CreatePredictor - Amazon Forecast

Amazon Forecast is no longer available to new customers. Existing customers of Amazon Forecast can continue to use the service as normal. Learn more"

CreatePredictor

Note

This operation creates a legacy predictor that does not include all the predictor functionalities provided by Amazon Forecast. To create a predictor that is compatible with all aspects of Forecast, use CreateAutoPredictor.

Creates an Amazon Forecast predictor.

Important

Amazon Forecast is no longer available to new customers. Existing customers of Amazon Forecast can continue to use the service as normal. Learn more"

In the request, provide a dataset group and either specify an algorithm or let Amazon Forecast choose an algorithm for you using AutoML. If you specify an algorithm, you also can override algorithm-specific hyperparameters.

Amazon Forecast uses the algorithm to train a predictor using the latest version of the datasets in the specified dataset group. You can then generate a forecast using the CreateForecast operation.

To see the evaluation metrics, use the GetAccuracyMetrics operation.

You can specify a featurization configuration to fill and aggregate the data fields in the TARGET_TIME_SERIES dataset to improve model training. For more information, see FeaturizationConfig.

For RELATED_TIME_SERIES datasets, CreatePredictor verifies that the DataFrequency specified when the dataset was created matches the ForecastFrequency. TARGET_TIME_SERIES datasets don't have this restriction. Amazon Forecast also verifies the delimiter and timestamp format. For more information, see Importing Datasets.

By default, predictors are trained and evaluated at the 0.1 (P10), 0.5 (P50), and 0.9 (P90) quantiles. You can choose custom forecast types to train and evaluate your predictor by setting the ForecastTypes.

AutoML

If you want Amazon Forecast to evaluate each algorithm and choose the one that minimizes the objective function, set PerformAutoML to true. The objective function is defined as the mean of the weighted losses over the forecast types. By default, these are the p10, p50, and p90 quantile losses. For more information, see EvaluationResult.

When AutoML is enabled, the following properties are disallowed:

  • AlgorithmArn

  • HPOConfig

  • PerformHPO

  • TrainingParameters

To get a list of all of your predictors, use the ListPredictors operation.

Note

Before you can use the predictor to create a forecast, the Status of the predictor must be ACTIVE, signifying that training has completed. To get the status, use the DescribePredictor operation.

Request Syntax

{ "AlgorithmArn": "string", "AutoMLOverrideStrategy": "string", "EncryptionConfig": { "KMSKeyArn": "string", "RoleArn": "string" }, "EvaluationParameters": { "BackTestWindowOffset": number, "NumberOfBacktestWindows": number }, "FeaturizationConfig": { "Featurizations": [ { "AttributeName": "string", "FeaturizationPipeline": [ { "FeaturizationMethodName": "string", "FeaturizationMethodParameters": { "string" : "string" } } ] } ], "ForecastDimensions": [ "string" ], "ForecastFrequency": "string" }, "ForecastHorizon": number, "ForecastTypes": [ "string" ], "HPOConfig": { "ParameterRanges": { "CategoricalParameterRanges": [ { "Name": "string", "Values": [ "string" ] } ], "ContinuousParameterRanges": [ { "MaxValue": number, "MinValue": number, "Name": "string", "ScalingType": "string" } ], "IntegerParameterRanges": [ { "MaxValue": number, "MinValue": number, "Name": "string", "ScalingType": "string" } ] } }, "InputDataConfig": { "DatasetGroupArn": "string", "SupplementaryFeatures": [ { "Name": "string", "Value": "string" } ] }, "OptimizationMetric": "string", "PerformAutoML": boolean, "PerformHPO": boolean, "PredictorName": "string", "Tags": [ { "Key": "string", "Value": "string" } ], "TrainingParameters": { "string" : "string" } }

Request Parameters

The request accepts the following data in JSON format.

AlgorithmArn

The Amazon Resource Name (ARN) of the algorithm to use for model training. Required if PerformAutoML is not set to true.

Supported algorithms:
  • arn:aws:forecast:::algorithm/ARIMA

  • arn:aws:forecast:::algorithm/CNN-QR

  • arn:aws:forecast:::algorithm/Deep_AR_Plus

  • arn:aws:forecast:::algorithm/ETS

  • arn:aws:forecast:::algorithm/NPTS

  • arn:aws:forecast:::algorithm/Prophet

Type: String

Length Constraints: Maximum length of 256.

Pattern: arn:([a-z\d-]+):forecast:.*:.*:.+

Required: No

AutoMLOverrideStrategy
Note

The LatencyOptimized AutoML override strategy is only available in private beta. Contact AWS Support or your account manager to learn more about access privileges.

Used to overide the default AutoML strategy, which is to optimize predictor accuracy. To apply an AutoML strategy that minimizes training time, use LatencyOptimized.

This parameter is only valid for predictors trained using AutoML.

Type: String

Valid Values: LatencyOptimized | AccuracyOptimized

Required: No

EncryptionConfig

An AWS Key Management Service (KMS) key and the AWS Identity and Access Management (IAM) role that Amazon Forecast can assume to access the key.

Type: EncryptionConfig object

Required: No

EvaluationParameters

Used to override the default evaluation parameters of the specified algorithm. Amazon Forecast evaluates a predictor by splitting a dataset into training data and testing data. The evaluation parameters define how to perform the split and the number of iterations.

Type: EvaluationParameters object

Required: No

FeaturizationConfig

The featurization configuration.

Type: FeaturizationConfig object

Required: Yes

ForecastHorizon

Specifies the number of time-steps that the model is trained to predict. The forecast horizon is also called the prediction length.

