Questions tagged with Amazon Forecast

Using the Amazon Forecast UI, I successfully created a forecast. But it does not allow me to update my dataset. In the dataset group dash board, I go to datasets, click on my target time series dataset name (previously uploaded), under 'dataset imports', I click 'create dataset import'. In the 'Create target time series dataset' page I go to dataset import details section and provide the the dataset import name and data location. Leave the lambda role the same and click 'start' but nothing happens. I don't get any error message. Simply nothing happens. Why?

I have a few questions regarding data preparation for Forecast. I have a dataset with about 3,000 item_id's, the data is recorded on weekdays only (no row for weekends/holidays), and the forecast horizon is 1 day. For example: [item_id | timestamp | target_value] item_A | 2022-01-19 (Wed) | 100 item_A | 2022-01-20 (Thurs) | 101 item_A | 2022-01-21 (Fri) | 99 item_A | 2022-01-24 (Mon) | 98 item_A | 2022-01-25 (Tues) | 102 Q1. Is it recommended that the weekends (1/22, 1/23) row is inserted to the dataset with NaN as the target_value? Q2. If target_value for a timestamp is NaN, do the RTS attributes get ignored regardless of what value it is? Q3. When the Forecast is training, does Forecast recognize that on Friday, the next value to predict is on Monday rather than on Saturday? Q4. If an item_id has small time series data points (for instance, global start~end date ranges from 2018 to 2021, but a particular item_id only has data recorded for a couple months in 2020), should I front-fill & back-fill with NaN to match the global start~end range? (I only intend to use the data for training, not creating a forecast)

When I am creating an AWS Forecast Export, it creates multiple small CSVs in the S3 bucket. Seemingly 1 for each item_id that I am making predictions for. My question is, is there a setting I can change to export all of this information as 1 CSV file? I am using the AWS CONSOLE, not the CLI. Thanks in advance, I am new to AWS Forecast and I could not find this question on the forums via search.

On Amazon Forecast, how can I export a model (a Predictor in Forecast lingo) that I already trained? For example, export an ARIMA or Prophet model weights to a file to be downloaded or stored on S3. Running forecasts on new data is just too slow and I would like to use Forecast to train models and eventually deploy them somewhere else.

How to detect overfitting with Amazon Forecast. Does SageMaker Debugger work with Amazon Forecast?

Does anyone know the detail about using AWS Forecast? I have the following 2 questions: 1. When I use AWS Forecast via boto3, if my training data is stored in S3, I create the dataset group, create dataset, then import dataset. If my dataset in S3 is updated, and I want to reimport the new data, do I need to repeat the process by creating dataset group, creating dataset, then importing dataset? Or I just need to import the dataset again (assuming the URL to the data in s3 hasn't change?)? 2. When use the DeepAR algorithm with CUSTOM data type, if my dataset has the dimensions column of item_id, will the algorithm recognize that item_id with the same dimensions will have a similar pattern? Or I need to put item_id and dimensions in metadata, so that the deeper algorithm can recognize that? Thanks!

Our customer tried AutoML on Amazon Forecast and they got the result below. Winning autoML algorithm was NPTS. How was NPTS selected? NPTS wQL[0.5]/MAPE 0.6676 RMSE 23.2451 wQL[0.9] 0.4503 wQL[0.1] 0.1988 ARIMA wQL[0.5]/MAPE 0.6184 RMSE 16.5769 wQL[0.9] 0.5333 wQL[0.1] 0.4487 Prophet wQL[0.5]/MAPE 1.4359 RMSE 43.3114 wQL[0.9] 1.2045 wQL[0.1] 0.5908 ETS wQL[0.5]/MAPE 0.6380 RMSE 18.0232 wQL[0.9] 31.9275 wQL[0.1] 31.7771

We get excellent results when we run separate Amazon Forecast analysis for two different data sets (for different cities). However, when we merge the two data sets and add a location as a new dimension in the target data set, the results aren't accurate anymore. What's the reason for this inaccuracy? We added the location as categorical (string) data. Should we add the location as one-hot encoding data?

Hi all, looking for amazon forecast demo for a customer workshop, do we have a guide or someone have a build a quick easy demo ?