allow missing data for "ts_forecast_panel" task

[int-chaos]

Why are these changes needed?

• allow missing data for "ts_forecast_panel" task
• address typo for hcrystalball_model
• add test for missing data scenario in test_forecast.py
• fix issue where fit_kwargs were not actually passed into the model fit()

Related issue number

• closes Time series gap detection for TFT tasks #754 tft gap detection on each time series independently #770
• addresses part 1 of pytorch TimeSeriesDataSet handling missing time stamps - for TemporalFusionTransformer model #771
• closes `hcrystaball_model` #839

[int-chaos]

there is a current error with missing timestamps test. please advise if you can.

missing timestamps causes error with cross validation.

With the current seeding, 04/01/2017 data is removed for Agency_13 SKU_04, but the time range 01/01/2017 - 06/01/2017 is used as the testing portion for holdout validation. Even though 04/01/2017 Agency_13 SKU_04 is not part of the testing data (in this case testing data is 01/01/2017 - 06/01/2017 for all agency and sku groups, with the exception of 04/01/2017 Agency_13 SKU_04), when the model predicts it still predicts a value for 04/01/2017 Agency_13 SKU_04, thus resulting in a ValueError where the testing length and prediction length are different.

I've been trying to come up with a way to remove the prediction for 04/01/2017 Agency_13 SKU_04, but there is no easy way to locate it since the raw prediction is just a tensor.

[sonichi]

there is a current error with missing timestamps test. please advise if you can.

missing timestamps causes error with cross validation.

With the current seeding, 04/01/2017 data is removed for Agency_13 SKU_04, but the time range 01/01/2017 - 06/01/2017 is used as the testing portion for holdout validation. Even though 04/01/2017 Agency_13 SKU_04 is not part of the testing data (in this case testing data is 01/01/2017 - 06/01/2017 for all agency and sku groups, with the exception of 04/01/2017 Agency_13 SKU_04), when the model predicts it still predicts a value for 04/01/2017 Agency_13 SKU_04, thus resulting in a ValueError where the testing length and prediction length are different.

I've been trying to come up with a way to remove the prediction for 04/01/2017 Agency_13 SKU_04, but there is no easy way to locate it since the raw prediction is just a tensor.

What does it mean by `With the current seeding, 04/01/2017 data is removed for Agency_13 SKU_04"? Why is it removed?