For example, if you configure a dataset for daily data collection (using the DataFrequency parameter of the CreateDataset operation) and set the forecast horizon to 10, the model returns predictions for 10 days.

The maximum forecast horizon is the lesser of 500 time-steps or 1/3 of the TARGET_TIME_SERIES dataset length.

Type: Integer

Required: Yes

ForecastTypes

Specifies the forecast types used to train a predictor. You can specify up to five forecast types. Forecast types can be quantiles from 0.01 to 0.99, by increments of 0.01 or higher. You can also specify the mean forecast with mean.

The default value is ["0.10", "0.50", "0.9"].

Type: Array of strings

Array Members: Minimum number of 1 item. Maximum number of 20 items.

Length Constraints: Minimum length of 2. Maximum length of 4.

Pattern: (^0?\.\d\d?$|^mean$)

Required: No

HPOConfig

Provides hyperparameter override values for the algorithm. If you don't provide this parameter, Amazon Forecast uses default values. The individual algorithms specify which hyperparameters support hyperparameter optimization (HPO). For more information, see Amazon Forecast Algorithms.

If you included the HPOConfig object, you must set PerformHPO to true.

Type: HyperParameterTuningJobConfig object

Required: No

InputDataConfig

Describes the dataset group that contains the data to use to train the predictor.

Type: InputDataConfig object

Required: Yes

OptimizationMetric

The accuracy metric used to optimize the predictor. The default value is AverageWeightedQuantileLoss.

Type: String

Valid Values: WAPE | RMSE | AverageWeightedQuantileLoss | MASE | MAPE

Required: No

PerformAutoML

Whether to perform AutoML. When Amazon Forecast performs AutoML, it evaluates the algorithms it provides and chooses the best algorithm and configuration for your training dataset.

The default value is false. In this case, you are required to specify an algorithm.

Set PerformAutoML to true to have Amazon Forecast perform AutoML. This is a good option if you aren't sure which algorithm is suitable for your training data. In this case, PerformHPO must be false.

Type: Boolean

Required: No

PerformHPO

Whether to perform hyperparameter optimization (HPO). HPO finds optimal hyperparameter values for your training data. The process of performing HPO is known as running a hyperparameter tuning job.

The default value is false. In this case, Amazon Forecast uses default hyperparameter values from the chosen algorithm.

To override the default values, set PerformHPO to true and, optionally, supply the HyperParameterTuningJobConfig object. The tuning job specifies a metric to optimize, which hyperparameters participate in tuning, and the valid range for each tunable hyperparameter. In this case, you are required to specify an algorithm and PerformAutoML must be false.

The following algorithms support HPO:

  • DeepAR+

  • CNN-QR

Type: Boolean

Required: No

PredictorName

A name for the predictor.

Type: String

Length Constraints: Minimum length of 1. Maximum length of 63.

Pattern: ^[a-zA-Z][a-zA-Z0-9_]*

Required: Yes

Tags

The optional metadata that you apply to the predictor to help you categorize and organize them. Each tag consists of a key and an optional value, both of which you define.

The following basic restrictions apply to tags:

  • Maximum number of tags per resource - 50.

  • For each resource, each tag key must be unique, and each tag key can have only one value.

  • Maximum key length - 128 Unicode characters in UTF-8.

  • Maximum value length - 256 Unicode characters in UTF-8.

  • If your tagging schema is used across multiple services and resources, remember that other services may have restrictions on allowed characters. Generally allowed characters are: letters, numbers, and spaces representable in UTF-8, and the following characters: + - = . _ : / @.

  • Tag keys and values are case sensitive.

  • Do not use aws:, AWS:, or any upper or lowercase combination of such as a prefix for keys as it is reserved for AWS use. You cannot edit or delete tag keys with this prefix. Values can have this prefix. If a tag value has aws as its prefix but the key does not, then Forecast considers it to be a user tag and will count against the limit of 50 tags. Tags with only the key prefix of aws do not count against your tags per resource limit.

Type: Array of Tag objects

Array Members: Minimum number of 0 items. Maximum number of 200 items.

Required: No

TrainingParameters

The hyperparameters to override for model training. The hyperparameters that you can override are listed in the individual algorithms. For the list of supported algorithms, see Amazon Forecast Algorithms.

Type: String to string map

Map Entries: Minimum number of 0 items. Maximum number of 100 items.

Key Length Constraints: Maximum length of 256.

Key Pattern: ^[a-zA-Z0-9\-\_\.\/\[\]\,\\]+$

Value Length Constraints: Maximum length of 256.

Value Pattern: ^[a-zA-Z0-9\-\_\.\/\[\]\,\"\\\s]+$

Required: No

Response Syntax

{ "PredictorArn": "string" }

Response Elements

If the action is successful, the service sends back an HTTP 200 response.

The following data is returned in JSON format by the service.

PredictorArn

The Amazon Resource Name (ARN) of the predictor.

Type: String

Length Constraints: Maximum length of 256.

Pattern: arn:([a-z\d-]+):forecast:.*:.*:.+

Errors

InvalidInputException

We can't process the request because it includes an invalid value or a value that exceeds the valid range.

HTTP Status Code: 400

LimitExceededException

The limit on the number of resources per account has been exceeded.

HTTP Status Code: 400

ResourceAlreadyExistsException

There is already a resource with this name. Try again with a different name.

HTTP Status Code: 400

ResourceInUseException

The specified resource is in use.

HTTP Status Code: 400

ResourceNotFoundException

We can't find a resource with that Amazon Resource Name (ARN). Check the ARN and try again.

HTTP Status Code: 400

See Also

For more information about using this API in one of the language-specific AWS SDKs, see the following: