{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "865de26a-88b3-4f97-b0df-b8f20d62170d", "metadata": {}, "outputs": [], "source": [ "import os, warnings, numpy as np, pandas as pd\n", "warnings.filterwarnings(\"ignore\")\n", "\n", "from sklearn.model_selection import train_test_split, GridSearchCV, StratifiedKFold\n", "from sklearn.preprocessing import OneHotEncoder, StandardScaler\n", "from sklearn.compose import ColumnTransformer\n", "from sklearn.pipeline import Pipeline\n", "from sklearn.impute import SimpleImputer\n", "from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, classification_report, confusion_matrix\n", "\n", "from sklearn.linear_model import LogisticRegression\n", "from sklearn.ensemble import RandomForestClassifier\n", "from sklearn.svm import SVC\n", "\n", "from sklearn.feature_selection import SelectKBest, mutual_info_classif, SelectFromModel\n", "\n", "from joblib import dump\n", "\n", "\n", "CSV_PATH = \"Breast_Cancer1.csv\" \n", "TARGET = \"Status\" \n", "RANDOM_STATE = 42\n", "\n", "OUT_DIR = \"ml_outputs\"\n", "os.makedirs(OUT_DIR, exist_ok=True)\n" ] }, { "cell_type": "code", "execution_count": 3, "id": "456e061a-19cc-4a9d-b5a1-d67d7890dccb", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "Age 0\n", "Race 0\n", "Marital Status 0\n", "T Stage 0\n", "N Stage 0\n", "6th Stage 0\n", "differentiate 0\n", "Grade 0\n", "A Stage 0\n", "Tumor Size 0\n", "dtype: int64" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "df = pd.read_csv(CSV_PATH)\n", "assert TARGET in df.columns, f\"Target '{TARGET}' not found! Available: {list(df.columns)}\"\n", "\n", "y = df[TARGET]\n", "X = df.drop(columns=[TARGET])\n", "\n", "num_cols = X.select_dtypes(include=[np.number]).columns.tolist()\n", "cat_cols = [c for c in X.columns if c not in num_cols]\n", "\n", "num_imputer = SimpleImputer(strategy=\"median\")\n", "cat_imputer = SimpleImputer(strategy=\"most_frequent\")\n", "\n", "if num_cols:\n", " X[num_cols] = num_imputer.fit_transform(X[num_cols])\n", "if cat_cols:\n", " X[cat_cols] = cat_imputer.fit_transform(X[cat_cols])\n", "\n", "df_clean = pd.concat([X, y], axis=1)\n", "df_clean.to_csv(CSV_PATH, index=False)\n", "\n", "\n", "display(df_clean.isnull().sum().sort_values(ascending=False).head(10))\n" ] }, { "cell_type": "code", "execution_count": 5, "id": "50ab35c7-1359-4f44-8daa-6a904a3e405a", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
phasemodelbest_paramscv_best_f1_macroaccuracyprecision_macrorecall_macrof1_macro
0fullLogReg{'clf__C': 1.0, 'clf__penalty': 'l1'}0.7589530.8956520.8347950.7185110.758134
1fullRandomForest{'clf__max_depth': None, 'clf__min_samples_spl...0.7909830.8944100.8387760.7077820.749800
2fullSVC{'clf__C': 10.0, 'clf__gamma': 'scale', 'clf__...0.7411480.8819880.7934080.6937870.727642
\n", "
" ], "text/plain": [ " phase model best_params \\\n", "0 full LogReg {'clf__C': 1.0, 'clf__penalty': 'l1'} \n", "1 full RandomForest {'clf__max_depth': None, 'clf__min_samples_spl... \n", "2 full SVC {'clf__C': 10.0, 'clf__gamma': 'scale', 'clf__... \n", "\n", " cv_best_f1_macro accuracy precision_macro recall_macro f1_macro \n", "0 0.758953 0.895652 0.834795 0.718511 0.758134 \n", "1 0.790983 0.894410 0.838776 0.707782 0.749800 \n", "2 0.741148 0.881988 0.793408 0.693787 0.727642 " ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "df = pd.read_csv(CSV_PATH)\n", "y = df[TARGET]\n", "X = df.drop(columns=[TARGET])\n", "\n", "num_cols = X.select_dtypes(include=[np.number]).columns.tolist()\n", "cat_cols = [c for c in X.columns if c not in num_cols]\n", "\n", "numeric_tf = Pipeline([(\"imputer\", SimpleImputer(strategy=\"median\")), (\"scaler\", StandardScaler())])\n", "categorical_tf = Pipeline([(\"imputer\", SimpleImputer(strategy=\"most_frequent\")), (\"onehot\", OneHotEncoder(handle_unknown=\"ignore\"))])\n", "\n", "preprocessor = ColumnTransformer([(\"num\", numeric_tf, num_cols), (\"cat\", categorical_tf, cat_cols)])\n", "\n", "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, stratify=y, random_state=RANDOM_STATE)\n", "\n", "models = {\n", " \"LogReg\": {\n", " \"pipe\": Pipeline([(\"prep\", preprocessor), (\"clf\", LogisticRegression(max_iter=2000, solver=\"liblinear\", random_state=RANDOM_STATE))]),\n", " \"grid\": {\"clf__penalty\": [\"l1\",\"l2\"], \"clf__C\": [0.1, 1.0, 10.0]}\n", " },\n", " \"RandomForest\": {\n", " \"pipe\": Pipeline([(\"prep\", preprocessor), (\"clf\", RandomForestClassifier(random_state=RANDOM_STATE))]),\n", " \"grid\": {\"clf__n_estimators\": [200, 400], \"clf__max_depth\": [None, 10], \"clf__min_samples_split\": [2, 5]}\n", " },\n", " \"SVC\": {\n", " \"pipe\": Pipeline([(\"prep\", preprocessor), (\"clf\", SVC(probability=False, random_state=RANDOM_STATE))]),\n", " \"grid\": {\"clf__kernel\": [\"rbf\",\"linear\"], \"clf__C\": [1.0, 10.0], \"clf__gamma\": [\"scale\",\"auto\"]}\n", " }\n", "}\n", "\n", "cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=RANDOM_STATE)\n", "\n", "def fit_grid_eval(name, pipe, grid):\n", " gs = GridSearchCV(pipe, grid, scoring=\"f1_macro\", cv=cv, n_jobs=-1, verbose=0)\n", " gs.fit(X_train, y_train)\n", " best = gs.best_estimator_\n", " y_pred = best.predict(X_test)\n", " row = {\n", " \"phase\": \"full\",\n", " \"model\": name,\n", " \"best_params\": str(gs.best_params_),\n", " \"cv_best_f1_macro\": gs.best_score_,\n", " \"accuracy\": accuracy_score(y_test, y_pred),\n", " \"precision_macro\": precision_score(y_test, y_pred, average=\"macro\", zero_division=0),\n", " \"recall_macro\": recall_score(y_test, y_pred, average=\"macro\", zero_division=0),\n", " \"f1_macro\": f1_score(y_test, y_pred, average=\"macro\", zero_division=0),\n", " }\n", " dump(best, os.path.join(OUT_DIR, f\"best_full_{name}.joblib\"))\n", " with open(os.path.join(OUT_DIR, f\"report_full_{name}.txt\"), \"w\", encoding=\"utf-8\") as f:\n", " f.write(classification_report(y_test, y_pred, zero_division=0))\n", " return row, confusion_matrix(y_test, y_pred)\n", "\n", "full_rows, cms_full = [], {}\n", "for name, spec in models.items():\n", " r, cm = fit_grid_eval(name, spec[\"pipe\"], spec[\"grid\"])\n", " full_rows.append(r); cms_full[name] = cm\n", "\n", "df_full = pd.DataFrame(full_rows).sort_values(\"f1_macro\", ascending=False)\n", "display(df_full)\n", "df_full.to_csv(os.path.join(OUT_DIR, \"results_full.csv\"), index=False)\n", "\n" ] }, { "cell_type": "code", "execution_count": 7, "id": "1b22123f-3903-4956-8fbb-a9bf9a1d5a87", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "KBest: ['Survival Months', 'N Stage', 'Reginol Node Positive']\n", "L1: ['Survival Months', 'N Stage']\n", "Final selected (intersection): ['N Stage', 'Survival Months']\n" ] } ], "source": [ "\n", "df = pd.read_csv(CSV_PATH)\n", "y = df[TARGET]\n", "X = df.drop(columns=[TARGET])\n", "\n", "num_cols = X.select_dtypes(include=[np.number]).columns.tolist()\n", "cat_cols = [c for c in X.columns if c not in num_cols]\n", "\n", "num_imputer = SimpleImputer(strategy=\"median\")\n", "cat_imputer = SimpleImputer(strategy=\"most_frequent\")\n", "if