4. SAP FedML/4.1 HR Use Case/notebooks/epp-end-to-end-with-fedml.ipynb

{"cells": [{"cell_type": "markdown", "id": "08ea8c44-ee92-4cfd-91c7-c55a639d693b", "metadata": {}, "source": "# Train and deploy a simple employee promotion prediction model\nThis notebook demonstrates a new approach with [SAP Federated ML Python libraries (FedML)](https://pypi.org/project/fedml-dsp/) for an end-to-end workflow in a predictive analytics use case, in which you train a classification model with data from [SAP Datasphere](https://help.sap.com/docs/SAP_DATASPHERE), deploy the model to [SAP AI Core](https://help.sap.com/docs/sap-ai-core), and use it for online inference. Comparing with other similar notebooks in this GitHub repository without SAP FedML, far less coding is needed to achieved the same results with FedML, which leverages SAP Datasphere's Data Federation architecture to intelligently source both SAP and non-SAP data for machine learning experiments. We developed and tested this notebook in [IBM\u00ae watsonx.ai\u2122 AI studio](https://www.ibm.com/products/watsonx-ai), but you should be able to run it in any Jupyter notebook environment you prefer with little or no code change.\n\nIn this notebook, you will learn:\n\n- How to access datasets stored in SAP Datasphere using FedML\n- How to train a Random Forest Classification model using Scikit-Learn\n- How to serve the model in SAP AI Core using FedML\n- How to make inference calls using FedML\n\n###  More about the use case\nThis notebook demonstrates the implementation of a Random Forest Classification model for predicting employee promotions. Random Forest is an ensemble learning method that constructs a multitude of decision trees during training and outputs the mode of the classes (classification) or mean prediction (regression) of the individual trees for robust and accurate predictions. To help understand and implement the Random Forest algorithm for predicting employee promotions, this notebook covers the process of extracting relevant features, training the Random Forest model, and evaluating its performance in predictions. Employee promotion prediction involves identifying factors that contribute to promotion decisions based on various features. The focus is on leveraging model training with employee historical data for accurate predictions."}, {"cell_type": "markdown", "id": "fbeebd8b", "metadata": {}, "source": "## Table of Contents\n1.  [Step 1: Connect to SAP Datasphere](#connect_dsp)\n\n1.  [Step 2: Load and prepare the data](#load_data)\n\n1.  [Step 3: Balance the dataset](#balance_dataset)\n\n1.  [Step 4: Data Preprocessing and Train-Test split](#process_data)\n\n1.  [Step 5: Train the Random forest classifier model](#train_model)\n\n1.  [Step 6: Model Evaluation](#evaluate_data)\n\n1.  [Step 7: Save the model as pickle](#save_model)\n\n1.  [Step 8: Deploy the model to SAP AI Core](#deploy_to_aicore)\n\n1.  [Step 9: Make predictions](#make_predictions)\n"}, {"cell_type": "markdown", "id": "e8787109", "metadata": {}, "source": "### Credentials\n\nYou need to provide the credentials required for accessing SAP Datasphere, SAP AI Core, GitHub repository, Docker registry, and IBM Cloud Object Storage (COS). Details about their parameters can be found in the documentation:\n- https://github.com/SAP-samples/datasphere-fedml/blob/main/Datasphere/fedml-dsp.md\n- https://cloud.ibm.com/docs/cloud-object-storage?topic=cloud-object-storage-python\n\n```\nfedml_config = {\n    \"address\":  <The IP address or host name of the database instance. Required. String>,\n    \"port\": <The port number of the database instance. Required>,\n    \"user\": <The database user. Required>,\n    \"password\": <The database user's password. Required>,\n    \"schema\": <The SAP Datasphere cloud Space Schema. Optional>,\n    \"encrypt\": <\"true\" . Denotes an encrypted connection>,\n    \"sslValidateCertificate\": <\"false\" . Specifies whether to validate the server's certificate>,\n    \"disableCloudRedirect\": < \"true\". Specifies if there should be a tenant redirection for a cloud instance,\n    \"communicationTimeout\": <\"0\". Value of 0 Disables any communication Timeouts>,\n    \"autocommit\": <\"true\". Sets auto commit to true for the database connection>,\n    \"sslUseDefaultTrustStore\": <\"true\". Denotes the use of client's default trust store>\n}\n\naicore_service_key = {\n    \"serviceurls\": {\n        \"AI_API_URL\": \"https://api.ai.xxx.yyy.zzz.ml.hana.ondemand.com\"\n    },\n    \"appname\": \"xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx!xxxxxxx|aicore!xxxx\",\n    \"clientid\": \"sb-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx!xxxxxxx|aicore!xxxx\",\n    \"clientsecret\": \"xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx$abcdefgh1234567890abcdefgh1234567890abcdefgh',\n    \"identityzone\": \"xyz123\",\n    \"identityzoneid\": \"xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx\",\n    \"url\": \"https://xyz123.authentication.us10.hana.ondemand.com\"\n}\n\ngithub_info = {\n    \"name\": \"<name>\",\n    \"url\": \"<url>\",\n    \"username\": \"<username>\",\n    \"password\": \"<password>\"\n}\n\ndocker_registry_secret = {\n    \"name\": \"<name_of_secret>\",\n    \"data\": {\n        \".dockerconfigjson\": \"{\\\"auths\\\":{\\\"https://index.docker.io\\\":{\\\"username\\\":\\\"<username>\\\",\\\"password\\\":\\\"<password>\\\"}}}\"\n    }\n}\n\nobject_storage_credential = {\n    'access_key_id' = '<your_access_key_id>',\n    'secret_access_key' = '<your_secret_access_key>',\n    'bucket' = '<your_bucket>',\n    'endpoint' = '<host>',\n    'region' = '<region>'\n}\n\ncos_credential = {\n    'COS_INSTANCE_ID': '<cos_instance_crn>',\n    'COS_API_KEY': '<cos_api_key>',\n    'COS_ENDPOINT': 'https://s3.