Note: This repository is still a work in progress

The custom models action manages custom inference models and their associated deployments in DataRobot via GitHub CI/CD workflows. These workflows allow you to create or delete models and deployments and modify settings. Metadata defined in YAML files enables the custom model action's control over models and deployments. Most YAML files for this action can reside in any folder within your custom model's repository. The YAML is searched, collected, and tested against a schema to determine if it contains the entities used in these workflows.

The following feature flags must be enabled in DataRobot:

• ENABLE_MLOPS
• ENABLE_CUSTOM_INFERENCE_MODEL
• ENABLE_CUSTOM_MODEL_GITHUB_CI_CD

(Please contact DataRobot support for more information)

This quickstart example uses a Python Scikit-Learn model template from the datarobot-user-model repository. To set up a custom models action that will create a custom inference model and deployment in DataRobot from a custom model repository in GitHub, take the following steps:

1. In the .github/workflows directory of your custom model repository, create a YAML file (with any filename) containing the following:

   name: Workflow CI/CD
   
   on:
     pull_request:
       branches: [ master ]
     push:
       branches: [ master ]
   
     # Allows you to run this workflow manually from the Actions tab
     workflow_dispatch:
   
   jobs:
     datarobot-custom-models:
       # Run this job on any action of a PR, but skip the job upon merging to the main branch. This
       # will be taken care of by the push event.
       if: ${{ github.event.pull_request.merged != true }}
   
       runs-on: ubuntu-latest
   
       steps:
         - uses: actions/checkout@v3
           with:
             fetch-depth: 0
   
         - name: DataRobot Custom Models Step
           id: datarobot-custom-models-step
           uses: datarobot-oss/[email protected]
           with:
             api-token: ${{ secrets.DATAROBOT_API_TOKEN }}
             webserver: https://app.datarobot.com/
             branch: master
             allow-model-deletion: true
             allow-deployment-deletion: true

   Configure the following fields:

   • branches: Provide the name of your repository's main branch (usually either master or main) for pull_request and push. If you created your repository in GitHub, you likely need to update these fields to main. While master and main are the most common branch names, you can target any branch; for example, you could run the workflow on a release branch or a test branch.

   • api-token: Provide a value for the ${{ secrets.DATAROBOT_API_TOKEN }} variable by creating an encrypted secret for GitHub Actions containing your DataRobot API key. Alternatively, you can set the token string directly to this field; however, this method is highly discouraged because your API key is extremely sensitive data. If you use this method, anyone with access to your repository can access your API key.

   • webserver: Provide your DataRobot webserver value here if it isn't the default DataRobot US server (https://app.datarobot.com/).

   • branch: Provide the name of your repository's main branch (usually either master or main). If you created your repository in GitHub, you likely need to update this field to main. While master and main are the most common branch names, you can target any branch; for example, you could run the workflow on a release branch or a test branch.

2. Commit the workflow YAML file and push it to the remote. After you complete this step, any push to the remote (or merged pull request) triggers the action.

3. In the folder for your DataRobot custom model, add a model definition YAML file (e.g., model.yaml) containing the following YAML and update the field values according to your model's characteristics:

   user_provided_model_id: user/model-unique-id-1
   target_type: Regression
   settings:
     name: My Awesome GitHub Model 1 [GitHub CI/CD]
     target_name: Grade 2014
   
   version:
     # Make sure this is the environment ID is in your system.
     # This one is the '[DataRobot] Python 3 Scikit-Learn Drop-In' environment
     model_environment_id: 5e8c889607389fe0f466c72d

   Configure the following fields:

   • user_provided_model_id: Provide any descriptive and unique string value. A good practice could be this kind of pattern: <user>/<model-unique-id>. Please note that, by default, this ID will reside in a unique namespace, which is the GitHub repository ID. Alternatively, the namespace can be configured as an input argument to the custom models action.
   • target_type: Provide the correct target type for your custom model.
   • target_name: Provide the correct target name for your custom model.
   • model_environment_id: Provide the DataRobot execution environment required for your custom model. You can find these environments in the DataRobot application under Model Registry > Custom Model Workshop > Environments.

