trusted-ai-pipeline.yaml

# Copyright 2021 The MLX Contributors
#
# SPDX-License-Identifier: Apache-2.0
apiVersion: tekton.dev/v1beta1
kind: PipelineRun
metadata:
  name: launch-trusted-ai-pipeline
  annotations:
    tekton.dev/output_artifacts: '{"adversarial-robustness-evaluation": [{"key": "artifacts/$PIPELINERUN/adversarial-robustness-evaluation/metric_path.tgz",
      "name": "adversarial-robustness-evaluation-metric_path", "path": "/tmp/outputs/metric_path/data"},
      {"key": "artifacts/$PIPELINERUN/adversarial-robustness-evaluation/robust_status.tgz",
      "name": "adversarial-robustness-evaluation-robust_status", "path": "/tmp/outputs/robust_status/data"}],
      "model-fairness-check": [{"key": "artifacts/$PIPELINERUN/model-fairness-check/metric_path.tgz",
      "name": "model-fairness-check-metric_path", "path": "/tmp/outputs/metric_path/data"}]}'
    tekton.dev/input_artifacts: '{}'
    tekton.dev/artifact_bucket: mlpipeline
    tekton.dev/artifact_endpoint: minio-service.kubeflow:9000
    tekton.dev/artifact_endpoint_scheme: http://
    tekton.dev/artifact_items: '{"adversarial-robustness-evaluation": [["metric_path",
      "$(results.metric-path.path)"], ["robust_status", "$(results.robust-status.path)"]],
      "model-fairness-check": [["metric_path", "$(results.metric-path.path)"]], "trust-ai-train-step":
      []}'
    sidecar.istio.io/inject: "false"
    pipelines.kubeflow.org/big_data_passing_format: $(workspaces.$TASK_NAME.path)/artifacts/$ORIG_PR_NAME/$TASKRUN_NAME/$TASK_PARAM_NAME
    pipelines.kubeflow.org/pipeline_spec: '{"description": "An example for trusted-ai
      integration.", "inputs": [{"default": "anonymous", "name": "namespace", "optional":
      true, "type": "String"}, {"default": "0.2", "name": "fgsm_attack_epsilon", "optional":
      true, "type": "String"}, {"default": "PyTorchModel.py", "name": "model_class_file",
      "optional": true, "type": "String"}, {"default": "ThreeLayerCNN", "name": "model_class_name",
      "optional": true, "type": "String"}, {"default": "processed_data/X_test.npy",
      "name": "feature_testset_path", "optional": true, "type": "String"}, {"default":
      "processed_data/y_test.npy", "name": "label_testset_path", "optional": true,
      "type": "String"}, {"default": "processed_data/p_test.npy", "name": "protected_label_testset_path",
      "optional": true, "type": "String"}, {"default": "0.0", "name": "favorable_label",
      "optional": true, "type": "String"}, {"default": "1.0", "name": "unfavorable_label",
      "optional": true, "type": "String"}, {"default": "[{''race'': 0.0}]", "name":
      "privileged_groups", "optional": true, "type": "String"}, {"default": "[{''race'':
      4.0}]", "name": "unprivileged_groups", "optional": true, "type": "String"},
      {"default": "torch.nn.CrossEntropyLoss()", "name": "loss_fn", "optional": true,
      "type": "String"}, {"default": "torch.optim.Adam(model.parameters(), lr=0.001)",
      "name": "optimizer", "optional": true, "type": "String"}, {"default": "(0, 1)",
      "name": "clip_values", "optional": true, "type": "String"}, {"default": "2",
      "name": "nb_classes", "optional": true, "type": "String"}, {"default": "(1,3,64,64)",
      "name": "input_shape", "optional": true, "type": "String"}], "name": "Trusted AI Pipeline"}'
spec:
  params:
  - name: clip_values
    value: (0, 1)
  - name: favorable_label
    value: '0.0'
  - name: feature_testset_path
    value: processed_data/X_test.npy
  - name: fgsm_attack_epsilon
    value: '0.2'
  - name: input_shape
    value: (1,3,64,64)
  - name: label_testset_path
    value: processed_data/y_test.npy
  - name: loss_fn
    value: torch.nn.CrossEntropyLoss()
  - name: model_class_file
    value: PyTorchModel.py
  - name: model_class_name
    value: ThreeLayerCNN
  - name: namespace
    value: anonymous
  - name: nb_classes
    value: '2'
  - name: optimizer
    value: torch.optim.Adam(model.parameters(), lr=0.001)
  - name: privileged_groups
    value: '[{''race'': 0.0}]'
  - name: protected_label_testset_path
    value: processed_data/p_test.npy
  - name: unfavorable_label
    value: '1.0'
  - name: unprivileged_groups
    value: '[{''race'': 4.0}]'
  pipelineSpec:
    params:
    - name: clip_values
      default: (0, 1)
    - name: favorable_label
      default: '0.0'
    - name: feature_testset_path
      default: processed_data/X_test.npy
    - name: fgsm_attack_epsilon
      default: '0.2'
    - name: input_shape
      default: (1,3,64,64)
    - name: label_testset_path
      default: processed_data/y_test.npy
    - name: loss_fn
      default: torch.nn.CrossEntropyLoss()
    - name: model_class_file
      default: PyTorchModel.py
    - name: model_class_name
      default: ThreeLayerCNN
    - name: namespace
      default: anonymous
    - name: nb_classes
      default: '2'
    - name: optimizer
      default: torch.optim.Adam(model.parameters(), lr=0.001)
    - name: privileged_groups
      default: '[{''race'': 0.0}]'
    - name: protected_label_testset_path
      default: processed_data/p_test.npy
    - name: unfavorable_label
