaled): does it mean to scale between 0 and 1 the trend?
• what do you mean by y_scale?
• what do you mean by y_shift?
• Once I have performed these transformations, can I apply the following formula? Yhat = trend * (1 + multiplicative terms) + additive terms

A code example will be great.

I would like to take this opportunity to thank you for the excellent work you are doing on this project.

Best,

[ourownstory]

merged in #791