Obisidian vault auto-backup: 29-09-2025 16:38:58 on . 1 files edited

ISEN/IA/CIPA4/TP/TP2/tp2_IA.ipynb

@@ -18,7 +18,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 1,
    "id": "5a9a29a0",
    "metadata": {},
    "outputs": [
@@ -62,7 +62,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": null,
    "id": "5be8782b",
    "metadata": {},
    "outputs": [],
@@ -81,7 +81,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": null,
    "id": "70575266",
    "metadata": {},
    "outputs": [
@@ -102,7 +102,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": null,
    "id": "eaffc443",
    "metadata": {},
    "outputs": [
@@ -171,7 +171,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": null,
    "id": "d333dc4d",
    "metadata": {},
    "outputs": [
@@ -256,7 +256,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": null,
    "id": "c0966e7f",
    "metadata": {},
    "outputs": [
@@ -318,7 +318,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": null,
    "id": "b0560d20",
    "metadata": {},
    "outputs": [
@@ -399,7 +399,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": null,
    "id": "bff45d57",
    "metadata": {},
    "outputs": [
@@ -449,7 +449,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": null,
    "id": "e02058de",
    "metadata": {},
    "outputs": [
@@ -1218,7 +1218,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": null,
    "id": "e73e18a3",
    "metadata": {},
    "outputs": [
@@ -1296,7 +1296,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": null,
    "id": "acaeacd9",
    "metadata": {},
    "outputs": [
@@ -1322,7 +1322,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": null,
    "id": "e5949779",
    "metadata": {},
    "outputs": [
@@ -1370,7 +1370,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 15,
+   "execution_count": null,
    "id": "5733c7b1",
    "metadata": {},
    "outputs": [
@@ -1447,7 +1447,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": null,
    "id": "970fce9c",
    "metadata": {},
    "outputs": [
@@ -1583,7 +1583,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": null,
    "id": "b4a6c352",
    "metadata": {},
    "outputs": [
@@ -1677,7 +1677,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 18,
+   "execution_count": null,
    "id": "1ad09da3",
    "metadata": {},
    "outputs": [
@@ -1777,7 +1777,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 19,
+   "execution_count": null,
    "id": "0a4ec288",
    "metadata": {},
    "outputs": [
@@ -1941,7 +1941,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 20,
+   "execution_count": null,
    "id": "7dc960b0",
    "metadata": {},
    "outputs": [
@@ -2037,7 +2037,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 22,
+   "execution_count": null,
    "id": "6f487ebf",
    "metadata": {},
    "outputs": [
@@ -2093,9 +2093,29 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
+      "/Volumes/SSD/Nextcloud/Documents/ISEN/Cours/Obsidian Vault/ISEN/IA/CIPA4/TP/TP2/.venv/lib/python3.13/site-packages/sklearn/linear_model/_stochastic_gradient.py:726: ConvergenceWarning: Maximum number of iteration reached before convergence. Consider increasing max_iter to improve the fit.\n",
+      "  warnings.warn(\n",
       "/Volumes/SSD/Nextcloud/Documents/ISEN/Cours/Obsidian Vault/ISEN/IA/CIPA4/TP/TP2/.venv/lib/python3.13/site-packages/sklearn/linear_model/_stochastic_gradient.py:726: ConvergenceWarning: Maximum number of iteration reached before convergence. Consider increasing max_iter to improve the fit.\n",
       "  warnings.warn(\n"
      ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "c. Évaluation avec données standardisées:\n",
+      "Fold 1: 0.9013 (90.14%)\n",
+      "Fold 2: 0.8939 (89.39%)\n",
+      "Fold 3: 0.9055 (90.55%)\n",
