{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Copyright (c) Microsoft Corporation. All rights reserved.\n",
        "\n",
        "Licensed under the MIT License."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "This notebook demonstrates how to run batch scoring job. __[Inception-V3 model](https://arxiv.org/abs/1512.00567)__  and unlabeled images from __[ImageNet](http://image-net.org/)__ dataset will be used. It registers a pretrained inception model in model registry then uses the model to do batch scoring on images in a blob container."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Prerequisites\n",
        "Make sure you go through the [00. Installation and Configuration](./00.configuration.ipynb) Notebook first if you haven't.\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import os\n",
        "from azureml.core import Workspace, Run, Experiment\n",
        "\n",
        "ws = Workspace.from_config()\n",
        "print('Workspace name: ' + ws.name, \n",
        "      'Azure region: ' + ws.location, \n",
        "      'Subscription id: ' + ws.subscription_id, \n",
        "      'Resource group: ' + ws.resource_group, sep = '\\n')\n",
        "\n",
        "# Also create a Project and attach to Workspace\n",
        "scripts_folder = \"scripts\"\n",
        "\n",
        "if not os.path.isdir(scripts_folder):\n",
        "    os.mkdir(scripts_folder)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "from azureml.core.compute import BatchAiCompute, ComputeTarget\n",
        "from azureml.core.datastore import Datastore\n",
        "from azureml.data.data_reference import DataReference\n",
        "from azureml.pipeline.core import Pipeline, PipelineData\n",
        "from azureml.pipeline.steps import PythonScriptStep\n",
        "from azureml.core.runconfig import CondaDependencies, RunConfiguration"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Create and attach Compute targets\n",
        "Use the below code to create and attach Compute targets. "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import os\n",
        "\n",
        "# choose a name for your cluster\n",
        "batchai_cluster_name = os.environ.get(\"BATCHAI_CLUSTER_NAME\", \"gpu-cluster\")\n",
        "cluster_min_nodes = os.environ.get(\"BATCHAI_CLUSTER_MIN_NODES\", 0)\n",
        "cluster_max_nodes = os.environ.get(\"BATCHAI_CLUSTER_MAX_NODES\", 1)\n",
        "vm_size = os.environ.get(\"BATCHAI_CLUSTER_SKU\", \"STANDARD_NC6\")\n",
        "autoscale_enabled = os.environ.get(\"BATCHAI_CLUSTER_AUTOSCALE_ENABLED\", True)\n",
        "\n",
        "\n",
        "if batchai_cluster_name in ws.compute_targets:\n",
        "    compute_target = ws.compute_targets[batchai_cluster_name]\n",
        "    if compute_target and type(compute_target) is BatchAiCompute:\n",
        "        print('found compute target. just use it. ' + batchai_cluster_name)\n",
        "else:\n",
        "    print('creating a new compute target...')\n",
        "    provisioning_config = BatchAiCompute.provisioning_configuration(vm_size = vm_size, # NC6 is GPU-enabled\n",
        "                                                                vm_priority = 'lowpriority', # optional\n",
        "                                                                autoscale_enabled = autoscale_enabled,\n",
        "                                                                cluster_min_nodes = cluster_min_nodes, \n",
        "                                                                cluster_max_nodes = cluster_max_nodes)\n",
        "\n",
        "    # create the cluster\n",
        "    compute_target = ComputeTarget.create(ws, batchai_cluster_name, provisioning_config)\n",
        "    \n",
        "    # can poll for a minimum number of nodes and for a specific timeout. \n",
        "    # if no min node count is provided it will use the scale settings for the cluster\n",
        "    compute_target.wait_for_completion(show_output=True, min_node_count=None, timeout_in_minutes=20)\n",
        "    \n",
        "     # For a more detailed view of current BatchAI cluster status, use the 'status' property    \n",
        "    print(compute_target.status.serialize())"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Python scripts to run"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Python scripts that run the batch scoring. `batchai_score.py` takes input images in `dataset_path`, pretrained models in `model_dir` and outputs a `results-label.txt` to `output_dir`."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "%%writefile $scripts_folder/batchai_score.py\n",
        "import os\n",
        "import argparse\n",
        "import datetime,time\n",
        "import tensorflow as tf\n",
        "from math import ceil\n",
        "import numpy as np\n",
        "import shutil\n",
        "from tensorflow.contrib.slim.python.slim.nets import inception_v3\n",
        "from azureml.core.model import Model\n",
        "\n",
        "slim = tf.contrib.slim\n",
        "\n",
        "parser = argparse.ArgumentParser(description=\"Start a tensorflow model serving\")\n",