4. In any directory in your repository, add a deployment definition YAML file (with any filename) containing the following YAML:

   user_provided_deployment_id: user/my-awesome-deployment-id
   user_provided_model_id: user/model-unique-id-1

   Configure the following fields:

   • user_provided_deployment_id: Provide any descriptive and unique string value. A good practice could be this kind of pattern: <user>/<deployment-unique-id>. Please note that, by default, this ID will reside in a unique namespace, which is the GitHub repository ID. Alternatively, the namespace can be configured as an input argument to the custom models action.
   • user_provided_model_id: Provide the exact user_provided_model_id you set in the model definition YAML file.

5. Commit these changes and push to the remote:

   • Navigate to your custom model repository in GitHub and click the Actions tab. You'll notice that the action is being executed.

   • Navigate to the DataRobot application. You'll notice that a new custom model was created along with an associated deployment. This action can take a few minutes.

Note: Creating two commits (or merging two pull requests) in quick succession can result in a ResourceNotFoundError. For example, you add a model definition with a training dataset, make a commit, and push to the remote. Then, you immediately delete the model definition, make a commit, and push to the remote. The training data upload action may begin after model deletion, resulting in an error. To avoid this scenario, wait for an action's execution to complete before pushing new commits or merging new pull requests to the remote repository.

After your workflow creates a model and a deployment in DataRobot, you can access the commit information from the model's version info and the deployment's overview:

1. In the Model Registry, click Custom Model Workshop.

2. On the Models tab, click a GitHub-sourced model from the list and then click the Versions tab.

3. Under Manage Versions, click the version you want to view the commit for.

4. Under Version Info, find the Git Commit Reference and then click the commit hash (or commit ID) to open the commit in GitHub that created the current version.

1. In the Model Registry, click Model Packages.

2. On the Model Packages tab, click a GitHub-sourced model package from the list.

3. Under Package Info, review the model information provided by your workflow, find the Git Commit Reference, and then click the commit hash (or commit ID) to open the commit that created the current model package.

1. In the Deployments inventory, click a GitHub-sourced deployment from the list.

2. On the deployment's Overview tab, review the model and deployment information provided by your workflow.

3. In the Content group box, find the Git Commit Reference and click the commit hash (or commit ID) to open the commit that created the deployment.

Datasets referenced in custom models action YAML files are expected to exist in the DataRobot AI catalog before configuring the action in GitHub. You should upload these datasets to the DataRobot AI catalog (via the UI or any other client) prior to configuring the GitHub action.

Environments referenced in custom models action YAML files are expected to exist in DataRobot before configuring the action in GitHub. You should validate the existence of the required drop-in environments prior to configuring the GitHub action. In addition, you can install new drop-in environments. For more information, see the Custom model environments documentation.

This GitHub action is implemented as a Python program, called with specific arguments provided in the GitHub workflow.

This action requires the following input arguments:

The action supports the following optional input arguments:

A namespace is a unique string that can be provided as an input argument to the action. The purpose is to guarantee that the custom model action handles models and deployments that only exist in the configured namespace. Any other models and deployments that are not in the configured namespace will remain untouched.

If not provided, the GitHub repository ID will be used as the namespace (see: GITHUB_REPOSITORY_ID in this link). It means that the custom models action will process models and deployments that were created from this repository only.

If, for instance, users would like to work with the same model and deployment definitions from different branches and still make sure that different entities will be created in DataRobot, they can simply configure a different namespace to the custom models action in the GitHub workflow.

Please note, that a change of the namespace input argument to the custom models action in a GitHub workflow, will result in new models and deployments in DataRobot. The existing ones will remain in DataRobot without a control from the GitHub action.

The GitHub action supports the following output arguments, which can later be used by follow-up steps in the same GitHub job (refer to the workflow example below):

The GitHub action requires the model's metadata in a YAML file. The model's full schema is defined by MODEL_SCHEMA, which can be found in here.

