feat: add new structure for augmentation. #6

notebooks/07_augmentation_clean_v2.ipynb

@@ -4,8 +4,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# 🔁 Notebook 07 — Data Augmentation\n",
-    "## YouTube Hate Speech Detection\n",
+    "# 🔁 Data Augmentation\n",
     "\n",
     "### ¿Qué hace este notebook?\n",
     "Evaluamos **3 estrategias de augmentation** sobre la clase tóxica del train set\n",
@@ -23,11 +22,6 @@
     "- EDA: mínima perturbación, preserva semántica\n",
     "- Back-translation: mejor calidad semántica, parafrasea naturalmente\n",
     "\n",
-    "### Limitación conocida\n",
-    "El augmentation se aplica **antes** del CV, no dentro del loop.\n",
-    "Con 1000 muestras implementarlo dentro del loop es costoso.\n",
-    "Se documenta como limitación metodológica.\n",
-    "\n",
     "### Modelo de referencia\n",
     "LR tuned cargado desde `final_model.joblib` — F1 test ≈ 0.7579."
    ]
@@ -151,7 +145,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -195,11 +189,13 @@
     "    f1_tr  = f1_score(y_tr, pred_train, average='weighted')\n",
     "    roc    = roc_auc_score(y_te, pipeline.predict_proba(X_te)[:,1])\n",
     "    cv_mean = cv_std = cv_gap = None\n",
+    "\n",
     "    if cv_scores is not None:\n",
     "        cv_mean = cv_scores['test_score'].mean()\n",
     "        cv_std  = cv_scores['test_score'].std()\n",
     "        cv_gap  = abs(cv_mean - f1_te) * 100\n",
     "    return {\n",
+    "        \n",
     "        'name'             : name,\n",
     "        'f1_test'          : round(f1_te, 4),\n",
     "        'f1_train'         : round(f1_tr, 4),\n",
@@ -312,11 +308,7 @@
    "source": [
     "## 5. Estrategia 1 — Synonym Replacement (WordNet)\n",
     "\n",
-    "Reemplaza palabras aleatorias por sinónimos usando WordNet.\n",
-    "\n",
-    "**Limitación conocida:** WordNet tiene cobertura baja en\n",
-    "lenguaje coloquial y jerga (palabras como 'thug', 'bullshit').\n",
-    "Los sinónimos formales pueden cambiar el estilo y confundir al modelo."
+    "Reemplaza palabras aleatorias por sinónimos usando WordNet."
    ]
   },
   {
@@ -370,7 +362,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": null,
    "metadata": {},
    "outputs": [
     {
@@ -401,6 +393,7 @@
     "                       X_test, y_test, 'LR + WordNet', cv_wn)\n",
     "\n",
     "delta = metrics_wn['f1_test'] - metrics_base['f1_test']\n",
+    "\n",
     "print(f\"WordNet: F1={metrics_wn['f1_test']:.4f} ({delta*100:+.2f}pp) | \"\n",
     "      f\"FN={metrics_wn['fn']} | FP={metrics_wn['fp']}\")"
    ]
@@ -537,8 +530,7 @@
     "- Los insultos raciales en inglés no siempre tienen equivalente directo\n",
     "  en español → el modelo de traducción busca el contexto más cercano\n",
     "- Añade variedad sintáctica real, no solo léxica\n",
-    "\n",
-    "**Requiere:** `pip install deep-translator` e internet."