<region>.cloud-object-storage.appdomain.cloud',\n    'IBM_AUTH_ENDPOINT': 'https://iam.cloud.ibm.com/oidc/token',\n    'BUCKET': '<bucket>',\n}\n```"}, {"cell_type": "code", "execution_count": 1, "id": "0f6e256f", "metadata": {}, "outputs": [], "source": "# The code was removed by Watson Studio for sharing."}, {"cell_type": "code", "execution_count": 2, "id": "a8b43b5d", "metadata": {}, "outputs": [], "source": "import os, types\nimport time\nimport json\nimport pandas as pd\nimport seaborn as sns\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import accuracy_score, classification_report, f1_score, recall_score, precision_score, roc_auc_score, roc_curve, auc\nimport matplotlib.pyplot as plt\nimport requests"}, {"cell_type": "code", "execution_count": 3, "id": "7cc5973b", "metadata": {}, "outputs": [], "source": "# Save fedml_config to a JSON file\nwith open(\"fedml_config.json\", \"w\") as outfile: \n    json.dump(fedml_config, outfile, indent = 4)\n\n# Save aicore_service_key to a JSON file\nwith open(\"aicore_service_key.json\", \"w\") as outfile: \n    json.dump(aicore_service_key, outfile, indent = 4)\n\n# Save github_info to a JSON file\nwith open(\"github_info.json\", \"w\") as outfile: \n    json.dump(github_info, outfile, indent = 4)\n\n# Save docker_registry_secret to a JSON file\nwith open(\"docker_registry_secret.json\", \"w\") as outfile: \n    json.dump(docker_registry_secret, outfile, indent = 4)"}, {"cell_type": "markdown", "id": "40c3c391", "metadata": {}, "source": "<a id=\"load_data\"></a>\n## Step 1: Connect to SAP Datasphere"}, {"cell_type": "markdown", "id": "189d2a0d", "metadata": {}, "source": "Install SAP FedML library."}, {"cell_type": "code", "execution_count": 4, "id": "674bc18d", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "No Nvidia GPU in system!\n"}], "source": "try:\n    import fedml_dsp\n    from fedml_dsp import DbConnection\nexcept:\n    %pip install -U fedml-dsp\n    import fedml_dsp\n    from fedml_dsp import DbConnection"}, {"cell_type": "markdown", "id": "b9147c1e", "metadata": {}, "source": "Create a connection to SAP Datasphere."}, {"cell_type": "code", "execution_count": 5, "id": "09b8a1d6", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "2024-04-18 04:43:26,835: fedml_dsp.logger INFO: Attempting DSP check.\n2024-04-18 04:43:27,025: fedml_dsp.logger INFO: DSP check completed. Connected to DSP.\n"}], "source": "db = DbConnection(url='fedml_config.json')"}, {"cell_type": "markdown", "id": "bccc3461", "metadata": {}, "source": "Seach for the training data stored in Datasphere, which is exposed as a view named \"epp_train_view\"."}, {"cell_type": "code", "execution_count": 6, "id": "cb831721", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "2024-04-18 04:43:27,038: fedml_dsp.logger INFO: Searching for views with name 'epp_train_view' in schema 'WATSONX'\n"}, {"data": {"text/plain": "([('epp_train_view', 'ROW')], ['VIEW_NAME', 'VIEW_TYPE'])"}, "execution_count": 6, "metadata": {}, "output_type": "execute_result"}], "source": "db.get_view_by_name(\"epp_train_view\")"}, {"cell_type": "markdown", "id": "6121c3f5", "metadata": {}, "source": "<a id=\"load_data\"></a>\n## Step 2: Load and prepare the data"}, {"cell_type": "markdown", "id": "5b63355d", "metadata": {}, "source": "Fetch the the training data from SAP Datasphere, and save it in a pandas DataFrame."}, {"cell_type": "code", "execution_count": 7, "id": "3aa87c97", "metadata": {}, "outputs": [{"data": {"text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>employee_id</th>\n      <th>department</th>\n      <th>region</th>\n      <th>education</th>\n      <th>gender</th>\n      <th>recruitment_channel</th>\n      <th>no_of_trainings</th>\n      <th>age</th>\n      <th>previous_year_rating</th>\n      <th>length_of_service</th>\n      <th>any_awards_won</th>\n      <th>is_promoted</th>\n      <th>kpis_met_above_80_percent</th>\n      <th>avg_training_score</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>45709</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_31</td>\n      <td>Bachelor's</td>\n      <td>f</td>\n      <td>other</td>\n      <td>1</td>\n      <td>29</td>\n      <td>None</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>49</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>66874</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_27</td>\n      <td>Bachelor's</td>\n      <td>f</td>\n      <td>other</td>\n      <td>1</td>\n      <td>30</td>\n      <td>None</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>50</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>36904</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_15</td>\n      <td>Bachelor's</td>\n      <td>m</td>\n      <td>other</td>\n      <td>1</td>\n      <td>29</td>\n      <td>3</td>\n      <td>2</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>51</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>32877</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_2</td>\n      <td>Bachelor's</td>\n      <td>f</td>\n      <td>other</td>\n      <td>1</td>\n      <td>40</td>\n      <td>3</td>\n      <td>12</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>50</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>58415</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_7</td>\n      <td>Bachelor's</td>\n      <td>m</td>\n      <td>other</td>\n      <td>1</td>\n      <td>45</td>\n      <td>4</td>\n      <td>5</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>50</td>\n    </tr>\n  </tbody>\n</table>\n</div>", "text/plain": "   employee_id         department     region   education gender  \\\n0        45709  Sales & Marketing  region_31  Bachelor's      f   \n1        66874  Sales & Marketing  region_27  Bachelor's      f   \n2        36904  Sales & Marketing  region_15  Bachelor's      m   \n3        32877  Sales & Marketing   region_2  Bachelor's      f   \n4        58415  Sales & Marketing   region_7  Bachelor's      m   \n\n  recruitment_channel  no_of_trainings  age previous_year_rating  \\\n0               other                1   29                 None   \n1               other                1   30                 None   \n2               other                1   29                    3   \n3               other                1   40                    3   \n4               other                1   45                    4   \n\n   length_of_service  any_awards_won  is_promoted  kpis_met_above_80_percent  \\\n0                  1               0            0                          0   \n1                  1               0            0                          0   \n2                  2               0            0                          0   \n3                 12               0            0                          0   \n4                  5               0            0                          0   \n\n   avg_training_score  \n0                  49  \n1                  50  \n2                  51  \n3                  50  \n4                  50  "}, "execution_count": 7, "metadata": {}, "output_type": "execute_result"}], "source": "data = db.get_data_with_headers(\"epp_train_view\")\nemployee_data = pd.DataFrame(data[0], columns=data[1])\nemployee_data.head()"}, {"cell_type": "code", "execution_count": 8, "id": "e1197684", "metadata": {}, "outputs": [{"data": {"text/plain": "employee_id                   int64\ndepartment                   object\nregion                       object\neducation                    object\ngender                       object\nrecruitment_channel          object\nno_of_trainings               int64\nage                           int64\nprevious_year_rating         object\nlength_of_service             int64\nany_awards_won                int64\nis_promoted                   int64\nkpis_met_above_80_percent     int64\navg_training_score            int64\ndtype: object"}, "execution_count": 8, "metadata": {}, "output_type": "execute_result"}], "source": "employee_data.dtypes"}, {"cell_type": "code", "execution_count": 9, "id": "0d3c07c0", "metadata": {}, "outputs": [{"data": {"text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>employee_id</th>\n      <th>department</th>\n      <th>region</th>\n      <th>education</th>\n      <th>gender</th>\n      <th>recruitment_channel</th>\n      <th>no_of_trainings</th>\n      <th>age</th>\n      <th>previous_year_rating</th>\n      <th>length_of_service</th>\n      <th>kpis_met_above_80_percent</th>\n      <th>any_awards_won</th>\n      <th>avg_training_score</th>\n      <th>is_promoted</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>45709</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_31</td>\n      <td>Bachelor's</td>\n      <td>f</td>\n      <td>other</td>\n      <td>1</td>\n      <td>29</td>\n      <td>NaN</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>49</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>66874</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_27</td>\n      <td>Bachelor's</td>\n      <td>f</td>\n      <td>other</td>\n      <td>1</td>\n      <td>30</td>\n      <td>NaN</td>\n      <td>1</td>\n      <td>0</td>\n      <td>0</td>\n      <td>50</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>36904</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_15</td>\n      <td>Bachelor's</td>\n      <td>m</td>\n      <td>other</td>\n      <td>1</td>\n      <td>29</td>\n      <td>3.0</td>\n      <td>2</td>\n      <td>0</td>\n      <td>0</td>\n      <td>51</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>32877</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_2</td>\n      <td>Bachelor's</td>\n      <td>f</td>\n      <td>other</td>\n      <td>1</td>\n      <td>40</td>\n      <td>3.0</td>\n      <td>12</td>\n      <td>0</td>\n      <td>0</td>\n      <td>50</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>58415</td>\n      <td>Sales &amp; Marketing</td>\n      <td>region_7</td>\n      <td>Bachelor's</td>\n      <td>m</td>\n      <td>other</td>\n      <td>1</td>\n      <td>45</td>\n      <td>4.0</td>\n      <td>5</td>\n      <td>0</td>\n      <td>0</td>\n      <td>50</td>\n      <td>0</td>\n    </tr>\n  </tbody>\n</table>\n</div>", "text/plain": "   employee_id         department     region   education gender  \\\n0        45709  Sales & Marketing  region_31  Bachelor's      f   \n1        66874  Sales & Marketing  region_27  Bachelor's      f   \n2        36904  Sales & Marketing  region_15  Bachelor's      m   \n3        32877  Sales & Marketing   region_2  Bachelor's      f   \n4        58415  Sales & Marketing   region_7  Bachelor's      m   \n\n  recruitment_channel  no_of_trainings  age  previous_year_rating  \\\n0               other                1   29                   NaN   \n1               other                1   30                   NaN   \n2               other                1   29                   3.0   \n3               other                1   40                   3.0   \n4               other                1   45                   4.0   \n\n   length_of_service  kpis_met_above_80_percent  any_awards_won  \\\n0                  1                          0               0   \n1                  1                          0               0   \n2                  2                          0               0   \n3                 12                          0               0   \n4                  5                          0               0   \n\n   avg_training_score  is_promoted  \n0                  49            0  \n1                  50            0  \n2                  51            0  \n3                  50            0  \n4                  50            0  "}, "execution_count": 9, "metadata": {}, "output_type": "execute_result"}], "source": "# Change dtype\nemployee_data['previous_year_rating'] = pd.to_numeric(employee_data['previous_year_rating'])\n\n# Change the order of the columns\nemployee_data = employee_data[['employee_id', 'department', 'region', 'education', 'gender',\n       'recruitment_channel', 'no_of_trainings', 'age', 'previous_year_rating',\n       'length_of_service', 'kpis_met_above_80_percent', 'any_awards_won',\n       'avg_training_score', 'is_promoted']]\n\nemployee_data.head()"}, {"cell_type": "code", "execution_count": 10, "id": "ebfc5df2", "metadata": {}, "outputs": [{"data": {"text/plain": "employee_id                    int64\ndepartment                    object\nregion                        object\neducation                     object\ngender                        object\nrecruitment_channel           object\nno_of_trainings                int64\nage                            int64\nprevious_year_rating         float64\nlength_of_service              int64\nkpis_met_above_80_percent      int64\nany_awards_won                 int64\navg_training_score             int64\nis_promoted                    int64\ndtype: object"}, "execution_count": 10, "metadata": {}, "output_type": "execute_result"}], "source": "employee_data.dtypes"}, {"cell_type": "code", "execution_count": 11, "id": "e8ae46ef", "metadata": {}, "outputs": [{"data": {"text/plain": "Index(['employee_id', 'department', 'region', 'education', 'gender',\n       'recruitment_channel', 'no_of_trainings', 'age', 'previous_year_rating',\n       'length_of_service', 'kpis_met_above_80_percent', 'any_awards_won',\n       'avg_training_score', 'is_promoted'],\n      dtype='object')"}, "execution_count": 11, "metadata": {}, "output_type": "execute_result"}], "source": "employee_data.columns"}, {"cell_type": "code", "execution_count": 12, "id": "73ccaafe", "metadata": {}, "outputs": [], "source": "employee_data = employee_data.drop(columns=[\"employee_id\", \"recruitment_channel\", \"region\"])"}, {"cell_type": "code", "execution_count": 13, "id": "434c2fd9", "metadata": {}, "outputs": [{"data": {"text/plain": "(54808, 11)"}, "execution_count": 13, "metadata": {}, "output_type": "execute_result"}], "source": "employee_data.shape"}, {"cell_type": "markdown", "id": "4564b675", "metadata": {}, "source": "<a id=\"balance_dataset\"></a>\n## Step 3: Balance the dataset\n\nBalancing the dataset with respect to the \"is_promoted\" variable is essential to avoid bias in the predictive model. When the dataset is imbalanced, with one class significantly outnumbering the other (e.g., promoted or not promoted employees), the model may exhibit a tendency to favor the majority class. In the context of employee promotion prediction, imbalance could lead to inaccurate predictions, especially if the majority of instances involve non-promotion."