        "parser.add_argument('--model_name', dest=\"model_name\", required=True)\n",
        "parser.add_argument('--label_dir', dest=\"label_dir\", required=True)\n",
        "parser.add_argument('--dataset_path', dest=\"dataset_path\", required=True)\n",
        "parser.add_argument('--output_dir', dest=\"output_dir\", required=True)\n",
        "parser.add_argument('--batch_size', dest=\"batch_size\", type=int, required=True)\n",
        "\n",
        "args = parser.parse_args()\n",
        "\n",
        "image_size = 299\n",
        "num_channel = 3\n",
        "\n",
        "# create output directory if it does not exist\n",
        "os.makedirs(args.output_dir, exist_ok=True)\n",
        "\n",
        "def get_class_label_dict(label_file):\n",
        "  label = []\n",
        "  proto_as_ascii_lines = tf.gfile.GFile(label_file).readlines()\n",
        "  for l in proto_as_ascii_lines:\n",
        "    label.append(l.rstrip())\n",
        "  return label\n",
        "\n",
        "\n",
        "class DataIterator:\n",
        "    def __init__(self, data_dir):\n",
        "        self.file_paths = []\n",
        "        image_list = os.listdir(data_dir)\n",
        "        total_size = len(image_list)\n",
        "        self.file_paths = [data_dir + '/' + file_name.rstrip() for file_name in image_list ]\n",
        "\n",
        "        self.labels = [1 for file_name in self.file_paths]\n",
        "\n",
        "    @property\n",
        "    def size(self):\n",
        "        return len(self.labels)\n",
        "\n",
        "    def input_pipeline(self, batch_size):\n",
        "        images_tensor = tf.convert_to_tensor(self.file_paths, dtype=tf.string)\n",
        "        labels_tensor = tf.convert_to_tensor(self.labels, dtype=tf.int64)\n",
        "        input_queue = tf.train.slice_input_producer([images_tensor, labels_tensor], shuffle=False)\n",
        "        labels = input_queue[1]\n",
        "        images_content = tf.read_file(input_queue[0])\n",
        "\n",
        "        image_reader = tf.image.decode_jpeg(images_content, channels=num_channel, name=\"jpeg_reader\")\n",
        "        float_caster = tf.cast(image_reader, tf.float32)\n",
        "        new_size = tf.constant([image_size, image_size], dtype=tf.int32)\n",
        "        images = tf.image.resize_images(float_caster, new_size)\n",
        "        images = tf.divide(tf.subtract(images, [0]), [255])\n",
        "\n",
        "        image_batch, label_batch = tf.train.batch([images, labels], batch_size=batch_size, capacity=5 * batch_size)\n",
        "        return image_batch\n",
        "\n",
        "def main(_):\n",
        "    start_time = datetime.datetime.now()\n",
        "    label_file_name = os.path.join(args.label_dir, \"labels.txt\")\n",
        "    label_dict = get_class_label_dict(label_file_name)\n",
        "    classes_num = len(label_dict)\n",
        "    test_feeder = DataIterator(data_dir=args.dataset_path)\n",
        "    total_size = len(test_feeder.labels)\n",
        "    count = 0\n",
        "    # get model from model registry\n",
        "    model_path = Model.get_model_path(args.model_name)\n",
        "    with tf.Session() as sess:\n",
        "        test_images = test_feeder.input_pipeline(batch_size=args.batch_size)\n",
        "        with slim.arg_scope(inception_v3.inception_v3_arg_scope()):\n",
        "            input_images = tf.placeholder(tf.float32, [args.batch_size, image_size, image_size, num_channel])\n",
        "            logits, _ = inception_v3.inception_v3(input_images,\n",
        "                                                        num_classes=classes_num,\n",
        "                                                        is_training=False)\n",
        "            probabilities = tf.argmax(logits, 1)\n",
        "\n",
        "        sess.run(tf.global_variables_initializer())\n",
        "        sess.run(tf.local_variables_initializer())\n",
        "        coord = tf.train.Coordinator()\n",
        "        threads = tf.train.start_queue_runners(sess=sess, coord=coord)\n",
        "        saver = tf.train.Saver()\n",
        "        saver.restore(sess, model_path)\n",
        "        out_filename = os.path.join(args.output_dir, \"result-labels.txt\")\n",
        "        with open(out_filename, \"w\") as result_file:\n",
        "            i = 0\n",
        "            while count < total_size and not coord.should_stop():\n",
        "                test_images_batch = sess.run(test_images)\n",
        "                file_names_batch = test_feeder.file_paths[i*args.batch_size: min(test_feeder.size, (i+1)*args.batch_size)]\n",
        "                results = sess.run(probabilities, feed_dict={input_images: test_images_batch})\n",
        "                new_add = min(args.batch_size, total_size-count)\n",
        "                count += new_add\n",
        "                i += 1\n",
        "                for j in range(new_add):\n",
        "                    result_file.write(os.path.basename(file_names_batch[j]) + \": \" + label_dict[results[j]] + \"\\n\")\n",
