File size: 12,930 Bytes
eccd289 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 |
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "b2b05c00",
"metadata": {},
"outputs": [],
"source": [
"!pip install -q transformers torch huggingface_hub pandas numpy kaggle\n",
"\n",
"import os\n",
"from pathlib import Path\n",
"\n",
"kaggle_json_path = Path.home() / '.kaggle' / 'kaggle.json'\n",
"\n",
"if not kaggle_json_path.exists():\n",
" print(\"Kaggle credentials not found.\")\n",
" print(\"\\nIf you have kaggle.json in the current directory:\")\n",
" if Path('kaggle.json').exists():\n",
" kaggle_json_path.parent.mkdir(exist_ok=True, parents=True)\n",
" import shutil\n",
" shutil.copy('kaggle.json', kaggle_json_path)\n",
" kaggle_json_path.chmod(0o600)\n",
" print(\"Kaggle credentials configured\")\n",
" else:\n",
" print(\"\\nPlease upload kaggle.json to this directory, then re-run this cell.\")\n",
" print(\"Download from: https://www.kaggle.com/settings\")\n",
"else:\n",
" print(\"Kaggle credentials found\")\n",
"\n",
"import numpy as np\n",
"import pandas as pd\n",
"import torch\n",
"from transformers import AutoTokenizer, AutoModelForSequenceClassification\n",
"from torch.cuda.amp import autocast\n",
"from huggingface_hub import hf_hub_download\n",
"import warnings\n",
"\n",
"warnings.filterwarnings(\"ignore\")\n",
"\n",
"class Config:\n",
" HF_REPO_ID = \"YOUR_USERNAME/emotion-classifier-deberta-v3\"\n",
" COMPETITION_NAME = \"2025-sep-dl-gen-ai-project\"\n",
" LABELS = [\"anger\", \"fear\", \"joy\", \"sadness\", \"surprise\"]\n",
" MAX_LEN = 128\n",
" BATCH_SIZE = 32\n",
" TEST_CSV = \"/kaggle/input/2025-sep-dl-gen-ai-project/test.csv\"\n",
" SUBMISSION_PATH = \"submission.csv\"\n",
"\n",
"CONFIG = Config()\n",
"\n",
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
"print(f\"Using device: {device}\")\n",
"if torch.cuda.is_available():\n",
" print(f\"GPU: {torch.cuda.get_device_name(0)}\")\n",
"\n",
"print(f\"Loading model from HuggingFace: {CONFIG.HF_REPO_ID}\")\n",
"\n",
"try:\n",
" print(\" Loading model...\")\n",
" model = AutoModelForSequenceClassification.from_pretrained(\n",
" CONFIG.HF_REPO_ID,\n",
" num_labels=len(CONFIG.LABELS),\n",
" problem_type=\"multi_label_classification\"\n",
" )\n",
" model.to(device)\n",
" model.eval()\n",
" print(\" Model loaded\")\n",
" \n",
" print(\" Loading tokenizer...\")\n",
" tokenizer = AutoTokenizer.from_pretrained(CONFIG.HF_REPO_ID)\n",
" print(\" Tokenizer loaded\")\n",
" \n",
" print(\" Loading optimized thresholds...\")\n",
" try:\n",
" threshold_path = hf_hub_download(\n",
" repo_id=CONFIG.HF_REPO_ID,\n",
" filename=\"best_thresholds.npy\"\n",
" )\n",
" best_thresholds = np.load(threshold_path)\n",
" print(\" Optimized thresholds loaded\")\n",
" print(f\"\\n Thresholds per label:\")\n",
" for i, label in enumerate(CONFIG.LABELS):\n",
" print(f\" {label}: {best_thresholds[i]:.3f}\")\n",
" except Exception as e:\n",
" print(f\" Could not load thresholds: {e}\")\n",
" print(\" Using default thresholds of 0.5\")\n",
" best_thresholds = np.array([0.5] * len(CONFIG.LABELS))\n",
" \n",
" print(\"\\nModel setup complete\")\n",
" \n",
"except Exception as e:\n",
" print(f\"\\nError loading model: {e}\")\n",
" print(\"\\nPlease ensure:\")\n",
" print(\"1. You've updated CONFIG.HF_REPO_ID with your actual repository ID\")\n",
" print(\"2. The model was successfully uploaded in the training notebook\")\n",
" print(\"3. The repository is public or you're logged in to HuggingFace\")\n",
" raise\n",
"\n",
"def ensure_text_column(df: pd.DataFrame) -> pd.DataFrame:\n",
" if \"text\" in df.columns:\n",
" return df\n",
