Class FormTrainingClient

Example usage:

const targetClient = new FormTrainingClient(targetEndpoint, new AzureKeyCredential(targetApiKey));
const authorization = await targetClient.getCopyAuthorization(targetResourceId, targetResourceRegion);

const sourceClient = new FormTrainingClient(endpoint, new AzureKeyCredential(apiKey));
const poller = await sourceClient.beginCopyModel(sourceModelId, authorization, {
  onProgress: (state) => {
    console.log(`Copy model status: ${state.status}`);
  }
});
const result = await poller.pollUntilDone();

summary

Copies custom model to target resource

The composed model will contain copies of all of its input submodels, and it will choose (using a machine learning algorithm) the most appropriate of its input models to use during form recognition.

This method returns a long-running operation poller that allows you to wait indefinitely until the operation is completed.

Notes:

• Only labeled models can be composed. Attempting to compose an unlabeled model will result in an error.
• the onProgress callback will not be invoked if the operation completes in the first request, and attempting to cancel a completed copy will result in an error being thrown.
• evena when the training operation fails, a model is still created in the Azure Form Recognizer resource.

Example usage:

const modelIds = ["<model ID 1>", "<model ID 2>", "<model ID 3>"];
const trainingClient = new FormTrainingClient(endpoint, new AzureKeyCredential(apiKey));

const poller = await trainingClient.beginCreateComposedModel(modelIds, {
  modelName: "<optional name for the composed model>",
  onProgress: (state) => { console.log("training status: "); console.log(state); }
});
const composedModel = await poller.pollUntilDone();

summary

Combines pre-existing models with labels into a single composed model.

If the useTrainingLabels parameter is set to true, then the operation will search for label files in addition to the training documents, and it will create a labeled model with the field names specified by the labels. Otherwise, it will create an unlabeled model automatically that returns generated field names for the items it determines are fields within the document structure.

This method returns a long-running operation poller that allows you to wait indefinitely until the operation is completed.

Notes:

• The onProgress callback will not be invoked if the operation completes in the first request, and attempting to cancel a completed copy will result in an error being thrown.
• Even when the training operation fails, a model is still created in the Azure Form Recognizer resource.

Example usage:

const trainingFilesUrl = "<url to the blob container storing training documents>";
const trainingClient = new FormTrainingClient(endpoint, new AzureKeyCredential(apiKey));

const poller = await trainingClient.beginTraining(trainingFilesUrl, false, {
  onProgress: (state) => { console.log("training status: "); console.log(state); }
});
const model = await poller.pollUntilDone();

summary

Creates and trains a custom form model.

This method should be called on a client that is authenticated using the target resource (where the model will be copied to) credentials, and the output can be passed as the target parameter to the beginCopyModel method of a source client.

The required resourceId and resourceRegion are properties of an Azure Form Recognizer resource and their values can be found in the Azure Portal.

.byPage() returns an async iterable iterator to list the blobs in pages.

Example using for await syntax:

const client = new FormTrainingClient(endpoint, new AzureKeyCredential(apiKey));
const result = client.listCustomModels();
let i = 1;
for await (const model of result) {
  console.log(`model ${i++}:`);
  console.log(model);
}

Example using iter.next():

let i = 1;
let iter = client.listCustomModels();
let modelItem = await iter.next();
while (!modelItem.done) {
  console.log(`model ${i++}: ${modelItem.value}`);
  modelItem = await iter.next();
}

Example using byPage():

 let i = 1;
 for await (const response of client.listCustomModels().byPage()) {
   for (const modelInfo of response.modelList!) {
     console.log(`model ${i++}: ${modelInfo.modelId}`);
   }
 }