+      "Moyenne: 0.9002 (90.02%)\n",
+      "\n",
+      "=== Comparaison des résultats ===\n",
+      "Sans standardisation: 0.8721 (87.21%)\n",
+      "Avec standardisation: 0.9002 (90.02%)\n",
+      "Amélioration: +0.0281 (+2.81 points de %)\n",
+      "✅ Les résultats sont MEILLEURS avec la standardisation\n"
+     ]
     }
    ],
    "source": [
@@ -2114,7 +2134,7 @@
     "# c. Évaluer le classifieur SGD avec les données standardisées\n",
     "scores_scaled = cross_val_score(sgd_multiclass, X_train_scaled, Y_train, cv=3, scoring='accuracy')\n",
     "\n",
-    "print(\"\\nc. Évaluation avec données standardisées:\")\n",
+    "print(\"\\nÉvaluation avec données standardisées:\")\n",
     "for i, score in enumerate(scores_scaled, 1):\n",
     "    print(f\"Fold {i}: {score:.4f} ({score*100:.2f}%)\")\n",
     "\n",
@@ -2135,11 +2155,159 @@
     "    print(\"❌ Les résultats ne sont PAS meilleurs avec la standardisation\")"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "ab0c9c23",
+   "metadata": {},
+   "source": [
+    "### 2-2- Matrice de confusion"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
    "id": "55bf66d7",
    "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "=== Matrice de confusion du classifieur SGD multi-classes ===\n",
+      "\n",
+      "Prédiction des classes avec cross_val_predict...\n"
+     ]
+    },
+    {
+     "ename": "",
+     "evalue": "",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[1;31mLe noyau s’est bloqué lors de l’exécution du code dans une cellule active ou une cellule précédente. \n",
+      "\u001b[1;31mVeuillez vérifier le code dans la ou les cellules pour identifier une cause possible de l’échec. \n",
+      "\u001b[1;31mCliquez <a href='https://aka.ms/vscodeJupyterKernelCrash'>ici</a> pour plus d’informations. \n",
+      "\u001b[1;31mPour plus d’informations, consultez Jupyter <a href='command:jupyter.viewOutput'>log</a>."
+     ]
+    }
+   ],
+   "source": [
+    "from sklearn.model_selection import cross_val_predict\n",
+    "from sklearn.metrics import confusion_matrix\n",
+    "import matplotlib.pyplot as plt\n",
+    "import numpy as np\n",
+    "\n",
+    "print(\"=== Matrice de confusion du classifieur SGD multi-classes ===\\n\")\n",
+    "\n",
+    "# Prédire les classes avec cross_val_predict pour le modèle multi-classes\n",
+    "print(\"Prédiction des classes avec cross_val_predict...\")\n",
+    "y_train_pred_multiclass = cross_val_predict(sgd_multiclass, X_train, Y_train, cv=3)\n",
+    "\n",
+    "print(f\"Forme des prédictions: {y_train_pred_multiclass.shape}\")\n",
+    "print(f\"Classes prédites uniques: {np.unique(y_train_pred_multiclass)}\")\n",
+    "\n",
+    "# a. Calculer la matrice de confusion normale et normalisée\n",
+    "cm_multiclass = confusion_matrix(Y_train, y_train_pred_multiclass)\n",
+    "print(\"\\na. Matrice de confusion (valeurs absolues) 10x10:\")\n",
+    "print(cm_multiclass)\n",
+    "\n",
+    "cm_normalized_multiclass = confusion_matrix(Y_train, y_train_pred_multiclass, normalize='true')\n",
+    "print(\"\\nMatrice de confusion normalisée:\")\n",
+    "print(cm_normalized_multiclass)\n",
+    "\n",
+    "# Visualisation des matrices de confusion\n",
+    "fig, axes = plt.subplots(1, 2, figsize=(15, 6))\n",
+    "\n",
+    "# Matrice de confusion absolue\n",
+    "im1 = axes[0].imshow(cm_multiclass, interpolation='nearest', cmap=plt.cm.Blues)\n",
+    "axes[0].set_title('Matrice de confusion (valeurs absolues)')\n",
+    "axes[0].set_xlabel('Classe prédite')\n",
+    "axes[0].set_ylabel('Classe réelle')\n",
+    "\n",
+    "# Ajouter les annotations pour les valeurs\n",
+    "for i in range(10):\n",
+    "    for j in range(10):\n",
+    "        axes[0].text(j, i, format(cm_multiclass[i, j], 'd'), \n",
