grosestq

> My expectation is to get a prediction that is not completely constant and can capture some volatility in the curve. Hi, is it really completely constant, as you said?...

> It learns absolutely no volatility compared to similar models regardless of features added or hyperparameter tuning. the change in forecast is minimal (hence almost constant). How does the prediction...

Is that working: model = TemporalFusionTransformer.load_from_checkpoint(best_model_path) predictions = model.predict(test).cpu() ?