03 Deploy as Web Service.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "3d7e0c9c-5565-4096-8dfb-1e09f02ed210",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "#Deploy model to Azure Databricks"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "dedf1279-7214-48aa-8f91-b60192f704aa",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "In this notebook, you will deploy the best performing model you selected previously as a web service hosted in in Azure Databricks cluster."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "4b2f964f-c27c-48b0-8679-92aaa5f52b70",
     "showTitle": false,
     "title": ""
    }
   },
   "outputs": [
    {
     "output_type": "display_data",
     "data": {
      "text/html": [
       "<style scoped>\n  .ansiout {\n    display: block;\n    unicode-bidi: embed;\n    white-space: pre-wrap;\n    word-wrap: break-word;\n    word-break: break-all;\n    font-family: \"Source Code Pro\", \"Menlo\", monospace;;\n    font-size: 13px;\n    color: #555;\n    margin-left: 4px;\n    line-height: 19px;\n  }\n</style>\n<div class=\"ansiout\"></div>"
      ]
     },
     "metadata": {
      "application/vnd.databricks.v1+output": {
       "addedWidgets": {},
       "arguments": {},
       "data": "<div class=\"ansiout\"></div>",
       "datasetInfos": [],
       "metadata": {},
       "removedWidgets": [],
       "type": "html"
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import mlflow\n",
    "import mlflow.spark\n",
    "from mlflow.tracking import MlflowClient\n",
    "import time\n",
    "from mlflow.entities.model_registry.model_version_status import ModelVersionStatus\n",
    "\n",
    "client = MlflowClient()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "b28982ae-77e1-47cc-9b0b-3d8c28ce6819",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "First, get the experiment from the prior notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "c2aaab8d-cea6-4509-9fc9-714510ccf28c",
     "showTitle": false,
     "title": ""
    }
   },
   "outputs": [
    {
     "output_type": "display_data",
     "data": {
      "text/html": [
       "<style scoped>\n  .ansiout {\n    display: block;\n    unicode-bidi: embed;\n    white-space: pre-wrap;\n    word-wrap: break-word;\n    word-break: break-all;\n    font-family: \"Source Code Pro\", \"Menlo\", monospace;;\n    font-size: 13px;\n    color: #555;\n    margin-left: 4px;\n    line-height: 19px;\n  }\n</style>\n<div class=\"ansiout\"></div>"
      ]
     },
     "metadata": {
      "application/vnd.databricks.v1+output": {
       "addedWidgets": {},
       "arguments": {},
       "data": "<div class=\"ansiout\"></div>",
       "datasetInfos": [],
       "metadata": {},
       "removedWidgets": [],
       "type": "html"
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "user_name = dbutils.notebook.entry_point.getDbutils().notebook().getContext().tags().apply('user')\n",
    "experiment_name = \"/Users/{user_name}/BigDataVis/Exercise 2/02 Train and Evaluate Models\".format(user_name=user_name)\n",
    "\n",
    "experiment = client.get_experiment_by_name(experiment_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "2165b071-c198-4969-9ff6-1d154c0c9619",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "Get the latest run from the experiment--as a reminder, this is the cross-validation model.  In order to retrieve the model itself, you will need to get the **parent** run ID, as that is where the model details are stored."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "d5752429-b3b1-40d4-b8c5-8a14ae05a2b0",
     "showTitle": false,
     "title": ""
    }
   },
   "outputs": [
    {
     "output_type": "display_data",
     "data": {
      "text/html": [
       "<style scoped>\n  .ansiout {\n    display: block;\n    unicode-bidi: embed;\n    white-space: pre-wrap;\n    word-wrap: break-word;\n    word-break: break-all;\n    font-family: \"Source Code Pro\", \"Menlo\", monospace;;\n    font-size: 13px;\n    color: #555;\n    margin-left: 4px;\n    line-height: 19px;\n  }\n</style>\n<div class=\"ansiout\">Out[10]: &#39;runs:/c043b4b1a8c7415299671dff1e2f48af/model&#39;</div>"
      ]
     },
     "metadata": {
      "application/vnd.databricks.v1+output": {
       "addedWidgets": {},
       "arguments": {},
       "data": "<div class=\"ansiout\">Out[10]: &#39;runs:/c043b4b1a8c7415299671dff1e2f48af/model&#39;</div>",
       "datasetInfos": [],
       "metadata": {},
       "removedWidgets": [],
       "type": "html"
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "experiment_id = experiment.experiment_id\n",
    "runs_df = client.search_runs(experiment_id, order_by=[\"attributes.start_time desc\"], max_results=1)\n",
    "run_id = runs_df[0].data.tags[\"mlflow.parentRunId\"]\n",
    "\n",
    "model_name = \"Delay Estimator\"\n",
    "\n",
    "artifact_path = \"model\"\n",
    "model_uri = \"runs:/{run_id}/{artifact_path}\".format(run_id=run_id, artifact_path=artifact_path)\n",
    "model_uri"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "fa713f1c-95f7-41cc-b79a-8815e46f35fe",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "The next step is to register this model with MLflow.  This may take anywhere fdrom 30 seconds to 5 minutes to complete."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "0a6206e7-2e95-4f50-ba7d-99eae1c2b6fb",