}, {"cell_type": "code", "execution_count": 14, "id": "807bb6e6", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "0    50140\n1     4668\nName: is_promoted, dtype: int64\n0    4668\n1    4668\nName: is_promoted, dtype: int64\n"}], "source": "# Separate the two classes\nn = employee_data[employee_data['is_promoted'] == 1].count()[0]\nclass_0_data = employee_data[employee_data['is_promoted'] == 0]\nclass_1_data = employee_data[employee_data['is_promoted'] == 1]\nprint(employee_data['is_promoted'].value_counts())\n# Take 3760 samples from class 0\nclass_0_sampled = class_0_data.sample(n, random_state=42)\n\n# Combine the two classes to create a balanced dataset\nemployee_data = pd.concat([class_0_sampled, class_1_data])\nprint(employee_data['is_promoted'].value_counts())"}, {"cell_type": "code", "execution_count": 15, "id": "ca9347e6", "metadata": {}, "outputs": [], "source": "employee_data.reset_index(drop=True,inplace=True)"}, {"cell_type": "markdown", "id": "7365c8ad", "metadata": {}, "source": "<a id=\"process_data\"></a>\n## Step 4: Data Preprocessing and Train-Test split\nData preprocessing, including handling missing values, encoding categorical columns, and splitting the dataset into training and test sets, with a check on the class distribution in the training data."}, {"cell_type": "code", "execution_count": 16, "id": "d14eba59-ef8a-4a2b-a3f9-be9499ae8ba3", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "Class Distribution in Training Data:\n0    3737\n1    3731\nName: is_promoted, dtype: int64\n"}], "source": "# Handle missing values\nemployee_data[\"education\"].fillna(employee_data[\"education\"].mode()[0], inplace=True)\nemployee_data[\"previous_year_rating\"].fillna(1, inplace=True)\n\n# Encode categorical columns\ncategorical_columns = employee_data.select_dtypes(include=['object']).columns.tolist()\nX_encoded = pd.get_dummies(employee_data, columns=categorical_columns, drop_first=True)\n\n# Split the data into features (X) and target (y)\ny = X_encoded[\"is_promoted\"]\nX_encoded = X_encoded.drop(\"is_promoted\", axis=1)\n\n# Splitting the data into training and test sets\nX_train, X_test, y_train, y_test = train_test_split(X_encoded, y, test_size=0.2, random_state=42)\n\n# Initializing a Random Forest Classifier\n# rf_classifier = RandomForestClassifier(n_estimators=100, random_state=42)\n# Check the class distribution in the training set\nclass_counts = y_train.value_counts()\nprint(\"Class Distribution in Training Data:\")\nprint(class_counts)"}, {"cell_type": "markdown", "id": "6743cf0b", "metadata": {}, "source": "<a id=\"train_model\"></a>\n## Step 5: Train the Random Forest Classifier model\nInitialization, training, and testing of a Random Forest Classifier with balanced class weights for predicting promotions in an employee dataset."}, {"cell_type": "code", "execution_count": 17, "id": "4bebb57d", "metadata": {}, "outputs": [], "source": "# Initialize a Random Forest Classifier with balanced class weights\nrf_classifier = RandomForestClassifier(n_estimators=100, random_state=100)\n\n# Training the model\nrf_classifier.fit(X_train, y_train)\n\n# Testing data prediction\ny_pred = rf_classifier.predict(X_test)\n\n# Traing data prediction\nyt_pred = rf_classifier.predict(X_train)"}, {"cell_type": "code", "execution_count": 18, "id": "af85e816", "metadata": {}, "outputs": [{"data": {"text/plain": "Index(['no_of_trainings', 'age', 'previous_year_rating', 'length_of_service',\n       'kpis_met_above_80_percent', 'any_awards_won', 'avg_training_score',\n       'department_Finance', 'department_HR', 'department_Legal',\n       'department_Operations', 'department_Procurement', 'department_R&D',\n       'department_Sales & Marketing', 'department_Technology',\n       'education_Bachelor's', 'education_Below Secondary',\n       'education_Master's & above', 'gender_m'],\n      dtype='object')"}, "execution_count": 18, "metadata": {}, "output_type": "execute_result"}], "source": "# List of feature columns \nX_encoded.columns"}, {"cell_type": "markdown", "id": "5bcf2209", "metadata": {}, "source": "<a id=\"evaluate_data\"></a>\n## Step 6: Model Evaluation\nEvaluating the training and testing accuracy scores of a Random Forest Classifier for employee promotion prediction."}, {"cell_type": "code", "execution_count": 19, "id": "4af14343", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "train_accuracy 0.9965184788430638\ntest_accuracy 0.8094218415417559\n"}], "source": "# Calculating accuracy\ntrain_accuracy = accuracy_score(y_train, yt_pred)\n\nprint(\"train_accuracy\",train_accuracy)\n\ntest_accuracy = accuracy_score(y_test, y_pred)\n\nprint(\"test_accuracy\",test_accuracy)"}, {"cell_type": "code", "execution_count": 20, "id": "d0e6be2f", "metadata": {}, "outputs": [], "source": "# Testing data \npred_df={\n    'no_of_trainings': {0: 1, 1: 1, 2: 1, 3: 2, 4: 1, 5: 2, 6: 1, 7: 1, 8: 1, 9: 1},\n    'age': {0: 35, 1: 30, 2: 34, 3: 39, 4: 45, 5: 31, 6: 31, 7: 33, 8: 28, 9: 32},\n    'previous_year_rating': {0: 5.0, 1: 5.0, 2: 3.0, 3: 1.0, 4: 3.0, 5: 3.0, 6: 3.0, 7: 3.0, 8: 4.0, 9: 5.0},\n    'length_of_service': {0: 8, 1: 4, 2: 7, 3: 10, 4: 2, 5: 7, 6: 5, 7: 6, 8: 5, 9: 5},\n    'kpis_met_above_80_percent': {0: 1, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 1},\n    'any_awards_won': {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    'avg_training_score': {0: 49, 1: 60, 2: 50, 3: 50, 4: 73, 5: 85, 6: 59, 7: 63, 8: 83, 9: 54},\n    'department_Finance': {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    'department_HR': {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    'department_Legal': {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    