Datasets
Models
Downloads: 1
Diff of
/03-Experiments/Temp-XGBoost_With_Custom_Loss.ipynb
[000000]
..
[b4c0b6]
Switch to side-by-side view
--- a +++ b/03-Experiments/Temp-XGBoost_With_Custom_Loss.ipynb @@ -0,0 +1,438 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "# Global Experiment Setup" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "<Experiment: artifact_location='/Users/arham/Downloads/Projects/03-Experiments/mlruns/2', creation_time=1713912394972, experiment_id='2', last_update_time=1713912394972, lifecycle_stage='active', name='XGBoost', tags={}>" + ] + }, + "execution_count": 1, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "import mlflow\n", + "\n", + "\n", + "# Set the MLflow tracking URI to a new SQLite URI\n", + "mlflow.set_tracking_uri(\"sqlite:///new_mlflow.db\")\n", + "mlflow.set_experiment(\"XGBoost\")" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "metadata": {}, + "outputs": [], + "source": [ + "import pandas as pd\n", + "from sklearn.model_selection import train_test_split\n", + "import matplotlib.pyplot as plt\n", + "import seaborn as sns\n", + "import numpy as np\n", + "from sklearn.preprocessing import MinMaxScaler\n", + "from sklearn.preprocessing import PolynomialFeatures\n", + "import lightgbm as lgb\n", + "from sklearn.metrics import accuracy_score\n", + "import warnings\n", + "from sklearn.tree import DecisionTreeClassifier\n", + "from sklearn.model_selection import cross_val_score\n", + "from sklearn.metrics import accuracy_score, precision_score, recall_score\n", + "import xgboost as xgb\n", + "from sklearn.metrics import accuracy_score, precision_score, recall_score\n", + "from sklearn.model_selection import cross_val_score\n", + "\n", + "\n", + "def load_data(path):\n", + " df = pd.read_csv(path)\n", + " train_df, test_df = train_test_split(df, test_size=0.35, random_state=42)\n", + " train_df, val_df, = train_test_split(train_df, test_size=0.20, random_state=42)\n", + " train_df = train_df.drop(['id'], axis=1).drop_duplicates().reset_index(drop=True)\n", + " test_df = test_df.drop(['id'], axis=1).drop_duplicates().reset_index(drop=True)\n", + " val_df = val_df.drop(['id'], axis=1).drop_duplicates().reset_index(drop=True)\n", + " return train_df, val_df, test_df\n", + "\n", + "def encode_target(train):\n", + " target_key = {'Insufficient_Weight': 0, 'Normal_Weight': 1, 'Overweight_Level_I': 2, 'Overweight_Level_II': 3, 'Obesity_Type_I': 4,'Obesity_Type_II' : 5, 'Obesity_Type_III': 6}\n", + " train['NObeyesdad'] = train['NObeyesdad'].map(target_key)\n", + " return train\n", + "\n", + "def make_gender_binary(train):\n", + " train['Gender'] = train['Gender'].map({'Male':0, 'Female':1})\n", + "\n", + "def datatypes(train):\n", + " train['Weight'] = train['Weight'].astype(float)\n", + " train['Age'] = train['Age'].astype(float)\n", + " train['Height'] = train['Height'].astype(float)\n", + " return train\n", + "\n", + "# def age_binning(train_df):\n", + "# # train_df['Age_Group'] = pd.cut(train_df['Age'], bins=[0, 20, 30, 40, 50, train_df['Age'].max()], labels=['0-20', '21-30', '31-40', '41-50', '50+'])\n", + "# train_df['Age_Group'] = pd.cut(train_df['Age'], bins=[0, 20, 30, 40, 50, train_df['Age'].max()], labels=[1, 2, 3, 4, 5])\n", + "# train_df['Age_Group'] = train_df['Age_Group'].astype(int)\n", + "# return