How to train multi item timeseries forecasting in Sagemaker?

Question

I have data like below.

   date      sku   unitprice    trand_item    target
2018-01-01    A      10            Black        3
2018-01-02    A      10            Black        7
2018-01-03    A      10            Black        0
2018-01-04    A      10            Black        13
.
.
.
2017-08-01    B      20            White        4
2017-08-02    B      20            White        0
2017-08-03    B      20            White        17
2017-08-04    B      20            White        9
.
.
.

Every timestamp is filled in 'D' without blank and 'sku' is item number. I have 25 items and i want to forecast 'target'. Also want to use 'unit price', 'trand_item' for meta data.

How can i train timeseries forecasting model in sagemaker? 1 model for 25 items. (For example, i want to forecast 30days for each item's 'target'.)

Please help me...

Answer 1

If you have at least 300 rows you can use DeepAR. It supports grouping, such as by SKU

Otherwise, you'd want to bring your own algorithm, such as sktime