'department_Operations': {0: 0, 1: 1, 2: 0, 3: 0, 4: 0, 5: 0, 6: 1, 7: 1, 8: 0, 9: 0},\n    'department_Procurement': {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    'department_R&D': {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    'department_Sales & Marketing': {0: 1, 1: 0, 2: 1, 3: 1, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 1},\n    'department_Technology': {0: 0, 1: 0, 2: 0, 3: 0, 4: 1, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    \"education_Bachelor's\": {0: 0, 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 1, 7: 0, 8: 1, 9: 0},\n    'education_Below Secondary': {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 0, 8: 0, 9: 0},\n    \"education_Master's & above\": {0: 1, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0, 7: 1, 8: 0, 9: 1},\n    'gender_m': {0: 0, 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 0, 7: 1, 8: 1, 9: 1}\n}\npred_df=pd.DataFrame(pred_df)"}, {"cell_type": "code", "execution_count": 21, "id": "a8573589", "metadata": {}, "outputs": [{"data": {"text/plain": "array([[0.18, 0.82],\n       [0.96, 0.04],\n       [1.  , 0.  ],\n       [0.99, 0.01],\n       [0.92, 0.08],\n       [0.74, 0.26],\n       [0.96, 0.04],\n       [0.91, 0.09],\n       [0.79, 0.21],\n       [0.57, 0.43]])"}, "execution_count": 21, "metadata": {}, "output_type": "execute_result"}], "source": "y_prob = rf_classifier.predict_proba(pred_df)\ny_prob"}, {"cell_type": "code", "execution_count": 22, "id": "b408b0f8", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "0.7873392680514342\n0.8495197438633938\n0.8172484599589321\n"}], "source": "from sklearn.metrics import precision_score, recall_score, f1_score\n\n# Compute precision\nprecision = precision_score(y_test, y_pred)\nprint(precision)\n\n# Compute recall\nrecall = recall_score(y_test, y_pred)\nprint(recall)\n\n# Compute F1-score\nf1 = f1_score(y_test, y_pred)\nprint(f1)"}, {"cell_type": "code", "execution_count": 23, "id": "bd813830", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "Classification Report:\n              precision    recall  f1-score   support\n\n           0       0.84      0.77      0.80       931\n           1       0.79      0.85      0.82       937\n\n    accuracy                           0.81      1868\n   macro avg       0.81      0.81      0.81      1868\nweighted avg       0.81      0.81      0.81      1868\n\nF1 Score: 0.82\nRecall: 0.85\nPrecision: 0.79\nAUC Score: 0.89\n"}, {"data": {"image/png": "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", "text/plain": "<Figure size 640x480 with 1 Axes>"}, "metadata": {}, "output_type": "display_data"}], "source": "# Get predicted probabilities for class 1\ny_prob = rf_classifier.predict_proba(X_test)[:, 1]\n\n# custom_threshold = 0.4  # You can adjust this value\n# y_pred = (y_prob >= custom_threshold).astype(int)\n\n# Classification Report\nclass_report = classification_report(y_test, y_pred)\nprint(\"Classification Report:\")\nprint(class_report)\n\n# F1 Score\nf1 = f1_score(y_test, y_pred)\nprint(f\"F1 Score: {f1:.2f}\")\n\n# Recall\nrecall = recall_score(y_test, y_pred)\nprint(f\"Recall: {recall:.2f}\")\n\n# Precision\nprecision = precision_score(y_test, y_pred)\nprint(f\"Precision: {precision:.2f}\")\n\n# AUC Score\nauc_score = roc_auc_score(y_test, y_prob)\nprint(f\"AUC Score: {auc_score:.2f}\")\n\n# ROC Curve\nfpr, tpr, thresholds = roc_curve(y_test, y_prob)\nroc_auc = auc(fpr, tpr)\n\n# Plot ROC curve\nplt.figure()\nplt.plot(fpr, tpr, color='darkorange', lw=2, label=f'ROC curve (area = {roc_auc:.2f})')\nplt.plot([0, 1], [0, 1], color='navy', lw=2, linestyle='--')\nplt.xlim([0.0, 1.0])\nplt.ylim([0.0, 1.05])\nplt.xlabel('False Positive Rate')\nplt.ylabel('True Positive Rate')\nplt.title('Receiver Operating Characteristic')\nplt.legend(loc=\"lower right\")\nplt.show()"}, {"cell_type": "markdown", "id": "df3abb9e-a0f4-4d31-811d-88162973f303", "metadata": {}, "source": "<a id=\"save_model\"></a>\n## Step 7: Save the model as pickle\nSaving a trained Random Forest Classifier model to a file and uploading it to a COS bucket for the next step. The Dockerfile found in the `container` directory, along with other files in it, can be used to build a runtime container image for serving the model on Docker, SAP AI Core, and other cloud container environments."}, {"cell_type": "code", "execution_count": 24, "id": "c3ed985a-d6f1-42e6-b956-cb5238eaef96", "metadata": {}, "outputs": [{"data": {"text/plain": "['epp.pkl']"}, "execution_count": 24, "metadata": {}, "output_type": "execute_result"}], "source": "import joblib\n\n# Save the model to a file\njoblib.dump(rf_classifier, 'epp.pkl')"}, {"cell_type": "code", "execution_count": 25, "id": "ac22b7d9", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "File Uploaded\n"}], "source": "from ibm_botocore.client import Config\nimport ibm_boto3\n\nclient = ibm_boto3.client(\n    service_name='s3',\n    ibm_api_key_id=cos_credential['COS_API_KEY'],\n    ibm_service_instance_id=cos_credential['COS_INSTANCE_ID'],\n    ibm_auth_endpoint=cos_credential['IBM_AUTH_ENDPOINT'],\n    config=Config(signature_version='oauth'),\n    endpoint_url=cos_credential['COS_ENDPOINT']\n)\n\ntry:\n    client.upload_file(Bucket=cos_credential['BUCKET'], Filename='epp.pkl', Key='scikit_model.pkl')\nexcept Exception as e:\n    print(Exception, e)\nelse:\n    print('File Uploaded')"}, {"cell_type": "markdown", "id": "140f6394", "metadata": {}, "source": "<a id=\"deploy_to_aicore\"></a>\n## Step 8: Deploy the model to SAP AI Core"}, {"cell_type": "markdown", "id": "fbddefa3", "metadata": {}, "source": "Before deploying a machine learning model to SAP AI Core, you need to initializes the connection to AI Core."