Merge pull request #91 from microsoft/mabou/patch

Mabou/patch

.ci/azure-pipelines-v2.yml

@@ -1,134 +1,30 @@
-# MLHyperparameterTuning Pipeline 
-
-trigger: none
-
-variables:
-  BuildConfiguration: Release
-  BuildBinariesDirectory: $(Build.BinariesDirectory)
-  BuildPlatform: any cpu
-  DotNetCoreBuildVersion: 2.2.108
-  DotNetRuntimeTarget: ubuntu.18.04-x64
-  AgentToolsDirectory: $(Agent.ToolsDirectory)
-  CloudPlatform: AzureCloud
-  ProductName: Trident
-  TridentWorkloadType: $(WorkloadType)
-  TridentWorkloadTypeShort: $(WorkloadTypeShort)
-  DeployLocation: eastus
-  Agent: agce-ai
-  azureSubscription: AG-AzureCAT-AIDevOps-Test-COGSNonProd-IO1685734(0ca618d2-22a8-413a-96d0-0f1b531129c3)
-  azure_subscription: 0ca618d2-22a8-413a-96d0-0f1b531129c3
-
-jobs:
-- job: MLHyperparameterTuningJob
-  timeoutInMinutes: 300
-  cancelTimeoutInMinutes: 2
-  pool:
-    vmImage: 'Ubuntu-16.04'
-
-  steps:
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      which conda
-      conda env create -f environment.yml
-      conda env list
-      conda activate MLHyperparameterTuning
-      conda env list  
-      echo Login Azure Account
-      az login -t $(sptenent) --service-principal -u $(spidentity) --password $(spsecret)
-      echo Try and figure out what account set takes
-      az account set -h
-      echo Try and set it.
-      az account set --subscription $(subscriptionid)
-#      papermill 01_Data_Prep.ipynb 01_Data_Prep_Output.ipynb --log-output --no-progress-bar -k python3 
-    displayName: 'Configuration'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 00_Data_Prep.ipynb
-      papermill 00_Data_Prep.ipynb 00_Data_Prep_Output.ipynb --log-output --no-progress-bar -k python3 
-    displayName: '00_Data_Prep.ipynb'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 01_Training_Script.ipynb
-      papermill 01_Training_Script.ipynb 01_Training_Script_Output.ipynb --log-output --no-progress-bar -k python3 
-    displayName: '01_Training_Script.ipynb'
- 
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 02_Testing_Script.ipynb
-      papermill 02_Testing_Script.ipynb 02_Testing_Script_Output.ipynb --log-output --no-progress-bar -k python3 
-    displayName: '02_Testing_Script.ipynb'
- 
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 03_Run_Locally.ipynb
-      papermill 03_Run_Locally.ipynb 03_Run_Locally_Output.ipynb --log-output --no-progress-bar -k python3 -p selected_subscription $(subscriptionid) -p resource_group $(azurergname) 
-    displayName: '03_Run_Locally.ipynb'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 04_Hyperparameter_Random_Search.ipynb
-      papermill 04_Hyperparameter_Random_Search.ipynb 04_Hyperparameter_Random_Search_Output.ipynb --log-output --no-progress-bar -k python3 -p max_total_runs $(dsmaxruns)
-    displayName: '04_Hyperparameter_Random_Search.ipynb'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 05_Train_Best_Model.ipynb
-      papermill 05_Train_Best_Model.ipynb 05_Train_Best_Model_Output.ipynb --log-output --no-progress-bar -k python3 
-    displayName: '05_Train_Best_Model.ipynb'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 06_Test_Best_Model.ipynb
-      papermill 06_Test_Best_Model.ipynb 06_Test_Best_Model_Output.ipynb --log-output --no-progress-bar -k python3 
-    displayName: '06_Test_Best_Model.ipynb'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 07_Train_With_AML_Pipeline.ipynb
-      papermill 07_Train_With_AML_Pipeline.ipynb 07_Train_With_AML_Pipeline_Output.ipynb --log-output --no-progress-bar -k python3 -p max_total_runs $(dsmaxruns)
-    displayName: '07_Train_With_AML_Pipeline.ipynb'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Executing 08_Tear_Down.ipynb
-      papermill 08_Tear_Down.ipynb 08_Tear_Down_Output.ipynb --log-output --no-progress-bar -k python3 
-    displayName: '08_Tear_Down.ipynb'
-
-  - bash: |
-      source /usr/share/miniconda/etc/profile.d/conda.sh
-      conda activate MLHyperparameterTuning
-      echo Execute Resource Group Delete
-      existResponse=$(az group exists -n $(azurergname))
-      if [ "$existResponse" == "true" ]; then
-        echo Deleting project resource group  
-        az group delete --name $(azurergname) --yes
-      else
-        echo Project resource group did not exist
-      fi
-      echo Done Cleanup
-    displayName: 'Backup Cleanup'
-    condition: or(canceled(),failed())
-
-  - task: CreateWorkItem@1
-    inputs:
-      workItemType: 'Issue'
-      title: $(System.TeamProject) - Build $(Build.BuildNumber) Failed
-      assignedTo: 'Mario Bourgoin <[email protected]>'
-      associate: true
-      teamProject: $(System.TeamProject)
-
-      fieldMappings: |
-        Description=Branch: Branch $(Build.SourceBranch) failed to build. Go to Boards>WorkItems and tag the failure type.
-    displayName: 'Create work item on failure'
-    condition: failed()
+# MLHyperparameterTuning  Pipeline
+#
+# A Github Service Connection must also be created with the name "AIArchitecturesAndPractices-GitHub"
+
+resources:
+  repositories:
+    - repository: aitemplates
+      type: github
+      name: microsoft/AI
+      endpoint: AIArchitecturesAndPractices-GitHub
+
+trigger:
+  branches:
+    include:
+    - master
+    - mabou/instrument
+
+pr:
+  autoCancel: true
+  branches:
+    include:
+    - master
+    - mabou/instrument
+
+stages:
+- template: .ci/stages/deploy_notebooks_stages_v2.yml@aitemplates
+  parameters:
+    jobDisplayName: MLScoreDeployJob
+    DefaultWorkingDirectory: $(System.DefaultWorkingDirectory)
+    workload_vars: ../vars/mlhyperparametertuning_vars.yml