num_cols:\n", " X[num_cols] = num_imputer.fit_transform(X[num_cols])\n", "if cat_cols:\n", " X[cat_cols] = cat_imputer.fit_transform(X[cat_cols])\n", "\n", "X_cat_oh = pd.get_dummies(X[cat_cols], drop_first=False) if cat_cols else pd.DataFrame(index=X.index)\n", "X_all = pd.concat([X[num_cols], X_cat_oh], axis=1)\n", "\n", "orig_to_expanded = {}\n", "for c in num_cols:\n", " orig_to_expanded[c] = [c]\n", "for c in cat_cols:\n", " cols = [col for col in X_all.columns if col.startswith(c + \"_\")]\n", " if not cols:\n", " cols = [c]\n", " orig_to_expanded[c] = cols\n", "\n", "# (1) SelectKBest(mutual_info)\n", "y_enc = y.astype(\"category\").cat.codes if not np.issubdtype(y.dtype, np.number) else y\n", "# k يعتمد على عدد الأعمدة الأصلية\n", "TOP_K = min(15, max(5, len(X.columns)//3))\n", "skb = SelectKBest(mutual_info_classif, k=min(TOP_K, X_all.shape[1]))\n", "skb.fit(X_all, y_enc)\n", "scores = pd.Series(skb.scores_, index=X_all.columns).fillna(0)\n", "kbest_scores_by_orig = {orig: float(scores.loc[[c for c in orig_to_expanded[orig] if c in scores]].sum()) for orig in X.columns}\n", "kbest_selected = list(pd.Series(kbest_scores_by_orig).sort_values(ascending=False).head(TOP_K).index)\n", "\n", "# (2) L1-Logistic (SelectFromModel) \n", "scaler = StandardScaler(with_mean=False)\n", "X_all_scaled = pd.DataFrame(scaler.fit_transform(X_all), columns=X_all.columns, index=X_all.index)\n", "\n", "l1_lr = LogisticRegression(penalty=\"l1\", solver=\"liblinear\", max_iter=3000, random_state=RANDOM_STATE)\n", "l1_lr.fit(X_all_scaled, y_enc)\n", "coefs = l1_lr.coef_\n", "abs_coef = np.abs(coefs).mean(axis=0) if coefs.ndim == 2 else np.abs(coefs)\n", "abs_coef_s = pd.Series(abs_coef.ravel(), index=X_all.columns)\n", "\n", "l1_scores_by_orig = {orig: float(abs_coef_s.loc[[c for c in orig_to_expanded[orig] if c in abs_coef_s]].sum()) for orig in X.columns}\n", "l1_rank = pd.Series(l1_scores_by_orig).sort_values(ascending=False)\n", "median_thr = l1_rank.median()\n", "l1_selected = list(l1_rank[l1_rank >= median_thr].index)\n", "if len(l1_selected) > TOP_K:\n", " l1_selected = list(l1_rank.head(TOP_K).index)\n", "\n", "final_selected = sorted(list(set(kbest_selected).intersection(l1_selected)))\n", "if len(final_selected) == 0: # احتياط\n", " final_selected = kbest_selected[:]\n", "\n", "print(\"KBest:\", kbest_selected)\n", "print(\"L1:\", l1_selected)\n", "print(\"Final selected (intersection):\", final_selected)\n", "\n", "df_selected = pd.concat([X[final_selected], y], axis=1)\n", "df_selected.to_csv(CSV_PATH, index=False)\n" ] }, { "cell_type": "code", "execution_count": 8, "id": "e4d1ae48-f075-4e00-bf8a-6c774e38057e", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
phasemodelbest_paramscv_best_f1_macroaccuracyprecision_macrorecall_macrof1_macro
1selectedRandomForest{'clf__max_depth': 10, 'clf__min_samples_split...0.7528130.8881990.8009560.7241080.753571
2selectedSVC{'clf__C': 10.0, 'clf__gamma': 'scale', 'clf__...0.7670110.8832300.7980370.6945200.729340
0selectedLogReg{'clf__C': 10.0, 'clf__penalty': 'l1'}0.7387210.8795030.7963870.6723290.709255
\n", "
