Skip to content

simple-train-cli.md

Latest commit

 

History

History
88 lines (65 loc) · 3.09 KB

simple-train-cli.md

File metadata and controls

88 lines (65 loc) · 3.09 KB

Train an image classification model with AML 2.0 CLI

  1. Remove any previous AML CLI extension installations

    az extension remove -n ml
az extension remove -n azure-cli-ml

  2. Install the latest AML 2.0 CLI, which is in public preview, and then verify installation

    az extension add -n ml
az ml -h

  3. Let's set some defaults for all subsequent "az ml" CLI commands

    az account set --subscription <subcription id>
az configure --defaults workspace=<azureml workspace name> group=<resource group>

  4. For this simple training job with AML 2.0 CLI, we have following project directory structure:

    simple-train-cli
|-- src
|   |-- train.py
|   |-- utils.py
|-- job.yml

    As you can see from above, the project simply contains a job YAML file and some Python training scripts. In general, this a very typical project setup for Azure Arc-enabled ML training. Let's take a look a job YAML file:

    experiment_name: Tutorial-sklearn-mnist
code:
  local_path: ./src
command: python train.py --data-folder {inputs.mnist} --regularization 0.5
environment:
  name: tutorial-env
  version: 1
  path: .
  conda_file: file:./environment.yml
  docker:
    image: mcr.microsoft.com/azureml/intelmpi2018.3-ubuntu16.04:20210301.v1
compute:
  target: azureml:<your compute target name>
  instance_type: <your instance type>
inputs:
  mnist:
    data: azureml:mnist_opendataset:1
    mode: mount

    Note: Instance type is optional parameter. If it's not given, the compute default instance type will be used. For this example to run, you would have created following assets in AML Workspace: compute target named <your compute name>, and file dataset named mnist_opendataset.

  5. Git clone preview Github repo and switch to simple-train-cli directory

    git clone https://github.com/Azure/AML-Kubernetes.git
cd AML-Kubernetes/examples/simple-train-cli

  6. Modify job YAML file to replace amlarc-ml with your own compute target name, and register open dataset MNIST as file dataset and named mnist_opendataset in AML Workspace from public URL: https://azureopendatastorage.blob.core.windows.net/mnist/*.gz

  7. Run the image classification training job

    az ml job create -f job.yml --web

    Creating this job uploads any specified local assets, like the source code directory, validates the YAML file, and submits the run. If needed, the environment is built, then the compute is scaled up and configured for running the job.

  8. Once the job is compute, you can download the outputs:

    az ml job download -n $run_id --outputs

That is it! You have successfully trained an image classification model and download outputs to local directory.

Additional resources