.ci/vars/agce_devops_sub_vars.yml

@@ -0,0 +1,3 @@
+variables:
+  azure_subscription: 0ca618d2-22a8-413a-96d0-0f1b531129c3
+  azureSubscription: AG-AzureCAT-AIDevOps-Test-COGSNonProd-IO1685734(0ca618d2-22a8-413a-96d0-0f1b531129c3)

.ci/vars/mlhyperparametertuning_vars.yml

@@ -0,0 +1,7 @@
+variables:
+  DeploymentName: MLScoreDeployJob
+  TridentWorkloadTypeShort: aimlscore
+  DeployLocation: eastus  
+  ProjectLocation: "."
+  PythonPath: "."
+  Template: MLTrainDeployAMLJob.yml

01_Training_Script.ipynb

@@ -440,7 +440,27 @@
    "metadata": {},
    "source": [
     "## Run the script to see that it works <a id='run'></a>\n",
-    "This should take around ten minutes."
+    "Set the effort expended to train the classifier."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "tags": [
+     "parameters"
+    ]
+   },
+   "outputs": [],
+   "source": [
+    "estimators = 1000"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Run the classifier script. This should take about 10 minutes."
    ]
   },
   {
@@ -451,7 +471,7 @@
    },
    "outputs": [],
    "source": [
-    "%run -t scripts/TrainClassifier.py --estimators 1000 --match 5 --ngrams 2 --min_child_samples 10 --save model"
+    "%run -t scripts/TrainClassifier.py --estimators $estimators --match 5 --ngrams 2 --min_child_samples 10 --save model"
    ]
   },
   {