     "showTitle": false,
     "title": ""
    }
   },
   "outputs": [
    {
     "output_type": "display_data",
     "data": {
      "text/html": [
       "<style scoped>\n  .ansiout {\n    display: block;\n    unicode-bidi: embed;\n    white-space: pre-wrap;\n    word-wrap: break-word;\n    word-break: break-all;\n    font-family: \"Source Code Pro\", \"Menlo\", monospace;;\n    font-size: 13px;\n    color: #555;\n    margin-left: 4px;\n    line-height: 19px;\n  }\n</style>\n<div class=\"ansiout\">Registered model &#39;Delay Estimator&#39; already exists. Creating a new version of this model...\nCreated version &#39;5&#39; of model &#39;Delay Estimator&#39;.\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: READY\n</div>"
      ]
     },
     "metadata": {
      "application/vnd.databricks.v1+output": {
       "addedWidgets": {},
       "arguments": {},
       "data": "<div class=\"ansiout\">Registered model &#39;Delay Estimator&#39; already exists. Creating a new version of this model...\nCreated version &#39;5&#39; of model &#39;Delay Estimator&#39;.\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: PENDING_REGISTRATION\nModel status: READY\n</div>",
       "datasetInfos": [],
       "metadata": {},
       "removedWidgets": [],
       "type": "html"
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "model_details = mlflow.register_model(model_uri=model_uri, name=model_name)\n",
    "\n",
    "# Wait until the model is ready\n",
    "def wait_until_ready(model_name, model_version):\n",
    "  client = MlflowClient()\n",
    "  for _ in range(500):\n",
    "    model_version_details = client.get_model_version(\n",
    "      name=model_name,\n",
    "      version=model_version,\n",
    "    )\n",
    "    status = ModelVersionStatus.from_string(model_version_details.status)\n",
    "    print(\"Model status: %s\" % ModelVersionStatus.to_string(status))\n",
    "    if status == ModelVersionStatus.READY:\n",
    "      break\n",
    "    time.sleep(1)\n",
    "\n",
    "wait_until_ready(model_details.name, model_details.version)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "efa6ba87-5199-4813-85eb-903670af44d0",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "Now that the model is registered, move it to Production.  This will make the current model the production model, allowing you to serve this iteration of the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "2cd5f04c-9ee2-41bc-9de6-6a72afdd0512",
     "showTitle": false,
     "title": ""
    }
   },
   "outputs": [
    {
     "output_type": "display_data",
     "data": {
      "text/html": [
       "<style scoped>\n  .ansiout {\n    display: block;\n    unicode-bidi: embed;\n    white-space: pre-wrap;\n    word-wrap: break-word;\n    word-break: break-all;\n    font-family: \"Source Code Pro\", \"Menlo\", monospace;;\n    font-size: 13px;\n    color: #555;\n    margin-left: 4px;\n    line-height: 19px;\n  }\n</style>\n<div class=\"ansiout\">The current model stage is: &#39;Production&#39;\n</div>"
      ]
     },
     "metadata": {
      "application/vnd.databricks.v1+output": {
       "addedWidgets": {},
       "arguments": {},
       "data": "<div class=\"ansiout\">The current model stage is: &#39;Production&#39;\n</div>",
       "datasetInfos": [],
       "metadata": {},
       "removedWidgets": [],
       "type": "html"
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "client.transition_model_version_stage(\n",
    "  name=model_details.name,\n",
    "  version=model_details.version,\n",
    "  stage='Production',\n",
    ")\n",
    "model_version_details = client.get_model_version(\n",
    "  name=model_details.name,\n",
    "  version=model_details.version,\n",
    ")\n",
    "print(\"The current model stage is: '{stage}'\".format(stage=model_version_details.current_stage))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "e7e5d221-0fda-4ff9-be20-8aada6b3afde",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "## Serve the Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "f6201fc6-75ea-4931-a429-eb18ff939a34",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "You have already created a model, but the next step will be to serve the model.  At present, the best way to do this is to select the **Models** menu option on the left-hand pane.  Note that you will only see this menu in the **Machine Learning** view.  If you are still in the **Data Science & Engineering** view, select the drop-down option from the menu and select **Machine Learning** first.\n",
    "\n",
    "In the Models menu, select the **Delay Estimator** model.  Navigate to the **Serving** menu and enable serving.  This will build a Databricks cluster to allow you to perform inference.\n",
    "\n",
    "Wait for both the Status indicator as well as the Production model indicator to read Ready and then copy the Model URL with \"Production\" in it.  This will take 5-10 minutes to complete."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "7bb8e084-129b-4327-9682-fea6404bf0ef",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "#Test the scoring web service"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "ceac1b04-73b7-44ca-831a-95a94f87db0c",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "In order to test the service, create two sample rows for testing and load them into a Pandas DataFrame."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "5bd7e7b6-e869-4cef-add9-2414f457d6c8",