" for c in [\"comment_text\", \"sentence\", \"content\", \"review\"]:\n",
" if c in df.columns:\n",
" return df.rename(columns={c: \"text\"})\n",
" raise ValueError(\"No text column found. Add/rename your text column to 'text'.\")\n",
"\n",
"class EmotionDS(torch.utils.data.Dataset):\n",
" def __init__(self, texts, tokenizer, max_len):\n",
" self.texts = texts\n",
" self.tok = tokenizer\n",
" self.max_len = max_len\n",
"\n",
" def __len__(self):\n",
" return len(self.texts)\n",
"\n",
" def __getitem__(self, i):\n",
" enc = self.tok(\n",
" self.texts[i],\n",
" truncation=True,\n",
" padding=\"max_length\",\n",
" max_length=self.max_len,\n",
" return_tensors=\"pt\",\n",
" )\n",
" return {k: v.squeeze(0) for k, v in enc.items()}\n",
"\n",
"print(f\"Loading test data from: {CONFIG.TEST_CSV}\")\n",
"\n",
"if not os.path.exists(CONFIG.TEST_CSV):\n",
" print(\"Test CSV not found. Please check the path.\")\n",
" print(\"\\nIf you're running locally, make sure you have the test data.\")\n",
" print(\"On Kaggle, ensure you've added the competition data as input.\")\n",
" raise FileNotFoundError(CONFIG.TEST_CSV)\n",
"\n",
"df_test = pd.read_csv(CONFIG.TEST_CSV)\n",
"df_test = ensure_text_column(df_test)\n",
"\n",
"print(f\"Test data loaded: {len(df_test)} samples\")\n",
"print(f\"\\nColumns: {df_test.columns.tolist()}\")\n",
"print(f\"\\nFirst few rows:\")\n",
"print(df_test.head())\n",
"\n",
"print(\"\\nGenerating predictions...\\n\")\n",
"\n",
"test_texts = df_test[\"text\"].tolist()\n",
"test_dataset = EmotionDS(test_texts, tokenizer, CONFIG.MAX_LEN)\n",
"test_loader = torch.utils.data.DataLoader(\n",
" test_dataset, \n",
" batch_size=CONFIG.BATCH_SIZE, \n",
" shuffle=False, \n",
" num_workers=2,\n",
" pin_memory=True\n",
")\n",
"\n",
"all_preds = []\n",
"\n",
"with torch.no_grad():\n",
" for batch_idx, batch in enumerate(test_loader):\n",
" batch = {k: v.to(device, non_blocking=True) for k, v in batch.items()}\n",
" \n",
" with autocast(enabled=True):\n",
" outputs = model(\n",
" input_ids=batch[\"input_ids\"], \n",
" attention_mask=batch[\"attention_mask\"]\n",
" )\n",
" \n",
" probs = torch.sigmoid(outputs.logits).float().cpu().numpy()\n",
" all_preds.append(probs)\n",
" \n",
" if (batch_idx + 1) % 10 == 0:\n",
" progress = (batch_idx + 1) * CONFIG.BATCH_SIZE\n",
" print(f\" Processed {min(progress, len(df_test))}/{len(df_test)} samples...\")\n",
"\n",
"all_probs = np.vstack(all_preds)\n",
"\n",
"print(f\"\\nPredictions generated for {len(all_probs)} samples\")\n",
"print(f\"Shape: {all_probs.shape}\")\n",
"\n",
"print(\"\\nApplying optimized thresholds...\\n\")\n",
"\n",
"final_predictions = (all_probs >= best_thresholds).astype(int)\n",
"\n",
"print(f\"Thresholds applied\")\n",
"print(f\"\\nPrediction distribution:\")\n",
"for i, label in enumerate(CONFIG.LABELS):\n",
" count = final_predictions[:, i].sum()\n",
" percentage = (count / len(final_predictions)) * 100\n",
" print(f\" {label:<12} {count:>6} samples ({percentage:>5.1f}%)\")\n",
"\n",
"avg_labels_per_sample = final_predictions.sum(axis=1).mean()\n",
"print(f\"\\n Average labels per sample: {avg_labels_per_sample:.2f}\")\n",
"\n",
"print(\"\\nCreating submission file...\\n\")\n",
"\n",
"submission = pd.DataFrame()\n",
"\n",
"if \"id\" in df_test.columns:\n",
" submission[\"id\"] = df_test[\"id\"]\n",
"else:\n",
" submission[\"id\"] = np.arange(len(df_test))\n",
"\n",
"for i, label in enumerate(CONFIG.LABELS):\n",
" submission[label] = final_predictions[:, i]\n",
"\n",
"submission.to_csv(CONFIG.SUBMISSION_PATH, index=False)\n",
"\n",
"print(f\"Submission file saved to: {CONFIG.SUBMISSION_PATH}\")\n",
"print(f\"\\nSubmission preview:\")\n",
"print(submission.head(10))\n",