}, {"cell_type": "code", "execution_count": 26, "id": "daf1323f", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "2024-04-17 03:48:45,624: fedml_dsp.logger INFO: Creating AI Core Connection ....\n2024-04-17 03:48:45,626: fedml_dsp.logger INFO: Sucessfully Created AI Core Connection\n"}], "source": "from fedml_dsp import Fedml\n\ntry:\n    fedml = Fedml(aic_service_key='aicore_service_key.json')\nexcept:\n    raise Exception(\"FedML constructor failed.\")"}, {"cell_type": "markdown", "id": "e4e840a6", "metadata": {}, "source": "Then, you need to onboard any ai core resources required, including your github repository, AI core resource group, and secret to provide AI core pull permissions to your Docker registry."}, {"cell_type": "code", "execution_count": 27, "id": "c3efbabc", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "2024-04-17 03:48:45,640: fedml_dsp.logger INFO: Onboarding Github Repository....\nName: aicore-fedml-demo\nURL: https://github.com/kevinxhuang/aicore-fedml-demo.git\nStatus: RepositoryStatus.COMPLETED\n2024-04-17 03:48:47,729: fedml_dsp.logger INFO: Creating secret....\n2024-04-17 03:48:48,233: fedml_dsp.logger INFO: {'message': 'secret has been created'}\n"}], "source": "try:\n    fedml.onboard_ai_core(\n        create_resource_group=False,\n        resource_group='default',\n        onboard_new_repo=True,\n        github_info_path=\"github_info.json\",\n        secret_path=\"docker_registry_secret.json\"\n    )\nexcept:\n    raise Exception(\"AI Core resource onboarding failed.\")"}, {"cell_type": "markdown", "id": "fe4d45df", "metadata": {}, "source": "Next, you can register a new application in AI Core. If you want to use an existing AI Core application for your deployment, skip this step."}, {"cell_type": "code", "execution_count": 28, "id": "5f3f4b08", "metadata": {"scrolled": true}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "2024-04-17 03:48:52,715: fedml_dsp.logger INFO: {'id': 'fedml-demo-app', 'message': 'Application has been successfully created.'}\n2024-04-17 03:48:57,868: fedml_dsp.logger INFO: successfully synced (all tasks run)\n"}], "source": "APPLICATION_NAME = \"fedml-demo-app\"\n\ntry:\n    application_details = {\n        \"application_name\": APPLICATION_NAME, \n        \"revision\": \"HEAD\", \n        \"repository_url\": \"https://github.com/kevinxhuang/aicore-fedml-demo.git\", \n        \"path\": \"demo-deployment\"\n    }\n\n    fedml.register_application(application_details=application_details)\nexcept:\n    raise Exception(\"Application registration failed.\")"}, {"cell_type": "markdown", "id": "6fb18b9a", "metadata": {}, "source": "Assuming you have built an image and pushed it to the Docker registry, you can now start the deployment of your application to AI Core."}, {"cell_type": "code", "execution_count": 29, "id": "a898d1e2", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "2024-04-17 03:49:58,625: fedml_dsp.logger INFO: {'id': 'a4120b99-88c2-43fb-9692-446f0ce8da92', 'message': 'Configuration created'}\n2024-04-17 03:49:59,166: fedml_dsp.logger INFO: {'id': 'a4120b99-88c2-43fb-9692-446f0ce8da92', 'name': 'fedml-demo-app', 'scenario_id': 'fedml-demo', 'executable_id': 'fedml-demo-serving-exectuable-20240404', 'parameter_bindings': [], 'input_artifact_bindings': [], 'created_at': datetime.datetime(2024, 4, 17, 3, 49, 58, tzinfo=datetime.timezone.utc), 'scenario': None}\n2024-04-17 03:49:59,660: fedml_dsp.logger INFO: {'id': 'd12da55bdfde7c9c', 'message': 'Deployment scheduled.', 'deployment_url': '', 'status': <Status.UNKNOWN: 'UNKNOWN'>, 'ttl': None}\n2024-04-17 03:50:15,167: fedml_dsp.logger INFO: Deployment status....Status.UNKNOWN\n2024-04-17 03:50:30,718: fedml_dsp.logger INFO: Deployment status....Status.UNKNOWN\n2024-04-17 03:50:46,240: fedml_dsp.logger INFO: Deployment status....Status.UNKNOWN\n2024-04-17 03:51:05,851: fedml_dsp.logger INFO: Deployment status....Status.PENDING\n2024-04-17 03:51:21,514: fedml_dsp.logger INFO: Deployment status....Status.PENDING\n2024-04-17 03:51:37,066: fedml_dsp.logger INFO: Deployment status....Status.PENDING\n2024-04-17 03:51:52,596: fedml_dsp.logger INFO: Deployment status....Status.PENDING\n2024-04-17 03:52:08,149: fedml_dsp.logger INFO: Deployment status....Status.PENDING\n2024-04-17 03:52:23,609: fedml_dsp.logger INFO: Deployment status....Status.PENDING\n2024-04-17 03:52:39,165: fedml_dsp.logger INFO: Deployment status....Status.PENDING\n2024-04-17 03:52:57,474: fedml_dsp.logger INFO: Deployment status....Status.RUNNING\n2024-04-17 03:52:57,476: fedml_dsp.logger INFO: Sucessfully deployed. {'id': 'd12da55bdfde7c9c', 'configuration_id': 'a4120b99-88c2-43fb-9692-446f0ce8da92', 'configuration_name': 