A model metadata YAML file may contain the schema of a single model's definition (as specified above) or the schema of multiple models' definitions.

The multiple models' schema is defined by MULTI_MODELS_SCHEMA, which can be found in here.

The single model's definition YAML file must be located inside the model's root directory. The multiple models' definition YAML file can be located anywhere in the repository.

For examples, please refer to the model definition examples section below.

At the top level, there are attributes you cannot change after a model is created:

• settings: Changes to the fields under this section result in changes to the model's settings without creating a new version.
• version: Changes to the fields under this section result in a new version.
• test: Contains attributes that control the custom inference model testing. If omitted, a test will not be executed.

The user is required to provide the deployment's metadata in a YAML file. The deployment's full schema is defined by DEPLOYMENT_SCHEMA, which can be found in here.

A deployment metadata YAML file may contain the schema of a single deployment's definition (as specified above) or the schema of multiple deployments' definitions.

The multiple deployments' schema is defined by MULTI_DEPLOYMENTS_SCHEMA, which can be found in here.

The deployment definition YAML file (single or multiple) can be located anywhere in the repository.

For examples, please refer to the deployment definition examples section below.

At the top level, some attributes shouldn't be changed once the deployment is created:

• user_provided_model_id: An exception that associates a model definition to the given deployment. A change in this field triggers model replacement or challenger creation, depending on the deployment's configuration.
• settings: Changes to the fields in this section will result in changes to the deployment's settings.

A GitHub workflow is a configurable process of one or more jobs. It is defined in a YAML file located under .github/workflows in the repository. For more information, refer to Using Workflows in GitHub.

To use the Custom Models Action, the following YAML should be included in the GitHub workflow definition:

1. The action should run on two events: pull_request and push. Therefore, the following should be defined:

   on:
     pull_request:
       branches: [ master ]
     push:
       branches: [ master ]

2. Use the DataRobot custom models action in a workflow job as follows:

   jobs:
     datarobot-custom-models:
       # Run this job on any action of a PR, but skip the job upon merging to the main branch.
       # This will be taken care of by the push event.
       if: ${{ github.event.pull_request.merged != true }}
   
       runs-on: ubuntu-latest
   
       steps:
         - uses: actions/checkout@v3
           with:
             fetch-depth: 0
   
         - name: DataRobot Custom Models Step
           id: datarobot-custom-models-step
           uses: datarobot-oss/[email protected]
           with:
             api-token: ${{ secrets.DATAROBOT_API_TOKEN }}
             webserver: ${{ secrets.DATAROBOT_WEBSERVER }}
             branch: master
             allow-model-deletion: true
             allow-deployment-deletion: true

For a complete example, refer to the workflow example below.

The top-level files and directories include the following:

• action.yaml: The YAML file containing the definition of the DataRobot custom models GitHub action.
• .github: The directory containing a GitHub workflow that executes the following jobs:
  • Linter
  • Code style checks
  • Unit-tests
  • Functional tests
• deps: The directory containing Python requirements for the development of this repository.
• src: The directory containing the source code that implements the related GitHub actions.
• tests: The directory containing the code source and resources to test the implementation. It includes the following:
  • datasets: The directory containing datasets used by the tests.
  • deployments: The directory containing a deployment definition that is used by tests.
  • functional: The directory containing the functional tests' source code.
  • models: The directory containing the model definition and source code used by the tests.
  • unit: The directory containing the unit-test source code.

Functional tests are written on top of the main entry point, simulating the GitHub actions execution. To enable communication with DataRobot, you must set two important environment variables:

• DATAROBOT_WEBSERVER: The DataRobot web server URL, which can be accessed publicly.
• DATAROBOT_API_TOKEN: The API key used to validate credentials with the DataRobot system.

In the current repository, there is a definition of one model under tests/models/py3_sklearn/ and one deployment under tests/deployments used by the functional test.