      default: '1.0'
    - name: unprivileged_groups
      default: '[{''race'': 4.0}]'
    tasks:
    - name: trust-ai-train-step
      params:
      - name: action
        value: create
      - name: output
        value: |
          - name: manifest
            valueFrom: '{}'
          - name: name
            valueFrom: '{.metadata.name}'
      - name: success-condition
        value: status.succeeded > 0
      - name: failure-condition
        value: status.failed > 0
      - name: set-ownerreference
        value: "false"
      - name: namespace
        value: $(params.namespace)
      taskSpec:
        params:
        - description: Action on the resource
          name: action
          type: string
        - default: strategic
          description: Merge strategy when using action patch
          name: merge-strategy
          type: string
        - default: ''
          description: An express to retrieval data from resource.
          name: output
          type: string
        - default: ''
          description: A label selector express to decide if the action on resource
            is success.
          name: success-condition
          type: string
        - default: ''
          description: A label selector express to decide if the action on resource
            is failure.
          name: failure-condition
          type: string
        - default: aipipeline/kubectl-wrapper:1.1.1
          description: Kubectl wrapper image
          name: image
          type: string
        - default: "false"
          description: Enable set owner reference for created resource.
          name: set-ownerreference
          type: string
        - name: namespace
        steps:
        - args:
          - --action=$(params.action)
          - --merge-strategy=$(params.merge-strategy)
          - |
            --manifest=apiVersion: batch/v1
            kind: Job
            metadata:
              name: trusted-ai-train-job-$(PIPELINERUN)
              namespace: $(inputs.params.namespace)
            spec:
              template:
                metadata:
                  annotations:
                    sidecar.istio.io/inject: 'false'
                spec:
                  containers:
                  - command:
                    - python
                    - -u
                    - gender_classification_training.py
                    - --data_bucket
                    - mlpipeline
                    - --result_bucket
                    - mlpipeline
                    env:
                    - name: S3_ENDPOINT
                      value: minio-service.kubeflow:9000
                    image: aipipeline/gender-classification:latest
                    name: classification-training
                  restartPolicy: Never
              ttlSecondsAfterFinished: 100
          - --output=$(params.output)
          - --success-condition=$(params.success-condition)
          - --failure-condition=$(params.failure-condition)
          - --set-ownerreference=$(params.set-ownerreference)
          image: $(params.image)
          name: main
          resources: {}
          env:
          - name: PIPELINERUN
            valueFrom:
              fieldRef:
                fieldPath: metadata.labels['tekton.dev/pipelineRun']
        results:
        - name: manifest
          description: '{}'
        - name: name
          description: '{.metadata.name}'
        metadata:
          labels:
            pipelines.kubeflow.org/pipelinename: ''
            pipelines.kubeflow.org/generation: ''
            pipelines.kubeflow.org/cache_enabled: "true"
          annotations:
            tekton.dev/template: ''
      timeout: 525600m
    - name: model-fairness-check
      params:
      - name: favorable_label
        value: $(params.favorable_label)
      - name: feature_testset_path
        value: $(params.feature_testset_path)
      - name: label_testset_path
        value: $(params.label_testset_path)
      - name: model_class_file
        value: $(params.model_class_file)
      - name: model_class_name
        value: $(params.model_class_name)
      - name: privileged_groups
        value: $(params.privileged_groups)
      - name: protected_label_testset_path
        value: $(params.protected_label_testset_path)
      - name: unfavorable_label
        value: $(params.unfavorable_label)
      - name: unprivileged_groups
        value: $(params.unprivileged_groups)
      taskSpec:
        steps:
        - name: main
          args:
          - -u
          - fairness_check.py
          - --model_id
          - training-example
          - --model_class_file
          - $(inputs.params.model_class_file)
          - --model_class_name
          - $(inputs.params.model_class_name)
          - --feature_testset_path
          - $(inputs.params.feature_testset_path)
          - --label_testset_path
          - $(inputs.params.label_testset_path)
          - --protected_label_testset_path
          - $(inputs.params.protected_label_testset_path)
          - --favorable_label
          - $(inputs.params.favorable_label)
