{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.14","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"nvidiaTeslaT4","dataSources":[{"sourceId":84896,"databundleVersionId":10305135,"sourceType":"competition"}],"dockerImageVersionId":30787,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"

imports

","metadata":{}},{"cell_type":"code","source":"!pip install -q scikit-learn==1.5.2","metadata":{"trusted":true,"_kg_hide-output":true,"_kg_hide-input":false},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import warnings\nfrom pathlib import Path\nwarnings.filterwarnings('ignore')","metadata":{"trusted":true,"_kg_hide-input":false,"_kg_hide-output":false},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import numpy as np\nimport polars as pl\nimport pandas as pd\nimport plotly.colors as pc\nimport plotly.express as px\nimport plotly.graph_objects as go","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import plotly.io as pio\npio.renderers.default = 'iframe'","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"pd.options.display.max_columns = None","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from catboost import CatBoostRegressor\nfrom sklearn.model_selection import KFold\nfrom sklearn.metrics import root_mean_squared_log_error as rmsle","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

configuration class

","metadata":{}},{"cell_type":"code","source":"class CFG:\n\n train_path = Path('/kaggle/input/playground-series-s4e12/train.csv')\n test_path = Path('/kaggle/input/playground-series-s4e12/test.csv')\n subm_path = Path('/kaggle/input/playground-series-s4e12/sample_submission.csv')\n\n batch_size = 262144\n color = '#EADDCA'\n\n early_stop = 200\n n_splits = 5\n\n ctb_params = {\n 'bootstrap_type': 'Bernoulli',\n 'loss_function': 'RMSE',\n 'learning_rate': 0.03,\n 'num_trees': 20000,\n 'random_state': 42,\n 'task_type': 'GPU',\n 'subsample': 0.85,\n 'reg_lambda': 8.0,\n 'depth': 10\n }","metadata":{"trusted":true,"_kg_hide-input":false},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

feature engineering

","metadata":{}},{"cell_type":"code","source":"class FE:\n\n def __init__(self, batch_size):\n self.batch_size = batch_size\n\n def load_data(self, path):\n\n return pl.read_csv(path, batch_size=self.batch_size).drop('id')\n\n def cast_datatypes(self, df):\n\n cat_cols = [\n 'Gender',\n 'Marital Status',\n 'Education Level',\n 'Occupation',\n 'Location',\n 'Policy Type',\n 'Policy Start Date',\n 'Customer Feedback',\n 'Smoking Status',\n 'Exercise Frequency',\n 'Property Type' \t\t\t\t\n ]\n\n '''\n Replace missing values with:\n\n `-1` for numeric columns\n `Unknown` for categorical columns\n '''\n\n for col in df.columns:\n\n if col in cat_cols:\n df = df.with_columns(pl.col(col).fill_null('Unknown').cast(pl.String))\n\n else:\n df = df.with_columns(pl.col(col).fill_null(-1).cast(pl.Float32)) \n\n return df \n\n def process_data(self, df):\n\n df = df.with_columns([\n \n pl.when(pl.col('Gender').eq('Male')).then(0).otherwise(1)\n .cast(pl.Int8).alias('Gender'),\n \n pl.when(pl.col('Smoking Status').eq('No')).then(0).otherwise(1)\n .cast(pl.Int8).alias('Smoking Status')\n \n ])\n\n # Regex pattern for date column (Policy Start Date): Year (4 digits) - Month (2 digits) - Day (2 digits)\n pattern = r\"(\\d{4})-(\\d{2})-(\\d{2})\"\n \n df = df.with_columns([\n pl.col('Policy Start Date').str.extract(pattern, 1).cast(pl.Int16).alias(\"Policy Start Year\"),\n pl.col('Policy Start Date').str.extract(pattern, 2).cast(pl.Int8).alias(\"Policy Start Month\"),\n pl.col('Policy Start Date').str.extract(pattern, 3).cast(pl.Int8).alias(\"Policy Start Day\"),\n ])\n\n return df.drop('Policy Start Date')\n\n def info(self, df):\n \n print(f'Shape of dataframe: {df.shape}') \n \n mem = df.memory_usage().sum() / 1024**2\n print('Memory usage: {:.2f} MB\\n'.format(mem))\n\n display(df.head())\n\n def apply_fe(self, path):\n\n df = self.load_data(path)\n df = self.cast_datatypes(df)\n df = self.process_data(df)\n df = df.to_pandas()\n\n self.info(df)\n \n cat_cols = [col for col in df.columns if df[col].dtype == pl.String]\n\n return df, cat_cols","metadata":{"trusted":true,"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"fe = FE(CFG.batch_size)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_data, cat_cols = fe.apply_fe(CFG.train_path)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"test_data, _ = fe.apply_fe(CFG.test_path)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

