Model Training

Once you’ve created a dataset, training your model on the TensorCI training cluster is easy:

$ tensorci train

Running the train command deploys the latest commit of your project’s git repo to the TensorCI training cluster. Once deployed, the following steps are performed, in order:

  1. All records in your project’s dataset are fetched from its TensorCI datatable.
  2. If specified, the prepro_data function specified in .tensorci.yml is called if the dataset has changed.
  3. The train function specified in .tensorci.yml is called.
  4. If specified, the test function specified in .tensorci.yml is called.
  5. The trained model specified in .tensorci.yml is uploaded to TensorCI model storage.

Watch Training Logs

Once a training deploy has succeeded, watching the real-time logs for steps 2-4 above can be initiated with the following command:

$ tensorci logs --follow

If the --follow flag is not provided, a dump of all the logs up to this point for this deploy are returned.

Fetch a Trained Model

Once training has finished and the trained model has been uploaded to TensorCI model storage, that model can be pulled locally by running the following command:

$ tensorci get model

This saves the model to the path specified in the .tensorci.yml config file, unless the --output argument is provided to specify another path.