train_df\n", + "\n", + "def age_binning(df):\n", + " age_groups = []\n", + " for age in df['Age']:\n", + " if age <= 20:\n", + " age_group = 1\n", + " elif age <= 30:\n", + " age_group = 2\n", + " elif age <= 40:\n", + " age_group = 3\n", + " elif age <= 50:\n", + " age_group = 4\n", + " else:\n", + " age_group = 5\n", + " age_groups.append(age_group)\n", + " df['Age_Group'] = age_groups\n", + " return df\n", + "\n", + "def age_scaling_log(train_df):\n", + " train_df['Age'] = train_df['Age'].astype(float)\n", + " train_df['Log_Age'] = np.log1p(train_df['Age'])\n", + " return train_df\n", + "\n", + "def age_scaling_minmax(train_df):\n", + " train_df['Age'] = train_df['Age'].astype(float)\n", + " scaler_age = MinMaxScaler()\n", + " train_df['Scaled_Age'] = scaler_age.fit_transform(train_df['Age'].values.reshape(-1, 1))\n", + " return train_df, scaler_age\n", + "\n", + "def weight_scaling_log(train_df):\n", + " train_df['Weight'] = train_df['Weight'].astype(float)\n", + " train_df['Log_Weight'] = np.log1p(train_df['Weight'])\n", + " return train_df\n", + "\n", + "def weight_scaling_minmax(train_df):\n", + " train_df['Weight'] = train_df['Weight'].astype(float)\n", + " scaler_weight = MinMaxScaler()\n", + " train_df['Scaled_Weight'] = scaler_weight.fit_transform(train_df['Weight'].values.reshape(-1, 1))\n", + " return train_df, scaler_weight\n", + "\n", + "def height_scaling_log(train_df):\n", + " train_df['Log_Height'] = np.log1p(train_df['Height'])\n", + " return train_df\n", + "\n", + "def height_scaling_minmax(train_df):\n", + " scaler_height = MinMaxScaler()\n", + " train_df['Scaled_Height'] = scaler_height.fit_transform(train_df['Height'].values.reshape(-1, 1))\n", + " return train_df, scaler_height\n", + "\n", + "def make_gender_binary(train):\n", + " train['Gender'] = train['Gender'].map({'Female':1, 'Male':0})\n", + " return train\n", + "\n", + "def fix_binary_columns(train):\n", + " Binary_Cols = ['family_history_with_overweight','FAVC', 'SCC','SMOKE']\n", + " # if yes then 1 else 0\n", + " for col in Binary_Cols:\n", + " train[col] = train[col].map({'yes': 1, 'no': 0})\n", + " # column datatype integer\n", + " train[col] = train[col].astype(int)\n", + " return train\n", + "\n", + "def freq_cat_cols(train):\n", + " # One hot encoding\n", + " cat_cols = ['CAEC', 'CALC']\n", + " for col in cat_cols:\n", + " train[col] = train[col].map({'no': 0, 'Sometimes': 1, 'Frequently': 2, 'Always': 3})\n", + " return train\n", + "\n", + "def Mtrans(train):\n", + " \"\"\"\n", + " Public_Transportation 8692\n", + " Automobile 1835\n", + " Walking 231\n", + " Motorbike 19\n", + " Bike 16\n", + " \"\"\"\n", + " # train['MTRANS'] = train['MTRANS'].map({'Public_Transportation': 3, 'Automobile': 5, 'Walking': 1, 'Motorbike': 4, 'Bike': 2})\n", + " # dummify column\n", + " train = pd.get_dummies(train, columns=['MTRANS'])\n", + " # convert these columns to integer\n", + " train['MTRANS_Automobile'] = train['MTRANS_Automobile'].astype(int)\n", + " train['MTRANS_Walking'] = train['MTRANS_Walking'].astype(int)\n", + " train['MTRANS_Motorbike'] = train['MTRANS_Motorbike'].astype(int)\n", + " train['MTRANS_Bike'] = train['MTRANS_Bike'].astype(int)\n", + " train['MTRANS_Public_Transportation'] = train['MTRANS_Public_Transportation'].astype(int)\n", + " return