+    "\n"
    ]
   },
   {

notebooks/08_transformers_v2.ipynb

@@ -25,8 +25,7 @@
     "10. Evaluación RoBERTa Hate  \n",
     "11. Comparación de modelos  \n",
     "12. Error Analysis  \n",
-    "13. Guardado del mejor modelo  \n",
-    "14. Conclusiones\n"
+    "13. Guardado del mejor modelo  \n"
    ]
   },
   {
@@ -39,7 +38,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 68,
+   "execution_count": null,
    "id": "072caf60",
    "metadata": {},
    "outputs": [
@@ -57,7 +56,6 @@
     }
    ],
    "source": [
-    "\n",
     "import os\n",
     "import sys\n",
     "import yaml\n",
@@ -117,7 +115,7 @@
     "if torch.cuda.is_available():\n",
     "    print(\"GPU:\", torch.cuda.get_device_name(0))\n",
     "else:\n",
-    "    print(\"⚠️ GPU no detectada\")\n"
+    "    print(\"⚠️ GPU no detectada\")"
    ]
   },
   {
@@ -130,7 +128,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 69,
+   "execution_count": null,
    "id": "0b9084dc",
    "metadata": {},
    "outputs": [
@@ -143,7 +141,6 @@
     }
    ],
    "source": [
-    "\n",
     "def set_seed(seed=42):\n",
     "\n",
     "    random.seed(seed)\n",
@@ -160,7 +157,7 @@
     "\n",
     "set_seed(RAND)\n",
     "\n",
-    "print(\"Seed configurado:\", RAND)\n"
+    "print(\"Seed configurado:\", RAND)"
    ]
   },
   {
@@ -173,7 +170,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 70,
+   "execution_count": null,
    "id": "0fb40c48",
    "metadata": {},
    "outputs": [
@@ -310,25 +307,13 @@
     }
    ],
    "source": [
-    "\n",
-    "DATA_PATH = (\n",
-    "    PROJECT_ROOT\n",
-    "    / \"data\"\n",
-    "    / \"processed\"\n",
-    "    / \"v2\"\n",
-    "    / \"comments_preprocessed.csv\"\n",
-    ")\n",
+    "DATA_PATH = (PROJECT_ROOT / \"data\" / \"processed\" / \"v2\" / \"comments_preprocessed.csv\")\n",
     "\n",
     "df = pd.read_csv(DATA_PATH)\n",
     "\n",
     "TEXT_COL = \"Text\"\n",
     "\n",
-    "df[TEXT_COL] = (\n",
-    "    df[TEXT_COL]\n",
-    "    .fillna(\"\")\n",
-    "    .astype(str)\n",
-    "    .str.strip()\n",
-    ")\n",
+    "df[TEXT_COL] = (df[TEXT_COL].fillna(\"\").astype(str).str.strip())\n",
     "\n",
     "df = df[df[TEXT_COL] != \"\"].copy()\n",
     "\n",
@@ -336,7 +321,7 @@
     "\n",
     "print(df.shape)\n",
     "\n",
-    "df.head()\n"
+    "df.head()"
    ]
   },
   {
@@ -349,7 +334,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 71,
+   "execution_count": null,
    "id": "85c08f41",
    "metadata": {},
    "outputs": [
@@ -367,14 +352,10 @@
     }
    ],
    "source": [
-    "\n",
     "X = df[TEXT_COL]\n",
     "y = df[TARGET]\n",
     "\n",
-    "# -------------------------------------------------\n",
     "# TEST FINAL\n",
-    "# -------------------------------------------------\n",
-    "\n",
     "X_temp, X_test, y_temp, y_test = train_test_split(\n",
     "    X,\n",
     "    y,\n",
@@ -383,10 +364,7 @@
     "    random_state=RAND,\n",
     ")\n",
     "\n",
-    "# -------------------------------------------------\n",
     "# VALIDATION\n",
-    "# -------------------------------------------------\n",
-    "\n",
     "X_train, X_valid, y_train, y_valid = train_test_split(\n",
     "    X_temp,\n",
     "    y_temp,\n",
@@ -401,7 +379,7 @@
     "\n",
     "print(\"Train:\", len(X_train))\n",
     "print(\"Validation:\", len(X_valid))\n",
-    "print(\"Test:\", len(X_test))\n"
+    "print(\"Test:\", len(X_test))"
    ]
   },
   {
@@ -414,7 +392,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 72,
+   "execution_count": null,
    "id": "4372ea97",
    "metadata": {},
    "outputs": [