'fedml-demo-app', 'scenario_id': 'fedml-demo', 'status': <Status.RUNNING: 'RUNNING'>, 'target_status': <TargetStatus.RUNNING: 'RUNNING'>, 'created_at': datetime.datetime(2024, 4, 17, 3, 49, 59, tzinfo=datetime.timezone.utc), 'modified_at': datetime.datetime(2024, 4, 17, 3, 52, 54, tzinfo=datetime.timezone.utc), 'status_message': None, 'status_details': None, 'submission_time': datetime.datetime(2024, 4, 17, 3, 50, 47, tzinfo=datetime.timezone.utc), 'start_time': datetime.datetime(2024, 4, 17, 3, 52, 44, tzinfo=datetime.timezone.utc), 'completion_time': None, 'deployment_url': 'https://api.ai.prod.us-east-1.aws.ml.hana.ondemand.com/v2/inference/deployments/d12da55bdfde7c9c', 'last_operation': <Operation.CREATE: 'CREATE'>, 'latest_running_configuration_id': 'a4120b99-88c2-43fb-9692-446f0ce8da92', 'details': {'resources': {'backend_details': {'predictor': {'resource_plan': 'starter'}}}, 'scaling': {'backend_details': {'predictor': {'max_replicas': '2', 'min_replicas': '1', 'running_replicas': 1}}}}, 'ttl': None}\n2024-04-17 03:52:57,478: fedml_dsp.logger INFO: Deployment url: https://api.ai.prod.us-east-1.aws.ml.hana.ondemand.com/v2/inference/deployments/d12da55bdfde7c9c\n2024-04-17 03:52:57,479: fedml_dsp.logger INFO: Invoke endpoint using: https://api.ai.prod.us-east-1.aws.ml.hana.ondemand.com/v2/inference/deployments/d12da55bdfde7c9c/v2/invocations\n"}], "source": "try:\n    deployment_config = {\n        \"name\": APPLICATION_NAME, \n        \"resource_group\": \"default\", \n        \"scenario_id\": \"fedml-demo\", \n        \"executable_id\": \"fedml-demo-serving-exectuable-20240404\"\n    }\n\n    time.sleep(60)\n    deployment_url = fedml.ai_core_deploy(deployment_config=deployment_config)\nexcept:\n    raise Exception(\"Deployment to AI Core failed\")"}, {"cell_type": "code", "execution_count": 30, "id": "5e39bad9", "metadata": {}, "outputs": [{"data": {"text/plain": "'https://api.ai.prod.us-east-1.aws.ml.hana.ondemand.com/v2/inference/deployments/d12da55bdfde7c9c/v2/invocations'"}, "execution_count": 30, "metadata": {}, "output_type": "execute_result"}], "source": "deployment_url"}, {"cell_type": "markdown", "id": "f6fb90d3", "metadata": {}, "source": "<a id=\"make_predictions\"></a>\n## Step 9: Make predictions\nFinally, after deployment is done, you can run inference on your deployed model."}, {"cell_type": "code", "execution_count": 31, "id": "8f20633c", "metadata": {}, "outputs": [{"data": {"text/plain": "'https://api.ai.prod.us-east-1.aws.ml.hana.ondemand.com/v2/inference/deployments/d12da55bdfde7c9c'"}, "execution_count": 31, "metadata": {}, "output_type": "execute_result"}], "source": "base_url = aicore_service_key[\"serviceurls\"][\"AI_API_URL\"] + \"/v2/inference/deployments/\" + deployment_url[-31:-15]\nbase_url"}, {"cell_type": "markdown", "id": "2ee5f703", "metadata": {}, "source": "You can use FedML library to make an inference call. The `fedml.get_ai_core_token()` function can retrieve an authorization token needed to make inferencing calls to SAP AI Core. The `ai_core_inference(endpoint, headers, body)` function makes an API call to the inference endpoint and returns a response."}, {"cell_type": "code", "execution_count": 32, "id": "a9cafe9f", "metadata": {}, "outputs": [{"name": "stdout", "output_type": "stream", "text": "<Response [200]>\n{'predictions': [{'fields': ['prediction', 'probability'], 'values': [[0, [0.95, 0.05]]]}]}\n"}], "source": "# Prepare a sample input\nsample_input = {\n    'input_data': [\n        {\n            'fields': [\n                'no_of_trainings',\n                'age',\n                'previous_year_rating',\n                'length_of_service',\n                'kpis_met_above_80_percent',\n                'any_awards_won', 'avg_training_score',\n                'department_Finance',\n                'department_HR', 'department_Legal',\n                'department_Operations', 'department_Procurement',\n                'department_R&D', 'department_Sales & Marketing',\n                'department_Technology',\n                \"education_Bachelor's\",\n                'education_Below Secondary',\n                \"education_Master's & above\",\n                'gender_m'],\n            'values': [[1,33,3,6,0,0,63,0,0,0,1,0,0,0,0,0,0,1,1]]\n        }\n    ]\n}\n\n# Make an inference call\ntry:\n    endpoint = f\"{base_url}/v2/predict\" # endpoint implemented in serving engine\n    headers = {\"Authorization\": fedml.get_ai_core_token(),\n               \"AI-Resource-Group\": \"default\",\n               \"Content-Type\": \"application/json\"}\n\n    response = fedml.ai_core_inference(\n        endpoint=endpoint,\n        headers=headers,\n        body=json.dumps(sample_input)\n    )\n\n    print(response)\n    print(response.json())\nexcept:\n    raise Exception(\"Model inferencing failed.\")"}, {"cell_type": "code", "execution_count": null, "id": "a98af79d", "metadata": {}, "outputs": [], "source": ""}], "metadata": {"kernelspec": {"display_name": "Python 3.10", "language": "python", "name": "python3"}, "language_info": {"codemirror_mode": {"name": "ipython", "version": 3}, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.14"}}, "nbformat": 4, "nbformat_minor": 5}