train\n", + "\n", + "\n", + "def other_features(train):\n", + " train['BMI'] = train['Weight'] / (train['Height'] ** 2)\n", + " # train['Age'*'Gender'] = train['Age'] * train['Gender']\n", + " polynomial_features = PolynomialFeatures(degree=2)\n", + " X_poly = polynomial_features.fit_transform(train[['Age', 'BMI']])\n", + " poly_features_df = pd.DataFrame(X_poly, columns=['Age^2', 'Age^3', 'BMI^2', 'Age * BMI', 'Age * BMI^2', 'Age^2 * BMI^2'])\n", + " train = pd.concat([train, poly_features_df], axis=1)\n", + " return train\n", + "\n", + "\n", + "def test_pipeline(test, scaler_age, scaler_weight, scaler_height):\n", + " test = datatypes(test)\n", + " test = encode_target(test)\n", + " test = age_binning(test)\n", + " test = age_scaling_log(test)\n", + " test['Scaled_Age'] = scaler_age.transform(test['Age'].values.reshape(-1, 1))\n", + " test = weight_scaling_log(test)\n", + " test['Scaled_Weight'] = scaler_weight.transform(test['Weight'].values.reshape(-1, 1))\n", + " test = height_scaling_log(test)\n", + " test['Scaled_Height'] = scaler_height.transform(test['Height'].values.reshape(-1, 1))\n", + " test = make_gender_binary(test)\n", + " test = fix_binary_columns(test)\n", + " test = freq_cat_cols(test)\n", + " test = Mtrans(test)\n", + " test = other_features(test)\n", + "\n", + " return test\n", + "\n", + "def train_model(params, X_train, y_train):\n", + " lgb_train = lgb.Dataset(X_train, y_train)\n", + " model = lgb.train(params, lgb_train, num_boost_round=1000)\n", + " return model\n", + "\n", + "def evaluate_model(model, X_val, y_val):\n", + " y_pred = model.predict(X_val)\n", + " y_pred = [np.argmax(y) for y in y_pred]\n", + " accuracy = accuracy_score(y_val, y_pred)\n", + " return accuracy\n", + "\n", + "def objective(trial, X_train, y_train):\n", + " params = {\n", + " 'objective': 'multiclass',\n", + " 'num_class': 7,\n", + " 'metric': 'multi_logloss',\n", + " 'boosting_type': 'gbdt',\n", + " 'learning_rate': trial.suggest_loguniform('learning_rate', 0.005, 0.5),\n", + " 'num_leaves': trial.suggest_int('num_leaves', 10, 1000),\n", + " 'max_depth': trial.suggest_int('max_depth', -1, 20),\n", + " 'bagging_fraction': trial.suggest_uniform('bagging_fraction', 0.6, 0.95),\n", + " 'feature_fraction': trial.suggest_uniform('feature_fraction', 0.6, 0.95),\n", + " 'verbosity': -1\n", + " }\n", + "\n", + " n_splits = 5\n", + " kf = StratifiedKFold(n_splits=n_splits, shuffle=True, random_state=42)\n", + " scores = []\n", + "\n", + " for train_index, val_index in kf.split(X_train, y_train):\n", + " X_tr, X_val = X_train.iloc[train_index], X_train.iloc[val_index]\n", + " y_tr, y_val = y_train.iloc[train_index], y_train.iloc[val_index]\n", + "\n", + " model = train_model(params, X_tr, y_tr)\n", + " accuracy = evaluate_model(model, X_val, y_val)\n", + " scores.append(accuracy)\n", + "\n", + " return np.mean(scores)\n", + "\n", + "def optimize_hyperparameters(X_train, y_train, n_trials=2):\n", + " study = optuna.create_study(direction='maximize')\n", + " study.optimize(lambda trial: objective(trial, X_train, y_train), n_trials=n_trials)\n", + " return study.best_params\n" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### XGB With Feature