How to set the start date for the predictions?

[ivihernandez]

The start_date of my forecast starts immediately after the last date of my test dataset, whereas I would expect it to start on the first date of my test dataset.

Lets say that test dataset starts on TS and ends on TE. After calling the predict function, the object gluonts.model.forecast.SampleForecast start_date shows TE + 1 instead of TS. Not sure what is going wrong. Below is a code sample:

train_dataset = PandasDataset.from_long_dataframe(
    dataframe=train,
    target='target',
    item_id='item_id',
    timestamp='timestamp',
    freq='D')

test_dataset = PandasDataset.from_long_dataframe(
    dataframe=test,
    target='target',
    item_id='item_id',
    timestamp='timestamp',
    freq='D')

estimator = DeepAREstimator(
    freq='D',
    prediction_length=30,
    trainer=Trainer(epochs=1)
)

predictor = estimator.train(train_dataset)

test_predictions = predictor.predict(
    test_dataset,
    num_samples=100) 

list_of_test_forecasts = list(test_predictions)

# some sample outputs
test_dataset.freq # 'D'
test_dataset.timestamp # timestamp
test.iloc[0].timestamp # Timestamp('2022-08-13 00:00:00')
test.iloc[len(test) -1].timestamp # Timestamp('2022-09-11 00:00:00')
list_of_test_forecasts[0].start_date # Period('2022-09-12', 'D') -- I would expect here 2022-08-13

[ivihernandez]

Hi @lostella , just checking to see if you had any insights for this by any chance. Thanks!