"print(f\"\\nTotal rows: {len(submission)}\")\n",
"print(f\"Columns: {submission.columns.tolist()}\")\n",
"\n",
"print(\"Verifying submission format...\\n\")\n",
"\n",
"required_columns = [\"id\"] + CONFIG.LABELS\n",
"submission_columns = submission.columns.tolist()\n",
"\n",
"if submission_columns == required_columns:\n",
" print(\"Submission format is correct\")\n",
" print(f\" Columns: {submission_columns}\")\n",
" \n",
" if submission[CONFIG.LABELS].isin([0, 1]).all().all():\n",
" print(\"All predictions are binary (0 or 1)\")\n",
" else:\n",
" print(\"Warning: Some predictions are not binary\")\n",
" \n",
" if not submission.isnull().any().any():\n",
" print(\"No missing values\")\n",
" else:\n",
" print(\"Missing values detected\")\n",
" print(submission.isnull().sum())\n",
"else:\n",
" print(\"Submission format is incorrect\")\n",
" print(f\" Expected: {required_columns}\")\n",
" print(f\" Got: {submission_columns}\")\n",
"\n",
"print(\"\\nSubmitting to Kaggle...\\n\")\n",
"\n",
"submission_message = f\"DeBERTa-v3 with optimized thresholds - HF: {CONFIG.HF_REPO_ID}\"\n",
"\n",
"try:\n",
" import kaggle\n",
" \n",
" kaggle.api.competition_submit(\n",
" file_name=CONFIG.SUBMISSION_PATH,\n",
" message=submission_message,\n",
" competition=CONFIG.COMPETITION_NAME\n",
" )\n",
" \n",
" print(\"Submission successful\")\n",
" print(f\"\\nSubmission message: {submission_message}\")\n",
" print(f\"\\nView your submission at:\")\n",
" print(f\" https://www.kaggle.com/c/{CONFIG.COMPETITION_NAME}/submissions\")\n",
" \n",
"except Exception as e:\n",
" print(f\"Submission failed: {e}\")\n",
" print(\"\\nPossible reasons:\")\n",
" print(\"1. Kaggle API credentials not configured\")\n",
" print(\"2. Competition name is incorrect\")\n",
" print(\"3. You've reached the daily submission limit\")\n",
" print(\"4. The competition has ended\")\n",
" print(\"\\nYou can manually upload the submission.csv file to Kaggle.\")\n",
"\n",
"print(\"\\n\" + \"=\"*60)\n",
"print(\"PREDICTION STATISTICS\")\n",
"print(\"=\"*60)\n",
"\n",
"labels_per_sample = final_predictions.sum(axis=1)\n",
"print(\"\\nLabels per sample distribution:\")\n",
"for i in range(6):\n",
" count = (labels_per_sample == i).sum()\n",
" percentage = (count / len(labels_per_sample)) * 100\n",
" print(f\" {i} labels: {count:>6} samples ({percentage:>5.1f}%)\")\n",
"\n",
"print(\"\\nMost common label combinations:\")\n",
"label_combinations = []\n",
"for pred in final_predictions:\n",
" active_labels = [CONFIG.LABELS[i] for i, val in enumerate(pred) if val == 1]\n",
" if active_labels:\n",
" label_combinations.append(\", \".join(sorted(active_labels)))\n",
" else:\n",
" label_combinations.append(\"(none)\")\n",
"\n",
"from collections import Counter\n",
"combo_counts = Counter(label_combinations)\n",
"for combo, count in combo_counts.most_common(10):\n",
" percentage = (count / len(label_combinations)) * 100\n",
" print(f\" {combo:<30} {count:>6} ({percentage:>5.1f}%)\")\n",
"\n",
"print(\"\\nAverage probability per label:\")\n",
"for i, label in enumerate(CONFIG.LABELS):\n",
" avg_prob = all_probs[:, i].mean()\n",
" std_prob = all_probs[:, i].std()\n",
" print(f\" {label:<12} {avg_prob:.4f} +/- {std_prob:.4f}\")\n",
"\n",
"print(\"\\n\" + \"=\"*60)\n",
"print(\"SUBMISSION COMPLETE\")\n",
"print(\"=\"*60)\n",
"print(f\"\\nSubmission file: {CONFIG.SUBMISSION_PATH}\")\n",
"print(f\"Model used: {CONFIG.HF_REPO_ID}\")\n",
"print(f\"Optimized thresholds: {best_thresholds}\")\n",
"print(\"\\nCheck Kaggle leaderboard for your score\")"
]
}
],
"metadata": {
"language_info": {
"name": "python"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|