        "                result_file.flush()\n",
        "            coord.request_stop()\n",
        "            coord.join(threads)\n",
        "            \n",
        "        # copy the file to artifacts\n",
        "        shutil.copy(out_filename, \"./outputs/\")\n",
        "        # Move the processed data out of the blob so that the next run can process the data.\n",
        "\n",
        "if __name__ == \"__main__\":\n",
        "    tf.app.run()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Prepare Model and Input data"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "### Download Model\n",
        "\n",
        "Download and extract model from http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz to `\"models\"`"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "# create directory for model\n",
        "model_dir = 'models'\n",
        "if not os.path.isdir(model_dir):\n",
        "    os.mkdir(model_dir)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import tarfile\n",
        "import urllib.request\n",
        "\n",
        "url=\"http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz\"\n",
        "response = urllib.request.urlretrieve(url, \"model.tar.gz\")\n",
        "tar = tarfile.open(\"model.tar.gz\", \"r:gz\")\n",
        "tar.extractall(model_dir)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "### Create a datastore that points to blob container containing sample images\n",
        "\n",
        "We have created a public blob container `sampledata` on an account named `pipelinedata` containing images from ImageNet evaluation set. In the next step, we create a datastore with name `images_datastore` that points to this container. The `overwrite=True` step overwrites any datastore that was created previously with that name. \n",
        "\n",
        "This step can be changed to point to your blob container by providing an additional `account_key` parameter with `account_name`. "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "account_name = \"pipelinedata\"\n",
        "sample_data = Datastore.register_azure_blob_container(ws, datastore_name=\"images_datastore\", container_name=\"sampledata\", \n",
        "                                                        account_name=account_name, \n",
        "                                                        overwrite=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Output datastore"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "We write the outputs to the default datastore"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "default_ds = ws.get_default_datastore()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Specify where the data is stored or will be written to"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "from azureml.core.conda_dependencies import CondaDependencies\n",
        "from azureml.data.data_reference import DataReference\n",
        "from azureml.pipeline.core import Pipeline, PipelineData\n",
        "from azureml.core import Datastore\n",
        "from azureml.core import Experiment"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "input_images = DataReference(datastore=sample_data, \n",
        "                             data_reference_name=\"input_images\",\n",
        "                             path_on_datastore=\"batchscoring/images\",\n",
        "                             mode=\"download\"\n",
        "                            )\n",
        "model_dir = DataReference(datastore=sample_data, \n",
        "                          data_reference_name=\"input_model\",\n",
        "                          path_on_datastore=\"batchscoring/models\",\n",
        "                          mode=\"download\"                          \n",
        "                         )\n",
        "label_dir = DataReference(datastore=sample_data, \n",
        "                          data_reference_name=\"input_labels\",\n",
        "                          path_on_datastore=\"batchscoring/labels\",\n",
        "                          mode=\"download\"                          \n",
        "                         )\n",
        "output_dir = PipelineData(name=\"scores\", \n",
        "                          datastore=default_ds, \n",
        "                          output_path_on_compute=\"batchscoring/results\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Register the model with Workspace"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import shutil\n",
        "from azureml.core.model import Model\n",
        "\n",
        "# register downloaded model \n",
        "model = Model.register(model_path = \"models/inception_v3.ckpt\",\n",
        "                       model_name = \"inception\", # this is the name the model is registered as\n",
        "                       tags = {'pretrained': \"inception\"},\n",
        "                       description = \"Imagenet trained tensorflow inception\",\n",
        "                       workspace = ws)\n",
        "# remove the downloaded dir after registration if you wish\n",