Changes in this repository should be submitted as pull requests. When a pull request is created, the associated GitHub workflow is triggered, and the following jobs are executed sequentially:

• Linter
• Code style checks
• Unit-tests.
• Functional test(s).

Note: To enable the full execution of the functional test, the two related variables (DATAROBOT_WEBSERVER and DATAROBOT_API_TOKEN) were set in the Secrets section of the GitHub repository. These are read by the workflow, which sets the proper environment variables.

user_provided_model_id: user/any-model-unique-id-1
target_type: Regression
settings:
  name: My Awsome GitHub Model 1 [GitHub CI/CD]
  target_name: Grade 2014

version:
  # Make sure this is the environment ID is in your system.
  # This one is the '[DataRobot] Python 3 Scikit-Learn Drop-In' environment
  model_environment_id: 5e8c889607389fe0f466c72d

user_provided_model_id: user/any-model-unique-id-1
target_type: Binary
settings:
  name: My Awsome GitHub Model 1 [GitHub CI/CD]
  description: My awesome model
  target_name: Grade 2014
  holdout_dataset_id: 627790ca5621558b55c78d78
  language: Python
  negative_class_label: '0'
  positive_class_label: '1'
  training_dataset_id: 627790ba56215587b3021632

version:
  # Make sure this is the environment ID is in your system.
  # This one is the '[DataRobot] Python 3 Scikit-Learn Drop-In' environment
  model_environment_id: 5e8c889607389fe0f466c72d
  exclude_glob_pattern:
    - README.md
    - out/
  include_glob_pattern:
    - ./
  memory: 256Mi
  replicas: 2
  egress_network_policy: NONE
  model_replacement_reason: DATA_DRIFT

test:
  memory: 256Mi
  skip: false
  test_data_id: 62779143562155aa34a3d65b
  checks:
    null_value_imputation:
      block_deployment_if_fails: true
      enabled: true
    performance:
      block_deployment_if_fails: false
      enabled: true
      max_execution_time: 100
      maximum_response_time: 50
      number_of_parallel_users: 3
    prediction_verification:
      block_deployment_if_fails: false
      enabled: true
      match_threshold: 0.9
      output_dataset_id: 627791f5562155d63f367b05
      passing_match_rate: 85
      predictions_column: Grade 2014
    side_effects:
      block_deployment_if_fails: true
      enabled: true
    stability:
      block_deployment_if_fails: true
      enabled: true
      maximum_payload_size: 1000
      minimum_payload_size: 100
      number_of_parallel_users: 1
      passing_rate: 95
      total_prediction_requests: 50

datarobot_models:
  - model_path: ./models/model_1
    model_metadata:
      user_provided_model_id: user/any-model-unique-id-1
      target_type: Regression
      settings:
        name: My Awsome GitHub Model 1 [GitHub CI/CD]
        target_name: Grade 2014

      version:
        # Make sure this is the environment ID is in your system.
        # This one is the '[DataRobot] Python 3 Scikit-Learn Drop-In' environment
        model_environment_id: 5e8c889607389fe0f466c72d

  - model_path: ./models/model_2
    model_metadata:
      user_provided_model_id: user/any-model-unique-string-2
      target_type: Regression
      settings:
        name: My Awsome GitHub Model 2 [GitHub CI/CD]
        target_name: Grade 2014

      version:
        # Make sure this is the environment ID is in your system.
        # This one is the '[DataRobot] Python 3 Scikit-Learn Drop-In' environment
        model_environment_id: 5e8c889607389fe0f466c72d

user_provided_deployment_id: user/my-awesome-deployment-id
user_provided_model_id: user/any-model-unique-id-1