     "showTitle": false,
     "title": ""
    }
   },
   "outputs": [
    {
     "output_type": "display_data",
     "data": {
      "text/html": [
       "<style scoped>\n  .ansiout {\n    display: block;\n    unicode-bidi: embed;\n    white-space: pre-wrap;\n    word-wrap: break-word;\n    word-break: break-all;\n    font-family: \"Source Code Pro\", \"Menlo\", monospace;;\n    font-size: 13px;\n    color: #555;\n    margin-left: 4px;\n    line-height: 19px;\n  }\n</style>\n<div class=\"ansiout\"></div>"
      ]
     },
     "metadata": {
      "application/vnd.databricks.v1+output": {
       "addedWidgets": {},
       "arguments": {},
       "data": "<div class=\"ansiout\"></div>",
       "datasetInfos": [],
       "metadata": {},
       "removedWidgets": [],
       "type": "html"
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import json\n",
    "import pandas as pd\n",
    "\n",
    "# Create two records for testing the prediction\n",
    "test_input1 = {\"OriginAirportCode\":\"SAT\",\"Month\":5,\"DayofMonth\":5,\"CRSDepHour\":13,\"DayOfWeek\":7,\"Carrier\":\"MQ\",\"DestAirportCode\":\"ORD\",\"WindSpeed\":9,\"SeaLevelPressure\":30.03,\"HourlyPrecip\":0}\n",
    "\n",
    "test_input2 = {\"OriginAirportCode\":\"ATL\",\"Month\":2,\"DayofMonth\":5,\"CRSDepHour\":8,\"DayOfWeek\":4,\"Carrier\":\"MQ\",\"DestAirportCode\":\"MCO\",\"WindSpeed\":3,\"SeaLevelPressure\":31.03,\"HourlyPrecip\":0}\n",
    "\n",
    "# package the inputs into a JSON string and test run() in local notebook\n",
    "inputs = pd.DataFrame([test_input1, test_input2])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "03ec59e5-f190-4c68-a45c-e434d42449cc",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "Fill in the values for `url` and `personal_access_token` in the function below and then run the following command to ensure that you get back results from the serving cluster.  The `url` is the model serving URL you created in the **Serve the Model** task above, and `personal_access_token` is the PAT you created in the hands-on lab."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "c1d8181e-7cb3-408e-a0c7-a5e71d9b7a16",
     "showTitle": false,
     "title": ""
    }
   },
   "outputs": [
    {
     "output_type": "display_data",
     "data": {
      "text/html": [
       "<style scoped>\n  .ansiout {\n    display: block;\n    unicode-bidi: embed;\n    white-space: pre-wrap;\n    word-wrap: break-word;\n    word-break: break-all;\n    font-family: \"Source Code Pro\", \"Menlo\", monospace;;\n    font-size: 13px;\n    color: #555;\n    margin-left: 4px;\n    line-height: 19px;\n  }\n</style>\n<div class=\"ansiout\"></div>"
      ]
     },
     "metadata": {
      "application/vnd.databricks.v1+output": {
       "addedWidgets": {},
       "arguments": {},
       "data": "<div class=\"ansiout\"></div>",
       "datasetInfos": [],
       "metadata": {},
       "removedWidgets": [],
       "type": "html"
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import os\n",
    "import requests\n",
    "import numpy as np\n",
    "\n",
    "def create_tf_serving_json(data):\n",
    "  return {'inputs': {name: data[name].tolist() for name in data.keys()} if isinstance(data, dict) else data.tolist()}\n",
    "\n",
    "def score_model(dataset):\n",
    "  url = 'https://adb-1404434138497066.6.azuredatabricks.net/model/Delay%20Estimator/Production/invocations' # Enter your URL here\n",
    "  personal_access_token = 'dapi20907ab7c44595a3e0c57984497889d7-3' # Enter your Personal Access Token here\n",
    "  headers = {'Authorization': f'Bearer {personal_access_token}'}\n",
    "  data_json = dataset.to_dict(orient='split') if isinstance(dataset, pd.DataFrame) else create_tf_serving_json(dataset)\n",
    "  response = requests.request(method='POST', headers=headers, url=url, json=data_json)\n",
    "  if response.status_code != 200:\n",
    "    raise Exception(f'Request failed with status {response.status_code}, {response.text}')\n",
    "  return response.json()\n",
    "\n",
    "score_model(inputs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "3a6ea164-09f8-4f15-b08e-9b60ee915247",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "# You are done!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "application/vnd.databricks.v1+cell": {
     "cellMetadata": {
      "byteLimit": 2048000,
      "rowLimit": 10000
     },
     "inputWidgets": {},
     "nuid": "82e43687-32ce-44d4-af31-f867af9d69fd",
     "showTitle": false,
     "title": ""
    }
   },
   "source": [
    "Congratulations, you have completed this team challenge!\n",
    "\n",
    "Please continue on to Exercise 3 in the hands-on lab document."
   ]
  }
 ],
 "metadata": {
  "application/vnd.databricks.v1+notebook": {
   "dashboards": [],
   "language": "python",
   "notebookMetadata": {
    "pythonIndentUnit": 4
   },
   "notebookName": "03 Deploy as Web Service",
   "notebookOrigID": 1036921577453893,
   "widgets": {}
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}