exploratory data analysis

","metadata":{}},{"cell_type":"code","source":"class EDA:\n \n def __init__(self, color, df):\n self.color = color \n self.df = df \n\n def template(self, fig, title):\n \n fig.update_layout(\n title=title,\n title_x=0.5, \n plot_bgcolor='rgba(40, 40, 43, 1)', \n paper_bgcolor='rgba(40, 40, 43, 1)', \n font=dict(color=self.color),\n margin=dict(l=72, r=72, t=72, b=72), \n height=720\n )\n \n return fig\n\n def distribution_plot(self, col, color):\n \n fig = px.histogram(\n self.df,\n x=col,\n nbins=100,\n title=f'Distribution of {col}',\n color_discrete_sequence=[color]\n )\n \n fig.update_layout(\n xaxis_title=col,\n yaxis_title='Count',\n bargap=0.1,\n xaxis=dict(gridcolor='grey'),\n yaxis=dict(gridcolor='grey', zerolinecolor='grey')\n )\n \n fig.update_traces(hovertemplate='Value: %{x:.0f}
Count: %{y:,}')\n \n fig = self.template(fig, f'Distribution of {col}')\n fig.show()\n\n def bubble_chart(self, col, colorscale):\n \n value_counts = self.df[col].value_counts().reset_index()\n value_counts.columns = [col, 'count']\n \n fig = px.scatter(\n value_counts,\n x=col,\n y='count',\n size='count',\n color='count',\n size_max=180,\n color_continuous_scale=colorscale,\n labels={col: col.capitalize(), 'count': 'Count'},\n hover_name=col,\n hover_data={'count': True}\n )\n \n fig.update_layout(\n title_text=f'Bubble Chart of {col.capitalize()}',\n title_x=0.5,\n plot_bgcolor='rgba(40, 40, 43, 1)',\n paper_bgcolor='rgba(40, 40, 43, 1)',\n font=dict(color=self.color),\n margin=dict(l=20, r=20, t=50, b=20),\n xaxis_title=col.capitalize(),\n yaxis_title='Count',\n height=720\n )\n \n fig.update_traces(\n marker=dict(line=dict(width=1, color='DarkSlateGrey')),\n hovertemplate=(\n 'Value: %{x}
'\n 'Count: %{y:,}
'\n ),\n hoverlabel=dict(\n font=dict(color=self.color), \n bgcolor='rgba(40, 40, 43, 1)'\n )\n )\n \n fig = self.template(fig, f'Bubble Chart of {col}')\n fig.show()\n\n def pie_chart(self, col):\n \n value_counts = self.df[col].value_counts().reset_index()\n value_counts.columns = [col, 'count']\n \n fig = px.pie(\n value_counts,\n names=col,\n values='count',\n color=col,\n color_discrete_sequence=px.colors.sequential.Redor,\n title=f'Pie Chart of {col.capitalize()}',\n hole=0.2\n )\n \n fig.update_layout(\n title_x=0.5,\n plot_bgcolor='rgba(40, 40, 43, 1)',\n paper_bgcolor='rgba(40, 40, 43, 1)',\n font=dict(color=self.color),\n margin=dict(l=20, r=20, t=50, b=20),\n height=720\n )\n \n fig.update_traces(\n textinfo='percent+label',\n hovertemplate=(\n 'Value: %{label}
'\n 'Count: %{value:,}
'\n 'Percentage: %{percent}
'\n ),\n hoverlabel=dict(\n font=dict(color=self.color),\n bgcolor='rgba(40, 40, 43, 1)'\n )\n )\n \n fig = self.template(fig, f'Pie Chart of {col}')\n fig.show()","metadata":{"trusted":true,"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda = EDA(CFG.color, train_data)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.distribution_plot(col='Premium Amount', color='#EADDCA') ","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.pie_chart(col='Property Type')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.bubble_chart(col='Location', colorscale='Magenta')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.pie_chart(col='Number of Dependents')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.bubble_chart(col='Education Level', colorscale='Burg')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.bubble_chart(col='Occupation', colorscale='Sunset')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.distribution_plot(col='Annual Income', color='#C19A6B') ","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.distribution_plot(col='Credit Score', color='#834333') ","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.distribution_plot(col='Age', color='#B87333') ","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.pie_chart(col='Marital Status')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.distribution_plot(col='Health Score', color='#C2B280') ","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eda.bubble_chart(col='Exercise Frequency', colorscale='Sunsetdark')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