03_Run_Locally.ipynb

@@ -39,7 +39,7 @@
    "metadata": {},
    "source": [
     "## Azure subscription <a id='subscription'></a>\n",
-    "If you have multiple subscriptions select the subscription you want to use. You may supply either the subscription's name or the subscription's ID. If you want to run this in a different location that supports HyperDrive, you may enter the one you want to use. You can also set the name of the resource group in which this tutorial will add resources. *IMPORTANT NOTE:* The last notebook in this example will delete this resource group and all associated resources."
+    "If you have multiple subscriptions select the subscription you want to use. You may supply either the subscription's name or the subscription's ID. If you want to run this in a different location that supports HyperDrive, you may enter the one you want to use. You can also set the name of the resource group in which this tutorial will add resources. *IMPORTANT NOTE:* The last notebook in this example will delete this resource group and all associated resources. We also define the number of estimators to use for the local run."
    ]
   },
   {
@@ -55,7 +55,8 @@
     "subscription_name=\"YOUR_SUBSCRIPTION_NAME\"\n",
     "subscription_id=\"YOUR_SUBSCRIPTION_ID\"\n",
     "location=\"eastus\"\n",
-    "resource_group=\"hypetuning\""
+    "resource_group=\"hypetuning\"\n",
+    "estimators = 1000"
    ]
   },
   {
@@ -179,10 +180,10 @@
     "est = Estimator(source_directory=os.path.join('.', 'scripts'), \n",
     "                entry_script='TrainClassifier.py',\n",
     "                script_params={'--data-folder': os.path.abspath('.'),\n",
-    "                               '--estimators': '1000',\n",
-    "                               '--match': '5',\n",
-    "                               '--ngrams': '2',\n",
-    "                               '--min_child_samples': '10',\n",
+    "                               '--estimators': estimators,\n",
+    "                               '--match': 5,\n",
+    "                               '--ngrams': 2,\n",
+    "                               '--min_child_samples': 10,\n",
     "                               \"--save\": \"local_model\"},\n",
     "                compute_target='local',\n",
     "                conda_packages=['pandas==0.23.4',\n",

04_Hyperparameter_Random_Search.ipynb

@@ -202,7 +202,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "This hyperparameter space specifies a grid of 9,360 unique configuration points (4 `ngrams` X 39 `match` X 30 `min_child_samples` X 2 `unweighted`). We control the resources used by the search through specifying a maximum number of configuration points to sample as `max_total_runs`."
+    "This hyperparameter space specifies a grid of 9,360 unique configuration points (4 `ngrams` X 39 `match` X 30 `min_child_samples` X 2 `unweighted`). We control the resources used by the search through specifying a maximum number of configuration points to sample as `max_total_runs`. We also define the number of estimators to use for each run."
    ]
   },
   {
@@ -215,7 +215,8 @@
    },
    "outputs": [],
    "source": [
-    "max_total_runs = 96"
+    "max_total_runs = 96\n",
+    "estimators = 1000"
    ]
   },
   {
@@ -270,7 +271,7 @@
     "estimator = Estimator(source_directory=os.path.join('.', 'scripts'),\n",
     "                      entry_script='TrainClassifier.py',\n",
     "                      script_params={'--data-folder': ds.as_mount(),\n",
-    "                                     '--estimators': 1000},\n",
+    "                                     '--estimators': estimators},\n",
     "                      compute_target=compute_target,\n",
     "                      conda_packages=['pandas==0.23.4',\n",
     "                                      'scikit-learn==0.21.3',\n",

05_Train_Best_Model.ipynb

@@ -166,8 +166,8 @@
    },
    "outputs": [],
    "source": [
-    "model_estimators = 8 * int(best_parameters['--estimators'])\n",
-    "model_estimators"
+    "estimators = 8 * int(best_parameters['--estimators'])\n",
+    "estimators"
    ]
   },
   {
@@ -186,7 +186,7 @@
     "ds = ws.get_default_datastore()\n",
     "model_parameters = best_parameters.copy()\n",
     "model_parameters['--data-folder'] = ds.as_mount()\n",
-    "model_parameters['--estimators'] = model_estimators\n",
+    "model_parameters['--estimators'] = estimators\n",
     "model_parameters['--save'] = 'FAQ_ranker'\n",
     "pd.Series(model_parameters, name='Value').to_frame()"
    ]

