- added notebook for testing webservice.

- added python file for deploying a web service from an Image.

deploy_service.py

@@ -21,7 +21,7 @@ def deploy_service(args):
     image = Image(workspace=workspace, name=args.image_name, version=args.image_id)
 
     # Get webservice configuration
-    aci_config = AciWebservice.deploy_configuration(cpu_cores=1, memory_gb=1,
+    aci_config = AciWebservice.deploy_configuration(cpu_cores=1, memory_gb=5,
                                                     description='Predict digit value from images '
                                                                 'of handwritten digits')
 

environment.yml

@@ -9,5 +9,6 @@ dependencies:
   - python=3.6.2
   - tensorflow==1.10.0
   - numpy==1.16.3
+  - pillow==6.0.0
   - pip:
     - azureml-defaults

notebooks/inference_graph_test.ipynb

@@ -11,18 +11,9 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 85,
+   "execution_count": 1,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "The autoreload extension is already loaded. To reload it, use:\n",
-      "  %reload_ext autoreload\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "# Configure Notebook\n",
     "import warnings\n",
@@ -89,17 +80,9 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 112,
+   "execution_count": null,
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "INFO:tensorflow:Restoring parameters from /home/sebastiangoodfellow/Documents/Code/mnist-azure/assets/outputs/checkpoints/model\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "# Get workspace\n",
     "ws = Workspace.get(name='mnist-azure', subscription_id='', \n",

notebooks/test_service.ipynb

@@ -11,9 +11,18 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": 70,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n"
+     ]
+    }
+   ],
    "source": [
     "# Configure Notebook\n",
     "import warnings\n",
@@ -51,14 +60,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": 71,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "['val_0.jpg', 'val_1.jpg', 'val_10.jpg', 'val_100.jpg', 'val_1000.jpg']\n"
+      "['val_2670.jpg', 'val_4752.jpg', 'val_7445.jpg', 'val_7043.jpg', 'val_3452.jpg']\n"
      ]
     }
    ],
@@ -73,12 +82,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": 72,
    "metadata": {},
    "outputs": [],
    "source": [
     "# Get workspace\n",
-    "workspace = Workspace.get(name='mnist-azure', subscription_id='', resource_group='')\n",
+    "workspace = Workspace.get(name='mnist-azure', subscription_id='', \n",
+    "                          resource_group='')\n",
     "\n",
     "# Get web service\n",
     "service = workspace.webservices['mnist-tf']\n",
@@ -87,9 +97,24 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 73,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "10c76329db2b4f759dde82e9336dd327",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "interactive(children=(IntSlider(value=0, description='Image ID', max=9999), Output()), _dom_classes=('widget-i…"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
    "source": [
     "def plot_prediction(file_name_id, file_names):\n",
     "\n",
@@ -99,36 +124,31 @@
     "    # Get prediction\n",
     "    image_array = imread(os.path.join(DATA_PATH, 'images', file_name)).reshape(image_shape)\n",
     "    \n",
-    "    test_samples = json.dumps({'data': image_array.tolist()})\n",
+    "    test_samples = json.dumps({'data': np.expand_dims(image_array, axis=0).tolist()})\n",
     "    test_samples = bytes(test_samples, encoding='utf8')\n",
+    "    test_samples = json.dumps([{'name': 'row1', 'image': encode_image(image_array)}])\n",
     "    print(test_samples)\n",
     "    \n",
-    "    result = json.loads(service.run(input_data=test_samples))\n",
-    "    print(result)\n",
-    "    \n",
+    "    # prediction = json.loads(service.run(input_data=test_samples))[0]\n",
+    "    service.run(input_data=test_samples)\n",
+    "\n",
     "    # Plot image\n",
     "    fig = plt.figure(figsize=(5, 5), facecolor='w')\n",
     "    fig.subplots_adjust(wspace=0, hspace=1.2)\n",
     "    ax1 = plt.subplot2grid((1, 1), (0, 0))\n",
-    "#     ax1.set_title('Prediction: {}\\nScore: {} %'.format(np.argmax(prediction), \n",
-    "#                                                      int(prediction[0][np.argmax(prediction)] * 100)), \n",
-    "#                   fontsize=16)\n",
+    "    ax1.set_title('Prediction: {}\\nScore: {} %'.format(np.argmax(prediction), \n",
+    "                                                       int(prediction[np.argmax(prediction)] * 100)), \n",
+    "                  fontsize=16)\n",
     "    ax1.imshow(image_array[:, :, 0], cmap='gray', vmin=0, vmax=255)\n",
     "    ax1.axes.get_xaxis().set_visible(False)\n",
     "    ax1.axes.get_yaxis().set_visible(False)\n",
     "    \n",
     "    plt.show()\n",
-    "\n",
-    "    \n",
-    "def get_prediction(file_name):\n",
-    "    image_array = imread(os.path.join(DATA_PATH, 'images', file_name)).reshape(image_shape)\n",
-    "    if graph_type is 'array':\n",
-    "        return image_array, sess.run(fetches=[prediction], feed_dict={images: [image_array]})[0]\n",
-    "    elif graph_type is 'string':\n",
-    "        image = cv2.imencode('.jpg', image_array)[1].tostring()\n",
-    "        # image = open(os.path.join(DATA_PATH, 'images', file_name), 'rb').read()\n",
-    "        return image_array, sess.run(fetches=[prediction], feed_dict={images: [image]})[0]\n",
     "    \n",
+    "def encode_image(image):\n",
+    "    \"\"\"b64 encode RGB numpy array.\"\"\"\n",
+    "    image_string = cv2.imencode('.jpg', image)[1].tostring()\n",
+    "    return base64.b64encode(image_string).decode(\"utf-8\")\n",
     "\n",
     "# Launch interactive plotting widget\n",
     "_ = interact(\n",
@@ -137,6 +157,34 @@
     "    file_names=fixed(file_names)\n",
     ") # 2634"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 62,
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "NameError",
+     "evalue": "name 'test_samples' is not defined",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
+      "\u001b[0;32m<ipython-input-62-8d436a965226>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtest_samples\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
+      "\u001b[0;31mNameError\u001b[0m: name 'test_samples' is not defined"
+     ]
+    }
+   ],
+   "source": [
+    "test_samples"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {

scoring.py

@@ -6,7 +6,10 @@
 # 3rd party imports
 import os
 import json
+import base64
 import numpy as np
+from PIL import Image
+from io import BytesIO
 import tensorflow as tf
 from azureml.core.model import Model
 
@@ -44,7 +47,17 @@ def init():
 
 
 def run(raw_data):
-    data = np.array(json.loads(raw_data)['data'])
-    out = predictions.eval(session=sess, feed_dict={images: data})
-    y_hat = np.argmax(out, axis=1)
-    return json.dumps(y_hat.tolist())
+    """Run model inference."""
+    # Convert raw data to a numpy array
+    # data = np.array(json.loads(raw_data)['data'])
+
+    # Load raw data
+    data = json.loads(raw_data)
+
+    # Get image arrays
+    inputs = np.array([np.array(Image.open(BytesIO(base64.b64decode(row['image']))), dtype=np.uint8) for row in data])
+
+    # Run model inverse with input data
+    output = predictions.eval(session=sess, feed_dict={images: inputs})
+
+    return json.dumps(output.tolist())