Engineering" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "metadata": {}, + "outputs": [], + "source": [ + "\n", + "path = '/Users/arham/Downloads/Projects/01-Dataset/01-Data-for-model-building/train.csv'\n", + "train_df, val_df, test_df = load_data(path)\n", + "\n", + "train_df = datatypes(train_df)\n", + "train_df = encode_target(train_df)\n", + "train_df = age_binning(train_df)\n", + "train_df, scaler_age = age_scaling_minmax(train_df)\n", + "train_df = age_scaling_log(train_df)\n", + "train_df, scaler_weight = weight_scaling_minmax(train_df)\n", + "train_df = weight_scaling_log(train_df)\n", + "train_df, scaler_height = height_scaling_minmax(train_df)\n", + "train_df = height_scaling_log(train_df)\n", + "train_df = make_gender_binary(train_df)\n", + "train_df = fix_binary_columns(train_df)\n", + "train_df = freq_cat_cols(train_df)\n", + "train_df = Mtrans(train_df)\n", + "train_df = other_features(train_df)\n", + "\n", + "val_df = test_pipeline(val_df, scaler_age, scaler_weight, scaler_height)\n", + "test_df = test_pipeline(test_df, scaler_age, scaler_weight, scaler_height)\n", + "\n", + "Target = 'NObeyesdad'\n", + "# features = train_df.columns.drop(Target)\n", + "features = ['Gender', 'Age', 'Height', 'Weight', 'family_history_with_overweight',\n", + " 'FAVC', 'FCVC', 'NCP', 'CAEC', 'SMOKE', 'CH2O', 'SCC', 'FAF', 'TUE',\n", + " 'CALC', 'Age_Group', \n", + " 'MTRANS_Automobile', 'MTRANS_Bike', 'MTRANS_Motorbike',\n", + " 'MTRANS_Public_Transportation', 'MTRANS_Walking', 'BMI', 'Age^2',\n", + " 'Age^3', 'BMI^2', 'Age * BMI', 'Age * BMI^2', 'Age^2 * BMI^2'] \n", + "\n", + " #'Scaled_Age', 'Log_Age', 'Scaled_Weight', 'Log_Weight', 'Scaled_Height', 'Log_Height',\n", + "\n", + "\n", + "X_train = train_df[features]\n", + "y_train = train_df[Target]\n", + "X_val = val_df[features]\n", + "y_val = val_df[Target]\n", + "X_test = test_df[features]\n", + "y_test = test_df[Target]\n", + "\n", + "# save X_train, y_train, X_val, X_test, y_test\n" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Cross-validation Scores (XGBoost): [0.90597499 0.90736452 0.89671144 0.89620019 0.90222428]\n", + "Mean CV Accuracy (XGBoost): 0.9016950833225661\n", + "\n", + "Accuracy (XGBoost): 0.9036680251945165\n", + "Precision (XGBoost): 0.9042803910684232\n", + "Recall (XGBoost): 0.9036680251945165\n", + "F1 (XGBoost): 0.9039741044249812\n", + "Recall for class 0: 0.9240506329113924\n", + "Recall for class 1: 0.9064171122994652\n", + "Recall for class 2: 0.7582089552238805\n", + "Recall for class 3: 0.8449848024316109\n", + "Recall for class 4: 0.8741092636579573\n", + "Recall for class 5: 0.9665071770334929\n", + "Recall for class 6: 0.9960474308300395\n" + ] + } + ], + "source": [ + "import xgboost as xgb\n", + "from sklearn.model_selection import cross_val_score\n", + "from sklearn.metrics import accuracy_score, precision_score, recall_score\n", + "import mlflow\n", + "import warnings\n", + "warnings.filterwarnings(\"ignore\")\n", + "# import precision_recall_fscore_support\n", + "from sklearn.metrics import precision_recall_fscore_support\n", + "\n", + "mlflow.sklearn.autolog(disable=True)\n", + "\n", + "with mlflow.start_run(run_name=\"XGB_with_Feature_Engineering\"):\n", + " class_counts_train = [y_train[y_train == i].count() for i in