GCP Auto ML - ghdrako/doc_snipets GitHub Wiki

AutoML comes into play when pretrained models are not fit for purpose. As an example, the Vision API can recognize a sofa, but what if we want to recognize a particular sofa that our company produces? The Vision API cannot do that for us.In such a case, we need to use AutoML or train our own model.

AutoML takes datasets from you, trains and deploys the model, and then serves it through the REST API.

There are five services available that allow you to train your custom model:

  • AutoML Vision: This classifies your images according to your own defined labels.
  • AutoML Translation: This performs translation queries, returning results specific to your domain.
  • AutoML Natural Language: This classifies English-language content into a custom set of categories.
  • AutoML Tables: This turns structured data into predictive insights.
  • AutoML Video Intelligence: This allows you to classify segments of video.

Example use AutoML, using the example of the Vision API, to recognize a table that your company produces. short, what you would do is the following:

  1. Take multiple photos of your table.
  2. Upload it to Cloud Storage.
  3. Create a CSV file with a label for your photos.
  4. Provide the CSV file to AutoML to train the model Once the model has been trained, you can access the model through the REST API, as you would with any other pretrained model.