@@ -447,7 +425,6 @@
     }
    ],
    "source": [
-    "\n",
     "print(\"=\" * 50)\n",
     "print(\"CLASS DISTRIBUTION\")\n",
     "print(\"=\" * 50)\n",
@@ -459,12 +436,12 @@
     "print(y_valid.value_counts(normalize=True))\n",
     "\n",
     "print(\"\\nTest\")\n",
-    "print(y_test.value_counts(normalize=True))\n"
+    "print(y_test.value_counts(normalize=True))"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 73,
+   "execution_count": null,
    "id": "49b2f924",
    "metadata": {},
    "outputs": [
@@ -495,7 +472,6 @@
     }
    ],
    "source": [
-    "\n",
     "train_lengths = X_train.str.split().apply(len)\n",
     "\n",
     "print(train_lengths.describe())\n",
@@ -507,7 +483,7 @@
     "plt.title(\"Distribución de longitud de comentarios\")\n",
     "plt.xlabel(\"Número de palabras\")\n",
     "\n",
-    "plt.show()\n"
+    "plt.show()"
    ]
   },
   {
@@ -525,7 +501,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "\n",
     "def build_hf_dataset(X, y):\n",
     "\n",
     "    df_local = pd.DataFrame({\n",
@@ -639,21 +614,14 @@
     "    plt.show()\n",
     "\n",
     "    return {\n",
-    "\n",
     "        \"accuracy\": accuracy_score(y_test, preds),\n",
-    "\n",
     "        \"precision\": precision_score(y_test, preds),\n",
-    "\n",
     "        \"recall\": recall_score(y_test, preds),\n",
-    "\n",
     "        \"f1\": f1_score(y_test, preds),\n",
-    "\n",
     "        \"roc_auc\": roc_auc_score(y_test, probs),\n",
-    "\n",
     "        \"preds\": preds,\n",
-    "\n",
     "        \"probs\": probs,\n",
-    "    }\n"
+    "    }"
    ]
   },
   {
@@ -666,7 +634,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 75,
+   "execution_count": null,
    "id": "f6f76741",
    "metadata": {},
    "outputs": [
@@ -682,12 +650,11 @@
     }
    ],
    "source": [
-    "\n",
     "hf_train_raw = build_hf_dataset(X_train, y_train)\n",
     "hf_valid_raw = build_hf_dataset(X_valid, y_valid)\n",
     "hf_test_raw  = build_hf_dataset(X_test, y_test)\n",
     "\n",
-    "print(hf_train_raw)\n"
+    "print(hf_train_raw)"
    ]
   },
   {
@@ -709,7 +676,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 76,
+   "execution_count": null,
    "id": "1e8f8d34",
    "metadata": {},
    "outputs": [
@@ -722,7 +689,6 @@
     }
    ],
    "source": [
-    "\n",
     "DISTIL_MODEL = \"distilbert-base-uncased\"\n",
     "\n",
     "MAX_LEN = 128\n",
@@ -730,12 +696,12 @@
     "EPOCHS = 3\n",
     "LR = 2e-5\n",
     "\n",
-    "print(DISTIL_MODEL)\n"
+    "print(DISTIL_MODEL)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 77,
+   "execution_count": null,
    "id": "b9ee6361",
    "metadata": {},
    "outputs": [
@@ -750,7 +716,6 @@
     }
    ],
    "source": [
-    "\n",
     "distil_tokenizer = AutoTokenizer.from_pretrained(\n",
     "    DISTIL_MODEL\n",
     ")\n",
@@ -771,25 +736,12 @@
     "    hf_test_raw,\n",
     "    distil_tokenizer,\n",
     "    MAX_LEN,\n",
-    ")\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 78,
-   "id": "02d2d984",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "\n",
-    "distil_collator = DataCollatorWithPadding(\n",
-    "    tokenizer=distil_tokenizer\n",
-    ")\n"
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 79,
+   "execution_count": null,
    "id": "8e867b40",
    "metadata": {},
    "outputs": [
@@ -825,6 +777,7 @@
     }
    ],
    "source": [
+    "distil_collator = DataCollatorWithPadding(tokenizer=distil_tokenizer)\n",
     "\n",
     "distil_model = AutoModelForSequenceClassification.from_pretrained(\n",
     "    DISTIL_MODEL,\n",
@@ -833,12 +786,12 @@
