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:
- All records in your project’s dataset are fetched from its TensorCI datatable.
- If specified, the
prepro_datafunction specified in.tensorci.ymlis called if the dataset has changed. - The
trainfunction specified in.tensorci.ymlis called. - If specified, the
testfunction specified in.tensorci.ymlis called. - The trained model specified in
.tensorci.ymlis 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.