          - --unfavorable_label
          - $(inputs.params.unfavorable_label)
          - --privileged_groups
          - $(inputs.params.privileged_groups)
          - --unprivileged_groups
          - $(inputs.params.unprivileged_groups)
          - --metric_path
          - $(results.metric-path.path)
          - --data_bucket_name
          - mlpipeline
          - --result_bucket_name
          - mlpipeline
          command:
          - python
          image: aipipeline/bias-detector:pytorch
          imagePullPolicy: Always
        params:
        - name: favorable_label
        - name: feature_testset_path
        - name: label_testset_path
        - name: model_class_file
        - name: model_class_name
        - name: privileged_groups
        - name: protected_label_testset_path
        - name: unfavorable_label
        - name: unprivileged_groups
        results:
        - name: metric-path
          description: /tmp/outputs/metric_path/data
        metadata:
          labels:
            pipelines.kubeflow.org/pipelinename: ''
            pipelines.kubeflow.org/generation: ''
            pipelines.kubeflow.org/cache_enabled: "true"
          annotations:
            platform: OpenSource
            pipelines.kubeflow.org/component_spec_digest: '{"name": "Model Fairness
              Check", "outputs": [{"description": "Path for fairness check output",
              "name": "metric_path", "type": "String"}], "version": "Model Fairness
              Check@sha256=1706e2527a14053300b36c47885e9ee211eea65930a98ca0f80dde88b675a034"}'
            tekton.dev/template: ''
      runAfter:
      - trust-ai-train-step
      timeout: 525600m
    - name: adversarial-robustness-evaluation
      params:
      - name: clip_values
        value: $(params.clip_values)
      - name: feature_testset_path
        value: $(params.feature_testset_path)
      - name: fgsm_attack_epsilon
        value: $(params.fgsm_attack_epsilon)
      - name: input_shape
        value: $(params.input_shape)
      - name: label_testset_path
        value: $(params.label_testset_path)
      - name: loss_fn
        value: $(params.loss_fn)
      - name: model_class_file
        value: $(params.model_class_file)
      - name: model_class_name
        value: $(params.model_class_name)
      - name: nb_classes
        value: $(params.nb_classes)
      - name: optimizer
        value: $(params.optimizer)
      taskSpec:
        steps:
        - name: main
          args:
          - -u
          - robustness_evaluation_fgsm_pytorch.py
          - --model_id
          - training-example
          - --model_class_file
          - $(inputs.params.model_class_file)
          - --model_class_name
          - $(inputs.params.model_class_name)
          - --feature_testset_path
          - $(inputs.params.feature_testset_path)
          - --label_testset_path
          - $(inputs.params.label_testset_path)
          - --epsilon
          - $(inputs.params.fgsm_attack_epsilon)
          - --loss_fn
          - $(inputs.params.loss_fn)
          - --optimizer
          - $(inputs.params.optimizer)
          - --clip_values
          - $(inputs.params.clip_values)
          - --nb_classes
          - $(inputs.params.nb_classes)
          - --input_shape
          - $(inputs.params.input_shape)
          - --metric_path
          - $(results.metric-path.path)
          - --robust_status
          - $(results.robust-status.path)
          - --data_bucket_name
          - mlpipeline
          - --result_bucket_name
          - mlpipeline
          - --adversarial_accuracy_threshold
          - '0.2'
          command:
          - python
          image: aipipeline/robustness-evaluation:pytorch
          imagePullPolicy: Always
        params:
        - name: clip_values
        - name: feature_testset_path
        - name: fgsm_attack_epsilon
        - name: input_shape
        - name: label_testset_path
        - name: loss_fn
        - name: model_class_file
        - name: model_class_name
        - name: nb_classes
        - name: optimizer
        results:
        - name: metric-path
          description: /tmp/outputs/metric_path/data
        - name: robust-status
          description: /tmp/outputs/robust_status/data
        metadata:
          labels:
            pipelines.kubeflow.org/pipelinename: ''
            pipelines.kubeflow.org/generation: ''
            pipelines.kubeflow.org/cache_enabled: "true"
          annotations:
            platform: OpenSource
            pipelines.kubeflow.org/component_spec_digest: '{"name": "Adversarial Robustness
              Evaluation", "outputs": [{"description": "Path for robustness check
              output", "name": "metric_path", "type": "String"}, {"description": "Path
              for robustness status output", "name": "robust_status", "type": "String"}],
              "version": "Adversarial Robustness Evaluation@sha256=28e3f0baf616b9f2b32085a66efbcb3b462db8ccb2d509c024b87ab0d3337fe1"}'
            tekton.dev/template: ''
      runAfter:
      - trust-ai-train-step
      timeout: 525600m
  timeout: 525600m