     "\n",
     "distil_model.to(device)\n",
     "\n",
-    "print(distil_model.__class__.__name__)\n"
+    "print(distil_model.__class__.__name__)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 80,
+   "execution_count": null,
    "id": "ed09755c",
    "metadata": {},
    "outputs": [
@@ -851,7 +804,6 @@
     }
    ],
    "source": [
-    "\n",
     "distil_args = TrainingArguments(\n",
     "\n",
     "    output_dir= PROJECT_ROOT / \"models\" / \"distilbert_results\",\n",
@@ -882,29 +834,17 @@
     "    report_to=\"none\",\n",
     "\n",
     "    seed=RAND,\n",
-    ")\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 81,
-   "id": "000c6a6b",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "\n",
-    "distil_early_stopping = EarlyStoppingCallback(\n",
-    "    early_stopping_patience=2\n",
-    ")\n"
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 82,
+   "execution_count": null,
    "id": "827eed3d",
    "metadata": {},
    "outputs": [],
    "source": [
+    "distil_early_stopping = EarlyStoppingCallback(early_stopping_patience=2)\n",
     "\n",
     "distil_trainer = Trainer(\n",
     "\n",
@@ -922,12 +862,12 @@
     "    compute_metrics=compute_metrics,\n",
     "\n",
     "    callbacks=[distil_early_stopping],\n",
-    ")\n"
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 83,
+   "execution_count": null,
    "id": "eea8b514",
    "metadata": {},
    "outputs": [
@@ -1024,12 +964,11 @@
     }
    ],
    "source": [
-    "\n",
     "print(\"=\" * 50)\n",
     "print(\"TRAINING DISTILBERT\")\n",
     "print(\"=\" * 50)\n",
     "\n",
-    "distil_trainer.train()\n"
+    "distil_trainer.train()"
    ]
   },
   {
@@ -1042,7 +981,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 84,
+   "execution_count": null,
    "id": "9647c651",
    "metadata": {},
    "outputs": [
@@ -1086,13 +1025,12 @@
     }
    ],
    "source": [
-    "\n",
     "distil_results = evaluate_model(\n",
     "    distil_trainer,\n",
     "    distil_test,\n",
     "    y_test,\n",
     "    \"DistilBERT\",\n",
-    ")\n"
+    ")"
    ]
   },
   {
@@ -1114,7 +1052,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 85,
+   "execution_count": null,
    "id": "40cbdf8d",
    "metadata": {},
    "outputs": [
@@ -1127,7 +1065,6 @@
     }
    ],
    "source": [
-    "\n",
     "HATE_MODEL = \"cardiffnlp/twitter-roberta-base-hate\"\n",
     "\n",
     "MAX_LEN = 128\n",
@@ -1135,12 +1072,12 @@
     "EPOCHS = 3\n",
     "LR = 2e-5\n",
     "\n",
-    "print(HATE_MODEL)\n"
+    "print(HATE_MODEL)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 86,
+   "execution_count": null,
    "id": "6d0d9245",
    "metadata": {},
    "outputs": [
@@ -1155,7 +1092,6 @@
     }
    ],
    "source": [
-    "\n",
     "hate_tokenizer = AutoTokenizer.from_pretrained(\n",
     "    HATE_MODEL\n",
     ")\n",
@@ -1176,25 +1112,12 @@
     "    hf_test_raw,\n",
     "    hate_tokenizer,\n",
     "    MAX_LEN,\n",
-    ")\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 87,
-   "id": "1052b914",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "\n",
-    "hate_collator = DataCollatorWithPadding(\n",
-    "    tokenizer=hate_tokenizer\n",
-    ")\n"
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 88,
+   "execution_count": null,
    "id": "5114bb65",
    "metadata": {},
    "outputs": [
@@ -1214,6 +1137,7 @@
     }
    ],
    "source": [
+    "hate_collator = DataCollatorWithPadding(tokenizer=hate_tokenizer)\n",
     "\n",
     "hate_model = AutoModelForSequenceClassification.from_pretrained(\n",
     "    HATE_MODEL,\n",
@@ -1223,7 +1147,7 @@