        "shutil.rmtree(\"models\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Specify environment to run the script"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "cd = CondaDependencies.create(pip_packages=[\"tensorflow-gpu==1.4.0\", \"azureml-defaults\"])\n",
        "\n",
        "# Runconfig\n",
        "batchai_run_config = RunConfiguration(conda_dependencies=cd)\n",
        "batchai_run_config.environment.docker.enabled = True\n",
        "batchai_run_config.environment.docker.gpu_support = True\n",
        "batchai_run_config.environment.docker.base_image = \"microsoft/mmlspark:gpu-0.12\"\n",
        "batchai_run_config.environment.spark.precache_packages = False"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Steps to run"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "A subset of the parameters to the python script can be given as input when we re-run a `PublishedPipeline`. In the current example, we define `batch_size` taken by the script as such parameter."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "from azureml.pipeline.core.graph import PipelineParameter\n",
        "batch_size_param = PipelineParameter(name=\"param_batch_size\", default_value=20)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "inception_model_name = \"inception_v3.ckpt\"\n",
        "\n",
        "batch_score_step = PythonScriptStep(\n",
        "    name=\"batch ai scoring\",\n",
        "    script_name=\"batchai_score.py\",\n",
        "    arguments=[\"--dataset_path\", input_images, \n",
        "               \"--model_name\", \"inception\",\n",
        "               \"--label_dir\", label_dir, \n",
        "               \"--output_dir\", output_dir, \n",
        "               \"--batch_size\", batch_size_param],\n",
        "    target=compute_target,\n",
        "    inputs=[input_images, label_dir],\n",
        "    outputs=[output_dir],\n",
        "    runconfig=batchai_run_config,\n",
        "    source_directory=scripts_folder\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "pipeline = Pipeline(workspace=ws, steps=[batch_score_step])\n",
        "pipeline_run = Experiment(ws, 'batch_scoring').submit(pipeline, pipeline_params={\"param_batch_size\": 20})"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Monitor run"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "from azureml.train.widgets import RunDetails\n",
        "RunDetails(pipeline_run).show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "pipeline_run.wait_for_completion(show_output=True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Download and review output"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "step_run = list(pipeline_run.get_children())[0]\n",
        "step_run.download_file(\"./outputs/result-labels.txt\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import pandas as pd\n",
        "df = pd.read_csv(\"result-labels.txt\", delimiter=\":\", header=None)\n",
        "df.columns = [\"Filename\", \"Prediction\"]\n",
        "df.head()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Publish a pipeline and rerun using a REST call"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Create a published pipeline"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "published_pipeline = pipeline_run.publish_pipeline(\n",
        "    name=\"Inception v3 scoring\", description=\"Batch scoring using Inception v3 model\", version=\"1.0\")\n",
        "\n",
        "published_id = published_pipeline.id"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Rerun using REST call"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Get AAD token"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "from azureml.core.authentication import AzureCliAuthentication\n",
        "import requests\n",
        "\n",
        "cli_auth = AzureCliAuthentication()\n",
        "aad_token = cli_auth.get_authentication_header()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Run published pipeline using its REST endpoint"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "from azureml.pipeline.core import PublishedPipeline\n",
        "\n",
        "rest_endpoint = published_pipeline.endpoint\n",
        "# specify batch size when running the pipeline\n",
        "response = requests.post(rest_endpoint, \n",
        "                         headers=aad_token, \n",
        "                         json={\"ExperimentName\": \"batch_scoring\",\n",
        "                               \"ParameterAssignments\": {\"param_batch_size\": 50}})\n",
        "run_id = response.json()[\"Id\"]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Monitor the new run"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "from azureml.pipeline.core.run import PipelineRun\n",
        "published_pipeline_run = PipelineRun(ws.experiments[\"batch_scoring\"], run_id)\n",
        "\n",
        "RunDetails(published_pipeline_run).show()"
      ]
    }
  ],
  "metadata": {
    "authors": [
      {
        "name": "hichando"
      }
    ],
    "kernelspec": {
      "display_name": "Python 3.6",
      "language": "python",
      "name": "python36"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.6"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 2
}