Difference in yhat with and without quantiles specified

[tuttoaposto]

I noticed the point forecasts in the test set shifted upwards and away from the actuals after specifying quantiles as in NeuralProphet(quantiles = [0.025, 0.975]), because I wanted to generate conformal prediction intervals in the next step. This increased MAPE from 0.034545 to 0.041148. Why is that?

What should I do to make the point forecasts stay the same while adding valid prediction intervals?

[leoniewgnr]

Did you set the random seed? `set_random_seed(0)'

[tuttoaposto]

Yes, I did.