Use the Google Cloud console to train an AutoML image classification model.After your dataset is created and data is imported, use theGoogle Cloud console to review the training images and begin modeltraining.

This tutorial has several pages:

1. Set up your project and environment.

2. Create an image classification dataset, and import images.

3. Train an AutoML image classification model.

4. Evaluate and analyze model performance.

5. Deploy a model to an endpoint, and send a prediction.

6. Clean up your project.

Each page assumes that you have already performed the instructions from theprevious pages of the tutorial.

Review imported images

After the dataset import, you are taken to theBrowse tab. You can also accessthis tab by selectingDatasets from the menu. Select theannotation set (set of single-label image annotations) associated with yournew dataset.

Go to the Datasets page

Begin AutoML model training

Choose one of the following options to begin training:

• ChooseTrain new model.

• SelectModels from the menu, and selectCreate.

1. Go to the Models page

2. SelectCreate to open theTrain new model window.

3. SelectSelect Training method, and select thetarget Datasetif they are not automatically selected. Make suretheradio_button_checkedAutoMLradio button is selected, and then chooseCONTINUE.

   

4. (Optional) SelectDefine your model, and enter theModel name. ClickCONTINUE.

   

5. SelectTrain options. Select a model option according to your accuracy and latency needs. Optionally, enable incremental training and clickCONTINUE.

   Incremental training considerations follow:

   • Incremental training can be enabled when there is at least one base modelthat has been trained in this project with the same objective.
   • Incremental training lets you use an existing base model as a starting pointto train a new model rather than training a new model from scratch.
   • Incremental training generally helps training to occur faster and savestraining time.
   • The base model can be trained from a different dataset.

   

6. SelectCompute and pricing. Specify a node-hour budget of8 node hours. SelectStart training.

   Node-hour budget is the maximum time (may vary slightly) that the modelspends training. This value is multiplied by theprice per node hourto calculate to total training cost. More training hours results in a moreaccurate (up to a point) model but results in a higher cost. For developmentpurposes, a low budget is fine but for production it's important to strike abalance between cost and accuracy.

Training takes several hours. An email notification is sent when the model training completes.

What's next

Follow thenext page of this tutorial to check theperformance of your trained AutoML model and explore ways of making it better.

FollowDeploy a model to an endpoint and make a prediction to deploy your trained AutoML model. An image is sent to the model for prediction.