Google AutoML Connector: Difference between revisions

From Okapi Framework
Jump to navigation Jump to search
Line 22: Line 22:
==Limitations==
==Limitations==


TODO
* AutoML Translation is currently in beta, so it isn't recommended for production use.
 
* There are costs associated with training a model. See the [https://cloud.google.com/translate/automl/pricing pricing] page for more information.
* The connector requires you have a trained model for each language pair you intend to use.
* Unlike the [[Google MT v2 Connector]], API keys cannot be used with this connector due to a [https://groups.google.com/forum/#!topic/google-translate-api/e3sOLcC1qZI limitation] of the AutoML Translation API.
* The connector does not retry API calls when they fail.
[[Category:Connectors]]
[[Category:Connectors]]

Revision as of 20:03, 26 October 2018

Overview

This is a connector for Google AutoML Translation, a service that allows users to train custom models and use them for translation. If you don't require a custom model, consider using the Google MT v2 Connector instead.

Warning: The end-user is solely responsible for the costs of using this service. The developers of the Okapi Framework do their best to provide bug-free components and they are not liable, in any way, for any cost incurred by the end-user even when caused by defective code.

Using the Connector

In order to use the connector, you'll need to create a Google service account and download a private key JSON file for the account.

You must train a model for each language pair you intend to use with the connector. Google has a tutorial on how to train a model through the AutoML Translation UI. You can also use the AutoML Translation API to train models programatically.

Parameters

Credential file path: Absolute path on your filesystem to a JSON file containing private key information for a Google service account. This is the same JSON file described in the previous section.

Model JSON map: A JSON string that maps a language pair to the resource name of the model that will process translations for that language pair. Model resource names are typically of the form projects/{projectId}/locations/{computeRegion}/models/{modelId}.

  • Example: {"en-US/ja-JP": "projects/my-project/locations/us-central1/models/ABC123", "en-US/de-DE": "projects/my-project/locations/us-central1/models/DEF456"}

Limitations

  • AutoML Translation is currently in beta, so it isn't recommended for production use.
  • There are costs associated with training a model. See the pricing page for more information.
  • The connector requires you have a trained model for each language pair you intend to use.
  • Unlike the Google MT v2 Connector, API keys cannot be used with this connector due to a limitation of the AutoML Translation API.
  • The connector does not retry API calls when they fail.