Using RELATED_TIME_SERIES Datasets
A RELATED_TIME_SERIES dataset includes timeseries data that isn't included in a TARGET_TIME_SERIES dataset and might improve the accuracy of your predictor.
For example, in the demand forecasting domain, a TARGET_TIME_SERIES dataset would
contain
timestamp
anditem_id
dimensions, while a complimentary
RELATED_TIME_SERIES dataset also includes the following supplementary features: item
price
, promotion
, and weather
.
A RELATED_TIME_SERIES dataset can contain up to 10 forecast dimensions (the same ones in your TARGET_TIME_SERIES dataset) and up to 13 related timeseries features.
You can use a RELATED_TIME_SERIES dataset only when training a predictor with the DeepAR+, NPTS, and Prophet algorithms. The R opensource algorithms (ARIMA and ETS) don't take data in a RELATED_TIME_SERIES dataset into consideration.
RELATED_TIME_SERIES Dataset Validation
A RELATED_TIME_SERIES dataset has the following restrictions:

It can't include the target value from the TARGET_TIME_SERIES.

It must include
item_id
andtimestamp
dimensions, and at least one related feature (such asstore
orlocation
). 
RELATED_TIME_SERIES feature data must be of the
int
orfloat
datatypes. 
Data frequency for a RELATED_TIME_SERIES dataset must match the TARGET_TIME_SERIES data frequency.
For example, if the forecast frequency for the TARGET_TIME_SERIES dataset is weekly, the data frequency for the RELATED_TIME_SERIES must also be weekly even if the TARGET_TIME_SERIES data frequency is daily.

The data for each item in the RELATED_TIME_SERIES dataset must start on or before the beginning
timestamp
of the correspondingitem_id
in the TARGET_TIME_SERIES dataset.For example, if the TARGET_TIME_SERIES data for
socks
starts at 20190101 and the TARGET_TIME_SERIES data forshoes
starts at 20190201, the RELATED_TIME_SERIES data forsocks
must begin on or before 20190101 and the data forshoes
must begin on or before 20190201. 
The last timestamp for every item in the RELATED_TIME_SERIES dataset must be on or after the last timestamp in the TARGET_TIME_SERIES plus the userdesignated forecast window (called the forecast horizon).
In the example RELATED_TIME_SERIES file below, the
timestamp
data for both socks and shoes must end on or after 20190701 (the last recorded timestamp) plus the forecast horizon. If data frequency in the TARGET_TIME_SERIES is daily and the forecast horizon is 10 days, daily data points must be provided in the RELATED_TIME_SERIES file until 20190711. 
The Forecast dimensions provided in the RELATED_TIME_SERIES dataset must be either equal to or a subset of the dimensions designated in the TARGET_TIME_SERIES dataset.
Important
Forecast doesn't support aggregations or filling missing values for RELATED_TIME_SERIES datasets as it does for TARGET_TIME_SERIES datasets.
Example: RELATED_TIME_SERIES File
The following table shows a correctly configured RELATED_TIME_SERIES dataset file. For this example, assume the following:

The last data point was recorded in the TARGET_TIME_SERIES dataset on 20190701.

The forecast horizon is 10 days.

The forecast frequency is daily (
D
).
This means that the user had to include data points up until 20190711.
A
"…
" row indicates all of the data points in between the previous and succeeding
rows.
timestamp 
item_id 
store 
price 

20190101  socks  NYC  10 
20190102  socks  NYC  10 
20190103  socks  NYC  15 
... 

20190601  socks  NYC  10 
... 

20190701  socks  NYC  10 
... 

20190711  socks  NYC  20 
20190105  socks  SFO  45 
... 

20190605  socks  SFO  10 
... 

20190701  socks  SFO  10 
... 

20190711  socks  SFO  30 
20190201  shoes  ORD  50 
... 

20190701  shoes  ORD  75 
... 

20190711  shoes  ORD  60 
Example: Forecasting Granularity
The following table shows compatible TARGET_TIME_SERIES and RELATED_TIME_SERIES frequencies for forecasting over the period of a week (the forecast granularity). Because data in a RELATED_TIME_SERIES dataset can't be aggregated, Forecast accepts only a RELATED_TIME_SEQUENCE data frequency that is the same as the chosen forecasting granularity.
Target Input Data Frequency  Related Time Series Frequency  Forecasting Granularity  Supported by Forecast? 

Daily  Weekly  Weekly  Yes 
Weekly  Weekly  Weekly  Yes 
N/A  Weekly  Weekly  Yes 
Daily  Daily  Weekly  No 
More Info

Step 2c: Creating a Related Dataset in Amazon Forecast: predicting timeseries at scale in the Amazon Forecast GitHub repository.