range(7)]\n", + " total_samples_train = len(y_train)\n", + " scale_pos_weights = [total_samples_train / (7 * count) for count in class_counts_train]\n", + "\n", + " xgb_classifier = xgb.XGBClassifier(objective='multi:softprob', num_class=7)\n", + " cv_scores_xgb = cross_val_score(xgb_classifier, X_train, y_train, cv=5, scoring='accuracy')\n", + " print(\"Cross-validation Scores (XGBoost):\", cv_scores_xgb)\n", + " print(\"Mean CV Accuracy (XGBoost):\", cv_scores_xgb.mean())\n", + " xgb_classifier.fit(X_train, y_train)\n", + " y_val_pred_proba_xgb = xgb_classifier.predict_proba(X_val)\n", + " y_val_pred_xgb = np.argmax(y_val_pred_proba_xgb, axis=1)\n", + " accuracy_xgb = accuracy_score(y_val, y_val_pred_xgb)\n", + " precision_xgb = precision_score(y_val, y_val_pred_xgb, average='weighted')\n", + " recall_xgb = recall_score(y_val, y_val_pred_xgb, average='weighted')\n", + " f1_xgb = 2 * (precision_xgb * recall_xgb) / (precision_xgb + recall_xgb)\n", + " print(\"\\nAccuracy (XGBoost):\", accuracy_xgb)\n", + " print(\"Precision (XGBoost):\", precision_xgb)\n", + " print(\"Recall (XGBoost):\", recall_xgb)\n", + " print(\"F1 (XGBoost):\", f1_xgb)\n", + " mlflow.log_metric('accuracy', accuracy_xgb)\n", + " mlflow.log_metric('precision', precision_xgb)\n", + " mlflow.log_metric('recall', recall_xgb)\n", + " mlflow.log_metric('f1', f1_xgb)\n", + "\n", + " precision_per_class, recall_per_class, f1_per_class, support_per_class = precision_recall_fscore_support(y_val, y_val_pred_xgb, average=None)\n", + " for i in range(len(recall_per_class)):\n", + " print(f\"Recall for class {i}: {recall_per_class[i]}\")\n", + " mlflow.log_metric(f'recall_class_{i}', recall_per_class[i])\n", + "\n", + " mlflow.xgboost.log_model(xgb_classifier, 'model')\n", + " mlflow.set_tag('experiments', 'Arham A.')\n", + " mlflow.set_tag('model_name', 'XGBoost')\n", + " mlflow.set_tag('preprocessing', 'Yes')\n" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "[2024-04-25 14:23:23 -0400] [11501] [INFO] Starting gunicorn 21.2.0\n", + "[2024-04-25 14:23:23 -0400] [11501] [INFO] Listening at: http://127.0.0.1:5000 (11501)\n", + "[2024-04-25 14:23:23 -0400] [11501] [INFO] Using worker: sync\n", + "[2024-04-25 14:23:23 -0400] [11502] [INFO] Booting worker with pid: 11502\n", + "[2024-04-25 14:23:23 -0400] [11503] [INFO] Booting worker with pid: 11503\n", + "[2024-04-25 14:23:23 -0400] [11504] [INFO] Booting worker with pid: 11504\n", + "[2024-04-25 14:23:23 -0400] [11505] [INFO] Booting worker with pid: 11505\n", + "^C\n", + "[2024-04-25 14:24:17 -0400] [11501] [INFO] Handling signal: int\n", + "[2024-04-25 14:24:17 -0400] [11505] [INFO] Worker exiting (pid: 11505)\n", + "[2024-04-25 14:24:17 -0400] [11504] [INFO] Worker exiting (pid: 11504)\n", + "[2024-04-25 14:24:17 -0400] [11502] [INFO] Worker exiting (pid: 11502)\n", + "[2024-04-25 14:24:17 -0400] [11503] [INFO] Worker exiting (pid: 11503)\n" + ] + } + ], + "source": [ + "!mlflow ui --backend-store-uri \"sqlite:////Users/arham/Downloads/Projects/03-Experiments/new_mlflow.db\"" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [] + } + ], + "metadata": { + "kernelspec": { + "display_name": "DataScience", + "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": "3.10.13" + } + }, + "nbformat": 4, + "nbformat_minor": 2 +}