     "\n",
     "hate_model.to(device)\n",
     "\n",
-    "print(hate_model.__class__.__name__)\n"
+    "print(hate_model.__class__.__name__)"
    ]
   },
   {
@@ -1236,7 +1160,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 89,
+   "execution_count": null,
    "id": "c3729fb1",
    "metadata": {},
    "outputs": [
@@ -1253,18 +1177,12 @@
     }
    ],
    "source": [
-    "\n",
-    "# -----------------------------------------------------\n",
     "# Congelar backbone\n",
-    "# -----------------------------------------------------\n",
     "\n",
     "for param in hate_model.base_model.parameters():\n",
     "    param.requires_grad = False\n",
     "\n",
-    "# -----------------------------------------------------\n",
     "# Classification head entrenable\n",
-    "# -----------------------------------------------------\n",
-    "\n",
     "classifier_found = False\n",
     "\n",
     "for name, param in hate_model.named_parameters():\n",
@@ -1279,10 +1197,7 @@
     "else:\n",
     "    print(\"⚠️ No se encontró classifier head\")\n",
     "\n",
-    "# -----------------------------------------------------\n",
     "# Verificación\n",
-    "# -----------------------------------------------------\n",
-    "\n",
     "total_params = 0\n",
     "trainable_params = 0\n",
     "\n",
@@ -1298,12 +1213,12 @@
     "print()\n",
     "print(f\"Trainable params: {trainable_params:,}\")\n",
     "print(f\"Total params: {total_params:,}\")\n",
-    "print(f\"Trainable %: {pct:.2f}%\")\n"
+    "print(f\"Trainable %: {pct:.2f}%\")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 90,
+   "execution_count": null,
    "id": "7a0a4821",
    "metadata": {},
    "outputs": [
@@ -1316,7 +1231,6 @@
     }
    ],
    "source": [
-    "\n",
     "hate_args = TrainingArguments(\n",
     "\n",
     "    output_dir= PROJECT_ROOT / \"models\" / \"roberta_hate_results\",\n",
@@ -1347,29 +1261,17 @@
     "    report_to=\"none\",\n",
     "\n",
     "    seed=RAND,\n",
-    ")\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 91,
-   "id": "13ffac9e",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "\n",
-    "hate_early_stopping = EarlyStoppingCallback(\n",
-    "    early_stopping_patience=2\n",
-    ")\n"
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 92,
+   "execution_count": null,
    "id": "6a3acd3d",
    "metadata": {},
    "outputs": [],
    "source": [
+    "hate_early_stopping = EarlyStoppingCallback(early_stopping_patience=2)\n",
     "\n",
     "hate_trainer = Trainer(\n",
     "\n",
@@ -1387,12 +1289,12 @@
     "    compute_metrics=compute_metrics,\n",
     "\n",
     "    callbacks=[hate_early_stopping],\n",
-    ")\n"
+    ")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 93,
+   "execution_count": null,
    "id": "a5752cbb",
    "metadata": {},
    "outputs": [
@@ -1489,12 +1391,11 @@
     }
    ],
    "source": [
-    "\n",
     "print(\"=\" * 50)\n",
     "print(\"TRAINING ROBERTA HATE\")\n",
     "print(\"=\" * 50)\n",
     "\n",
-    "hate_trainer.train()\n"
+    "hate_trainer.train()"
    ]
   },
   {
@@ -1507,7 +1408,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 94,
+   "execution_count": null,
    "id": "d58a64a5",
    "metadata": {},
    "outputs": [
@@ -1551,13 +1452,12 @@
     }
    ],
    "source": [
-    "\n",
     "hate_results = evaluate_model(\n",
     "    hate_trainer,\n",
     "    hate_test,\n",
     "    y_test,\n",
     "    \"RoBERTa Hate\",\n",
-    ")\n"
+    ")"
    ]
   },
   {
@@ -1570,7 +1470,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 95,
+   "execution_count": null,
