Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

AWS Forecast. Too few observations for number of items

I am trying to do some tests with Amazon Forecast. I have tried now two different datasets which look like so:

13,2013-03-31 19:25:00,93.10999 14,2013-03-31 19:35:00,93.5 15,2013-03-31 19:40:00,93.64 16,2013-03-31 19:50:00,93.74 17,2013-03-31 20:00:00,93.8 18,2013-03-31 20:10:00,92.5

It is:

id, date, value

Both of them return back with this error:

Too few observations (633334) for number of items (633334), averaging 1.000 observations per item

But with different counts of item.

Forecast is very new so there isn't anything coming up on searches. Documentation talks nothing about this error. Really have no idea where to even start trying to get past this.

like image 580
rygo6 Avatar asked Sep 09 '19 04:09

rygo6


People also ask

Which datasets are required for generating a retail forecast with Amazon forecast?

You can provide Amazon Forecast with related time series datasets, such as the price or the number of web hits the item received on a particular date. The more information that you provide, the more accurate the forecast.

What is forecasted amount in AWS?

A forecast is a prediction of how much you will use AWS services over the forecast time period that you selected. This forecast is based on your past usage. You can use a forecast to estimate your AWS bill and set alarms and budgets for based on predictions.

What forecasting method does Amazon use?

Amazon Forecast is a time-series forecasting service based on machine learning (ML) and built for business metrics analysis.

What is the value of HPO in case of AutoML in Amazon forecast?

HPO and AutoML. By default, Amazon Forecast uses the 0.1 (P10), 0.5 (P50), and 0.9 (P90) quantiles for hyperparameter tuning during hyperparameter optimization (HPO) and for model selection during AutoML.


1 Answers

You need to identify a unique item in the data set that has multiple observations. The error you got means that none of the item in your dataset has more than 1 observation. To understand better consider this dataset:

2014-10-30 18:00:00, 144.2786069651745, client_12
2014-10-30 19:00:00, 139.30348258706476, client_12
2014-10-30 20:00:00, 133.29187396351574, client_12
2014-10-30 21:00:00, 83.95522388059702, client_12
2014-10-30 22:00:00, 60.116086235489256, client_12
2014-10-30 23:00:00, 56.17744610281925, client_12
2014-01-01 01:00:00, 56.07302533532045, client_10
2014-01-01 02:00:00, 48.435171385991076, client_10
2014-01-01 03:00:00, 43.21907600596124, client_10
2014-01-01 04:00:00, 39.1207153502235, client_10
2014-01-01 05:00:00, 37.81669150521605, client_10
2014-01-01 06:00:00, 34.836065573770455, client_10
2014-01-01 07:00:00, 34.6497764530551, client_10
2014-01-01 08:00:00, 30.737704918032748, client_10
2014-01-01 09:00:00, 27.1982116244411, client_10

You can see there are three fields, timestamp, target and item. The item i.e. client must have multiple target value observations recorded over a period of time.

To fix the error you need identify an item and the target value. Hope this helps.

like image 176
Ajit Singh Avatar answered Sep 20 '22 07:09

Ajit Singh