+    "                    ha=\"center\", va=\"center\", \n",
+    "                    color=\"white\" if cm_multiclass[i, j] > cm_multiclass.max()/2 else \"black\",\n",
+    "                    fontsize=8)\n",
+    "\n",
+    "axes[0].set_xticks(range(10))\n",
+    "axes[0].set_yticks(range(10))\n",
+    "axes[0].set_xticklabels(range(10))\n",
+    "axes[0].set_yticklabels(range(10))\n",
+    "\n",
+    "# Matrice de confusion normalisée\n",
+    "im2 = axes[1].imshow(cm_normalized_multiclass, interpolation='nearest', cmap=plt.cm.Blues)\n",
+    "axes[1].set_title('Matrice de confusion normalisée')\n",
+    "axes[1].set_xlabel('Classe prédite')\n",
+    "axes[1].set_ylabel('Classe réelle')\n",
+    "\n",
+    "# Ajouter les annotations pour les pourcentages\n",
+    "for i in range(10):\n",
+    "    for j in range(10):\n",
+    "        axes[1].text(j, i, format(cm_normalized_multiclass[i, j], '.2f'), \n",
+    "                    ha=\"center\", va=\"center\", \n",
+    "                    color=\"white\" if cm_normalized_multiclass[i, j] > 0.5 else \"black\",\n",
+    "                    fontsize=8)\n",
+    "\n",
+    "axes[1].set_xticks(range(10))\n",
+    "axes[1].set_yticks(range(10))\n",
+    "axes[1].set_xticklabels(range(10))\n",
+    "axes[1].set_yticklabels(range(10))\n",
+    "\n",
+    "plt.tight_layout()\n",
+    "plt.show()\n",
+    "\n",
+    "# b. Interprétation des résultats\n",
+    "print(\"\\n=== Interprétation des résultats ===\")\n",
+    "\n",
+    "# Calcul des métriques globales\n",
+    "accuracy_total = np.trace(cm_multiclass) / np.sum(cm_multiclass)\n",
+    "print(f\"\\nAccuracy globale: {accuracy_total:.4f} ({accuracy_total*100:.2f}%)\")\n",
+    "\n",
+    "# Analyse par classe (diagonale = bonnes prédictions)\n",
+    "print(f\"\\nPerformances par classe (rappel):\")\n",
+    "for i in range(10):\n",
+    "    rappel_classe = cm_normalized_multiclass[i, i]\n",
+    "    nb_instances = cm_multiclass[i, :].sum()\n",
+    "    print(f\"  Classe {i}: {rappel_classe:.3f} ({rappel_classe*100:.1f}%) - {nb_instances} instances\")\n",
+    "\n",
+    "# Classes les mieux classées\n",
+    "diagonale = np.diag(cm_normalized_multiclass)\n",
+    "meilleures_classes = np.argsort(diagonale)[::-1][:3]\n",
+    "pires_classes = np.argsort(diagonale)[:3]\n",
+    "\n",
+    "print(f\"\\nMeilleures classes (rappel élevé):\")\n",
+    "for classe in meilleures_classes:\n",
+    "    print(f\"  Classe {classe}: {diagonale[classe]:.3f} ({diagonale[classe]*100:.1f}%)\")\n",
+    "\n",
+    "print(f\"\\nPires classes (rappel faible):\")\n",
+    "for classe in pires_classes:\n",
+    "    print(f\"  Classe {classe}: {diagonale[classe]:.3f} ({diagonale[classe]*100:.1f}%)\")\n",
+    "\n",
+    "# Erreurs les plus fréquentes (hors diagonale)\n",
+    "cm_erreurs = cm_multiclass.copy()\n",
+    "np.fill_diagonal(cm_erreurs, 0)  # Enlever la diagonale\n",
+    "erreurs_max = np.unravel_index(np.argmax(cm_erreurs), cm_erreurs.shape)\n",
+    "\n",
+    "print(f\"\\nErreurs les plus fréquentes:\")\n",
+    "indices_erreurs = np.argsort(cm_erreurs.ravel())[::-1][:5]\n",
+    "for idx in indices_erreurs:\n",
+    "    i, j = np.unravel_index(idx, cm_erreurs.shape)\n",
+    "    if cm_erreurs[i, j] > 0:\n",
+    "        print(f\"  Classe {i} prédite comme {j}: {cm_erreurs[i, j]} fois\")\n",
+    "\n",
+    "print(f\"\\n=== Conclusion ===\")\n",
+    "print(\"La matrice 10x10 montre les performances du classifieur sur les 10 chiffres.\")\n",
+    "print(\"La diagonale représente les bonnes classifications.\")\n",
+    "print(\"Les valeurs hors diagonale montrent les confusions entre classes.\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "16a1368f",
+   "metadata": {},
    "outputs": [],
    "source": []
   }