    "id": "d7f2b05f",
    "metadata": {},
    "outputs": [
@@ -1638,7 +1538,6 @@
     }
    ],
    "source": [
-    "\n",
     "comparison_df = pd.DataFrame({\n",
     "\n",
     "    \"Model\": [\n",
@@ -1672,12 +1571,12 @@
     "    ],\n",
     "})\n",
     "\n",
-    "comparison_df\n"
+    "comparison_df"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 96,
+   "execution_count": null,
    "id": "2c757b76",
    "metadata": {},
    "outputs": [
@@ -1693,7 +1592,6 @@
     }
    ],
    "source": [
-    "\n",
     "plt.figure(figsize=(8, 5))\n",
     "\n",
     "sns.barplot(\n",
@@ -1703,8 +1601,7 @@
     ")\n",
     "\n",
     "plt.title(\"Comparación F1 Score\")\n",
-    "\n",
-    "plt.show()\n"
+    "plt.show()"
    ]
   },
   {
@@ -1717,7 +1614,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 97,
+   "execution_count": null,
    "id": "0f32e457",
    "metadata": {},
    "outputs": [
@@ -1730,29 +1627,24 @@
     }
    ],
    "source": [
-    "\n",
     "if hate_results[\"f1\"] >= distil_results[\"f1\"]:\n",
     "\n",
     "    best_name = \"RoBERTa Hate\"\n",
-    "\n",
     "    best_preds = hate_results[\"preds\"]\n",
-    "\n",
     "    best_probs = hate_results[\"probs\"]\n",
     "\n",
     "else:\n",
     "\n",
     "    best_name = \"DistilBERT\"\n",
-    "\n",
     "    best_preds = distil_results[\"preds\"]\n",
-    "\n",
     "    best_probs = distil_results[\"probs\"]\n",
     "\n",
-    "print(\"Best model:\", best_name)\n"
+    "print(\"Best model:\", best_name)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 98,
+   "execution_count": null,
    "id": "1d7148ce",
    "metadata": {},
    "outputs": [
@@ -1858,15 +1750,11 @@
     }
    ],
    "source": [
-    "\n",
     "error_df = pd.DataFrame({\n",
     "\n",
     "    \"text\": X_test.values,\n",
-    "\n",
     "    \"real\": y_test.values,\n",
-    "\n",
     "    \"pred\": best_preds,\n",
-    "\n",
     "    \"prob_toxic\": best_probs,\n",
     "})\n",
     "\n",
@@ -1875,13 +1763,12 @@
     ")\n",
     "\n",
     "print(\"Errores:\", error_df[\"is_error\"].sum())\n",
-    "\n",
-    "error_df.head()\n"
+    "error_df.head()"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 99,
+   "execution_count": null,
    "id": "64457e22",
    "metadata": {},
    "outputs": [
@@ -1940,12 +1827,7 @@
     }
    ],
    "source": [
-    "\n",
-    "false_negatives = error_df[\n",
-    "    (error_df[\"real\"] == 1)\n",
-    "    &\n",
-    "    (error_df[\"pred\"] == 0)\n",
-    "]\n",
+    "false_negatives = error_df[(error_df[\"real\"] == 1) & (error_df[\"pred\"] == 0)]\n",
     "\n",
     "print(\"=\" * 80)\n",
     "print(\"FALSE NEGATIVES\")\n",
@@ -1954,9 +1836,7 @@
     "for idx, row in false_negatives.head(10).iterrows():\n",
     "\n",
     "    print(\"\\nProb toxicidad:\", round(row[\"prob_toxic\"], 4))\n",
-    "\n",
     "    print(\"-\" * 60)\n",
-    "\n",
     "    print(row[\"text\"])\n"
    ]
   },
@@ -1970,7 +1850,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 100,
+   "execution_count": null,
    "id": "8cc4001e",
    "metadata": {},
    "outputs": [
@@ -1997,7 +1877,6 @@
     }
    ],
    "source": [
-    "\n",
     "SAVE_DIR = PROJECT_ROOT / \"models\"\n",
     "\n",
     "if best_name == \"RoBERTa Hate\":\n",
@@ -2015,10 +1894,8 @@
     "    save_path = SAVE_DIR / \"best_distilbert\"\n",
     "\n",
     "final_model.save_model(save_path)\n",
-    "\n",
     "final_tokenizer.save_pretrained(save_path)\n",
-    "\n",
-    "print(\"Modelo guardado en:\", save_path)\n"
+    "print(\"Modelo guardado en:\", save_path)"
    ]
   }
  ],