model development

","metadata":{}},{"cell_type":"code","source":"class MD:\n \n def __init__(self, early_stop, n_splits, color):\n self.early_stop = early_stop\n self.n_splits = n_splits\n self.color = color\n \n def _plot_cv(self, fold_scores, title, metric='RMSLE'):\n \n fold_scores = [round(score, 3) for score in fold_scores]\n mean_score = round(np.mean(fold_scores), 3)\n\n fig = go.Figure()\n\n fig.add_trace(go.Scatter(\n x = list(range(1, len(fold_scores) + 1)),\n y = fold_scores,\n mode = 'markers', \n name = 'Fold Scores',\n marker = dict(size = 27, color=self.color, symbol='diamond'),\n text = [f'{score:.3f}' for score in fold_scores],\n hovertemplate = 'Fold %{x}: %{text}',\n hoverlabel=dict(font=dict(size=18)) \n ))\n\n fig.add_trace(go.Scatter(\n x = [1, len(fold_scores)],\n y = [mean_score, mean_score],\n mode = 'lines',\n name = f'Mean: {mean_score:.3f}',\n line = dict(dash = 'dash', color = '#FFBF00'),\n hoverinfo = 'none'\n ))\n \n fig.update_layout(\n title = f'{title} | Cross-validation {metric} scores: {mean_score}',\n xaxis_title = 'Fold',\n yaxis_title = f'{metric} Score',\n plot_bgcolor = 'rgba(40, 40, 43, 1)', \n paper_bgcolor = 'rgba(40, 40, 43, 1)',\n font = dict(color=self.color), \n xaxis = dict(\n gridcolor = 'grey',\n tickmode = 'linear',\n tick0 = 1,\n dtick = 1,\n range = [0.5, len(fold_scores) + 0.5],\n zerolinecolor = 'grey'\n ),\n yaxis = dict(\n gridcolor = 'grey',\n zerolinecolor = 'grey'\n )\n )\n \n fig.show()\n \n def train_model(self, data, cat_cols, params, title):\n\n for col in cat_cols:\n data[col] = data[col].astype('category')\n \n X = data.drop(['Premium Amount'], axis=1)\n y = data['Premium Amount']\n\n cv = KFold(n_splits=self.n_splits, shuffle=True, random_state=42)\n\n models, scores = [], []\n\n # Initialize out-of-fold predictions array\n oof_preds = np.zeros(len(X))\n\n for fold, (train_index, valid_index) in enumerate(cv.split(X, y), 1):\n \n X_train, X_valid = X.iloc[train_index], X.iloc[valid_index]\n y_train, y_valid = y.iloc[train_index], y.iloc[valid_index]\n\n if title.startswith('LightGBM'):\n \n model = lgb.LGBMRegressor(**params)\n \n model.fit(X_train, y_train,\n eval_set=[(X_valid, y_valid)],\n eval_metric='rmse',\n callbacks=[lgb.early_stopping(self.early_stop, verbose=0), lgb.log_evaluation(0)])\n \n elif title.startswith('CatBoost'):\n \n model = CatBoostRegressor(**params, verbose=0, cat_features=cat_cols)\n \n model.fit(X_train, y_train,\n eval_set=(X_valid, y_valid),\n early_stopping_rounds=self.early_stop, verbose=0)\n \n models.append(model)\n\n # Store out-of-fold predictions - Ensure predictions are non-negative\n oof_preds[valid_index] = np.clip(model.predict(X_valid), 0, None)\n \n score = rmsle(y_valid, oof_preds[valid_index])\n scores.append(score)\n\n self._plot_cv(scores, title)\n \n return models\n\n def infer_model(self, data, cat_cols, models):\n\n for col in cat_cols:\n data[col] = data[col].astype('category')\n\n return np.mean([model.predict(data) for model in models], axis=0)","metadata":{"trusted":true,"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"md = MD(CFG.early_stop, CFG.n_splits, CFG.color)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"ctb_models = md.train_model(train_data, cat_cols, CFG.ctb_params, 'CatBoost')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

inference

","metadata":{}},{"cell_type":"code","source":"ctb_preds = md.infer_model(test_data, cat_cols, ctb_models)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"subm_data = pd.read_csv(CFG.subm_path)\nsubm_data['Premium Amount'] = ctb_preds","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"subm_data.to_csv('submission.csv', index=False)\ndisplay(subm_data.head())","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}