user_provided_deployment_id: user/my-awesome-deployment-id
user_provided_model_id: user/any-model-unique-string-2
prediction_environment_name: "https://eks-test.orm.company.com"
settings:
  label: "My Awesome Deployment (model-2)"
  description: "This is a more detailed description."
  importance: LOW  # NOTE: a higher importance value than "LOW" will trigger a review process
                   # for any operation, such as 'create', 'update', 'delete', etc. So, practically
                   # the user will need to wait for approval from a reviewer in order to be able
                   # to apply new changes and merge them to the main branch.
  association:
    association_id_column: id
    required_in_pred_request: true
    actual_values_column: Animal
    actuals_dataset_id: 6d8c889607389fe0f466c72e
  enable_target_drift: true
  enable_feature_drift: true
  enable_predictions_collection: true
  enable_challenger_models: true
  segment_analysis:
    enabled: true
    # NOTE: the 'segment_analysis' may contain an 'attributes' section, where users can specify
    # attributes that are categorical features in the associated model.
    # Be aware that if you enabled segment analysis, without specifying attribute, you can still
    # access various statistics by segment of built-in attributes in DataRobot.
    #
    # attributes:
    # - <categorical-attr-1>
    # - <categorical-attr-2>

- user_provided_deployment_id: user/any-deployment-unique-id-1
  user_provided_model_id: user/any-model-unique-id-1

- user_provided_deployment_id: user/any-deployment-unique-id-2
  user_provided_model_id: user/any-model-unique-string-2

- user_provided_deployment_id: user/any-deployment-unique-id-3
  user_provided_model_id: user/any-model-unique-id-3

This is an example of a GitHub workflow definition. The YAML file should be located at the following location: .github/workflows/workflow.yaml.

The YAML file should contain the following:

name: Workflow CI/CD

on:
  pull_request:
    branches: [ master ]
  push:
    branches: [ master ]

  # Allows you to run this workflow manually from the Actions tab
  workflow_dispatch:

jobs:
  datarobot-custom-models:
    # Run this job on any action of a PR, but skip the job upon merging to the main branch. This
    # will be taken care of by the push event.
    if: ${{ github.event.pull_request.merged != true }}

    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v3
        with:
          fetch-depth: 0

      - name: DataRobot Custom Models Step
        id: datarobot-custom-models-step
        uses: datarobot-oss/[email protected]
        with:
          api-token: ${{ secrets.DATAROBOT_API_TOKEN }}
          webserver: ${{ secrets.DATAROBOT_WEBSERVER }}
          branch: master
          allow-model-deletion: true
          allow-deployment-deletion: true

      - name: DataRobot Custom Models Action Metrics
        run: |
          echo "Total affected models: ${{ steps.datarobot-custom-models-step.outputs.models--total-affected }}"
          echo "Total created models: ${{ steps.datarobot-custom-models-step.outputs.models--total-created }}"
          echo "Total deleted models: ${{ steps.datarobot-custom-models-step.outputs.models--total-deleted }}"
          echo "Total models whose settings were updated: ${{ steps.datarobot-custom-models-step.outputs.models--total-updated-settings }}"
          echo "Total created model versions: ${{ steps.datarobot-custom-models-step.outputs.models--total-created-versions }}"

          echo "Total affected deployments: ${{ steps.datarobot-custom-models-step.outputs.deployments--total-affected }}"
          echo "Total created deployments: ${{ steps.datarobot-custom-models-step.outputs.deployments--total-created }}"
          echo "Total deleted deployments: ${{ steps.datarobot-custom-models-step.outputs.deployments--total-deleted }}"
          echo "Total deployments whose settings were updated: ${{ steps.datarobot-custom-models-step.outputs.deployments--total-updated-settings }}"

          echo "Message: ${{ steps.datarobot-custom-models-step.outputs.message }}"

Custom Models GitHub Action is Copyright 2022 DataRobot, Inc. All rights reserved. Licensed under a Modified 3-Clause BSD License (the "License"). See the LICENSE file. You may not use this software except in compliance with the License.

Software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT ANY EXPRESS OR IMPLIED WARRANTIES OF ANY KIND AND WITHOUT ANY LICENSE TO ANY PATENTS OR TRADEMARKS. See the LICENSE file for the specific language governing permissions and limitations under the License.