Section 10: Model - 10 (Turkey Problem) Results mismatch

[ysoniwal]

Hi

In model 10 (adding turkey data to bitcoin price prediction), my model is not performing as bad as it is in the lecture videos. I went through the course book website and I think there is a mismatch in results that I am not able to understand.

When we evaluate the model on (X_test, y_test), the loss is ~638. But when we use evaluate_pred method to calculate results for turkey_preds, the mae value is ~17144. This should be exactly equal to 638 because the loss function for the model is mae.

Refer the screenshot from https://dev.mrdbourke.com/tensorflow-deep-learning/10_time_series_forecasting_in_tensorflow/#building-a-turkey-model-model-to-predict-on-turkey-data

Let me know if I am missing anything.