" ], "text/plain": [ " phase model best_params \\\n", "1 selected RandomForest {'clf__max_depth': 10, 'clf__min_samples_split... \n", "2 selected SVC {'clf__C': 10.0, 'clf__gamma': 'scale', 'clf__... \n", "0 selected LogReg {'clf__C': 10.0, 'clf__penalty': 'l1'} \n", "\n", " cv_best_f1_macro accuracy precision_macro recall_macro f1_macro \n", "1 0.752813 0.888199 0.800956 0.724108 0.753571 \n", "2 0.767011 0.883230 0.798037 0.694520 0.729340 \n", "0 0.738721 0.879503 0.796387 0.672329 0.709255 " ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "df = pd.read_csv(CSV_PATH)\n", "y = df[TARGET]\n", "X = df.drop(columns=[TARGET])\n", "\n", "# preprocessor\n", "num_cols = X.select_dtypes(include=[np.number]).columns.tolist()\n", "cat_cols = [c for c in X.columns if c not in num_cols]\n", "\n", "numeric_tf = Pipeline([(\"imputer\", SimpleImputer(strategy=\"median\")), (\"scaler\", StandardScaler())])\n", "categorical_tf = Pipeline([(\"imputer\", SimpleImputer(strategy=\"most_frequent\")), (\"onehot\", OneHotEncoder(handle_unknown=\"ignore\"))])\n", "\n", "preprocessor = ColumnTransformer([(\"num\", numeric_tf, num_cols), (\"cat\", categorical_tf, cat_cols)])\n", "\n", "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, stratify=y, random_state=RANDOM_STATE)\n", "\n", "models = {\n", " \"LogReg\": {\n", " \"pipe\": Pipeline([(\"prep\", preprocessor), (\"clf\", LogisticRegression(max_iter=2000, solver=\"liblinear\", random_state=RANDOM_STATE))]),\n", " \"grid\": {\"clf__penalty\": [\"l1\",\"l2\"], \"clf__C\": [0.1, 1.0, 10.0]}\n", " },\n", " \"RandomForest\": {\n", " \"pipe\": Pipeline([(\"prep\", preprocessor), (\"clf\", RandomForestClassifier(random_state=RANDOM_STATE))]),\n", " \"grid\": {\"clf__n_estimators\": [200, 400], \"clf__max_depth\": [None, 10], \"clf__min_samples_split\": [2, 5]}\n", " },\n", " \"SVC\": {\n", " \"pipe\": Pipeline([(\"prep\", preprocessor), (\"clf\", SVC(probability=False, random_state=RANDOM_STATE))]),\n", " \"grid\": {\"clf__kernel\": [\"rbf\",\"linear\"], \"clf__C\": [1.0, 10.0], \"clf__gamma\": [\"scale\",\"auto\"]}\n", " }\n", "}\n", "cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=RANDOM_STATE)\n", "\n", "def fit_grid_eval_selected(name, pipe, grid):\n", " gs = GridSearchCV(pipe, grid, scoring=\"f1_macro\", cv=cv, n_jobs=-1, verbose=0)\n", " gs.fit(X_train, y_train)\n", " best = gs.best_estimator_\n", " y_pred = best.predict(X_test)\n", " row = {\n", " \"phase\": \"selected\",\n", " \"model\": name,\n", " \"best_params\": str(gs.best_params_),\n", " \"cv_best_f1_macro\": gs.best_score_,\n", " \"accuracy\": accuracy_score(y_test, y_pred),\n", " \"precision_macro\": precision_score(y_test, y_pred, average=\"macro\", zero_division=0),\n", " \"recall_macro\": recall_score(y_test, y_pred, average=\"macro\", zero_division=0),\n", " \"f1_macro\": f1_score(y_test, y_pred, average=\"macro\", zero_division=0),\n", " }\n", " dump(best, os.path.join(OUT_DIR, f\"best_selected_{name}.joblib\"))\n", " with open(os.path.join(OUT_DIR, f\"report_selected_{name}.txt\"), \"w\", encoding=\"utf-8\") as f:\n", " f.write(classification_report(y_test, y_pred, zero_division=0))\n", " return row, confusion_matrix(y_test, y_pred)\n", "\n", "sel_rows, cms_sel = [], {}\n", "for name, spec in models.items():\n", " r, cm = fit_grid_eval_selected(name, spec[\"pipe\"], spec[\"grid\"])\n", " sel_rows.append(r); cms_sel[name] = cm\n", "\n", "df_selected_phase = pd.DataFrame(sel_rows).sort_values(\"f1_macro\", ascending=False)\n", "display(df_selected_phase)\n", "df_selected_phase.to_csv(os.path.join(OUT_DIR, \"results_selected.csv\"), index=False)\n" ] }, { "cell_type": "code", "execution_count": null, "id": "6fc7937c-b7d5-4c50-8dc0-1ae7bf03694d", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "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": "3.12.11" } }, "nbformat": 4, "nbformat_minor": 5 }