deploy_model/fix

tutorials/get-started-notebooks/deploy-model.ipynb

@@ -434,18 +434,26 @@
    },
    "outputs": [],
    "source": [
-    "from azure.ai.ml.entities import ManagedOnlineDeployment\n",
+    "from azure.ai.ml.entities import ManagedOnlineDeployment, Environment, CodeConfiguration\n",
     "\n",
     "# Choose the latest version of the registered model for deployment\n",
     "model = ml_client.models.get(name=registered_model_name, version=latest_model_version)\n",
     "\n",
+    "# define the deployment environment\n",
+    "env = Environment(\n",
+    "    conda_file=\"./deploy/deployment-env.yaml\",\n",
+    "    image=\"mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu22.04:latest\",\n",
+    ")\n",
+    "\n",
     "# define an online deployment\n",
     "# if you run into an out of quota error, change the instance_type to a comparable VM that is available.\n",
     "# Learn more on https://azure.microsoft.com/en-us/pricing/details/machine-learning/.\n",
     "blue_deployment = ManagedOnlineDeployment(\n",
     "    name=\"blue\",\n",
     "    endpoint_name=online_endpoint_name,\n",
     "    model=model,\n",
+    "    environment=env,\n",
+    "    code_configuration=CodeConfiguration(code=\"./deploy\", scoring_script=\"score.py\"),\n",
     "    instance_type=\"Standard_DS3_v2\",\n",
     "    instance_count=1,\n",
     ")"
@@ -637,13 +645,21 @@
     "# pick the model to deploy. Here you use the latest version of the registered model\n",
     "model = ml_client.models.get(name=registered_model_name, version=latest_model_version)\n",
     "\n",
+    "# define the deployment environment\n",
+    "env = Environment(\n",
+    "    conda_file=\"./deploy/deployment-env.yaml\",\n",
+    "    image=\"mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu22.04:latest\",\n",
+    ")\n",
+    "\n",
     "# define an online deployment using a more powerful instance type\n",
     "# if you run into an out of quota error, change the instance_type to a comparable VM that is available.\n",
     "# Learn more on https://azure.microsoft.com/en-us/pricing/details/machine-learning/.\n",
     "green_deployment = ManagedOnlineDeployment(\n",
     "    name=\"green\",\n",
     "    endpoint_name=online_endpoint_name,\n",
     "    model=model,\n",
+    "    environment=env,\n",
+    "    code_configuration=CodeConfiguration(code=\"./deploy\", scoring_script=\"score.py\"),\n",
     "    instance_type=\"Standard_F4s_v2\",\n",
     "    instance_count=1,\n",
     ")\n",

tutorials/get-started-notebooks/deploy/deployment-env.yaml

@@ -0,0 +1,12 @@
+channels:
+- conda-forge
+dependencies:
+- python=3.9.18
+- pip
+- pip:
+  - azureml-inference-server-http
+  - scikit-learn==0.24.2
+  - numpy==1.23.5
+  - scipy==1.9.3
+  - cloudpickle==2.2.0
+name: deployment-env

tutorials/get-started-notebooks/deploy/score.py

@@ -0,0 +1,22 @@
+import os
+import json
+import pickle
+import numpy as np
+
+
+def init():
+    global model
+    model_dir = os.getenv("AZUREML_MODEL_DIR")
+    # Handle possible directory layouts
+    model_path = os.path.join(model_dir, "model.pkl")
+    if not os.path.exists(model_path):
+        model_path = os.path.join(model_dir, "credit_defaults_model", "model.pkl")
+    with open(model_path, "rb") as f:
+        model = pickle.load(f)
+
+
+def run(raw_data):
+    data = json.loads(raw_data)
+    input_data = data.get("input_data", data)
+    result = model.predict(np.array(input_data["data"]))
+    return json.dumps(result.tolist())