07_Train_With_AML_Pipeline.ipynb

@@ -292,7 +292,7 @@
    "metadata": {},
    "source": [
     "## Create AML Pipeline Tuning Step <a id='aml_pipeline_tune_step'></a>\n",
-    "We create a HyperDrive step in the AML pipeline to perform a search for hyperparameters. The `tune_estimators` pipeline parameter that controls the number of estimators used in tuning deliberately has a low default value for the speed of pipeline testing. The `tune_steps_data` output pipeline data is only used to synchronize with the next pipeline step."
+    "We create a HyperDrive step in the AML pipeline to perform a search for hyperparameters. The `tune_estimators` pipeline parameter that controls the number of estimators used in tuning deliberately has a low default value for the speed of pipeline testing."
    ]
   },
   {
@@ -302,15 +302,13 @@
    "outputs": [],
    "source": [
     "tune_step_name=\"tune_model\"\n",
-    "tune_steps_data = PipelineData(\"tune_steps_data\", datastore=ds)\n",
     "tune_estimators = PipelineParameter(name=\"tune_estimators\", default_value=1)  # Set to 1000 when running the pipeline.\n",
     "tune_step = HyperDriveStep(\n",
     "    name=tune_step_name,\n",
     "    hyperdrive_config=hyperdrive_run_config,\n",
     "    estimator_entry_script_arguments=[\"--data-folder\", data_folder,\n",
     "                                      \"--estimators\", tune_estimators],\n",
     "    inputs=[data_folder],\n",
-    "    outputs=[tune_steps_data],\n",
     "    allow_reuse=False)"
    ]
   },
@@ -404,7 +402,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Creating PythonScript Step for AML pipeline to get the best run's hyperparameters. The `tune_steps_data` input pipeline data is only used to synchronize with the previous pipeline step."
+    "Creating PythonScript Step for AML pipeline to get the best run's hyperparameters."
    ]
   },
   {
@@ -428,18 +426,18 @@
     "    arguments=[\"--hd-step\", tune_step_name,\n",
     "               \"--output-steps-data\", bh_steps_data,\n",
     "               \"--hyperparameters\", bh_hyperparameters_file],\n",
-    "    inputs=[tune_steps_data],\n",
     "    outputs=[bh_steps_data],\n",
     "    runconfig=bh_run_config,\n",
-    "    allow_reuse=False)"
+    "    allow_reuse=False)\n",
+    "bh_step.run_after(tune_step)"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
     "## Create AML Pipeline Best Model Step <a id='aml_pipeline_estimator_step'></a>\n",
-    "This step passes the hyperparameters file from the previous step to the training script to create the best model. The `best_estimators` pipeline parameter that controls the number of estimators used in getting the best model deliberately has a low default value for the speed of pipeline testing. The `bm_steps_data` output pipeline data is only used to synchronize with the next pipeline step."
+    "This step passes the hyperparameters file from the previous step to the training script to create the best model. The `best_estimators` pipeline parameter that controls the number of estimators used in getting the best model deliberately has a low default value for the speed of pipeline testing."
    ]
   },
   {
@@ -449,7 +447,6 @@
    "outputs": [],
    "source": [
     "bm_step_name=\"best_model\"\n",
-    "bm_steps_data = PipelineData(\"bm_steps_data\", datastore=ds)\n",
     "bm_estimators = PipelineParameter(name=\"best_estimators\", default_value=1)  # Set to 8000 when running the pipeline\n",
     "bm_estimator = Estimator(source_directory=os.path.join('.', 'scripts'),  # Use a new Estimator as a bug workaround\n",
     "                         entry_script='TrainClassifier.py',\n",
@@ -467,7 +464,6 @@
     "                                      \"--save\", model_name],\n",
     "    compute_target=compute_target,\n",
     "    inputs=[data_folder, bh_steps_data],\n",
-    "    outputs=[bm_steps_data],\n",
     "    allow_reuse=False)"
    ]
   },
@@ -532,7 +528,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Creating PythonScript Step for AML pipeline to register the best model. The `bm_steps_data` input pipeline data is only used to synchronize with the previous pipeline step."
+    "Creating PythonScript Step for AML pipeline to register the best model."
    ]
   },
   {
@@ -554,9 +550,9 @@
     "    arguments=[\"--es-step\", bm_step_name,\n",
     "               \"--outputs\", \"outputs\",\n",
     "               \"--model-name\", model_name],\n",
-    "    inputs=[bm_steps_data],\n",
     "    runconfig=rm_run_config,\n",
-    "    allow_reuse=False)"
+    "    allow_reuse=False)\n",
+    "rm_step.run_after(bm_step)"
    ]
   },
   {
@@ -671,7 +667,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.7.3"
+   "version": "3.6.7"
   }
  },
  "nbformat": 4,