Why the results are different for make_evaluation_prediction and predictor.predict

[ethanqi1109]

For example,
Training dataset: 2020-01-01 ~ 2020-01-25
Testing dataset: 2020-01-01 ~ 2020-01-31

after trained a DeepAR model with training dataset,
predictor = model.train(train_set)

the forecast of evaluation for 2020-01-25 ~ 2020-01-31 is,
forecast_it, ts_it = make_evaluation_predictions(
dataset=test_set,
predictor=predictor,
num_samples=num_samples
)

As my think, there is another way to evaluate will get the same result:
forecast_it = predictor.predict(train_set)

However, I got two results with big gap, therefore I'd like to know is there any difference between "make_evaluation_predictions" and "predictor.predict"?

Thanks in advance.

[vafl]

That should give the same forecasts if your train set is the same as the test set except that the the last prediction_length time points have been removed from each time series -- this is what make_evaluation_prediction does before the forecast is generated. Can you maybe double check this?