Spaces:
Running
Running
File size: 10,204 Bytes
7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 7a920b1 5515ef5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 |
import pandas as pd
from typing import Optional
from sklearn.metrics import (
precision_score,
recall_score,
accuracy_score,
classification_report,
)
from tqdm.asyncio import tqdm as async_tqdm
import asyncio
from src.modules.vlm_inference import analyze_product_image_async
from src.modules.data_processing import image_to_base64
from pathlib import Path
DATA_PATH = Path(__file__).parents[2] / "data"
async def _process_single_row(
row_data: dict,
model: str,
api_key: str,
provider: str,
semaphore: asyncio.Semaphore,
) -> dict:
"""
Process a single row with semaphore control
Args:
row_data: Dictionary with 'image' and 'id' keys
model: Model to use for inference
api_key: API key for the provider
provider: Provider to use
semaphore: Asyncio semaphore for rate limiting
Returns:
dict: Prediction result
"""
async with semaphore:
try:
img_b64 = image_to_base64(row_data["image"])
prediction = await analyze_product_image_async(
image_url=img_b64,
model=model,
api_key=api_key,
provider=provider,
)
return {
"id": row_data["id"],
"pred_masterCategory": prediction.master_category,
"pred_gender": prediction.gender,
"pred_subCategory": prediction.sub_category,
"pred_description": prediction.description,
}
except Exception as e:
return {
"id": row_data["id"],
"pred_masterCategory": None,
"pred_gender": None,
"pred_subCategory": None,
"pred_description": f"Error: {str(e)}",
}
async def run_inference_on_dataframe_async(
df: pd.DataFrame,
model: str = "accounts/fireworks/models/qwen2p5-vl-72b-instruct",
api_key: Optional[str] = None,
provider: str = "Fireworks",
max_concurrent_requests: int = 10,
) -> pd.DataFrame:
"""
Run VLM inference on entire dataframe of images with concurrent requests
Args:
df: DataFrame containing images
model: Model to use for inference
api_key: API key for the provider
provider: Provider to use (Fireworks or OpenAI)
max_concurrent_requests: Maximum number of concurrent API requests (default: 10)
Returns:
pd.DataFrame: Results with columns:
- id: Image ID
- pred_masterCategory: Predicted master category
- pred_gender: Predicted gender
- pred_subCategory: Predicted sub-category
- pred_description: Predicted description
"""
# Create semaphore for rate limiting
semaphore = asyncio.Semaphore(max_concurrent_requests)
# Prepare all rows as dictionaries
rows_data = [
{"image": row.image, "id": row.id}
for row in df.itertuples(index=False, name="columns")
]
# Create all tasks (coroutines, not awaited yet)
tasks = [
_process_single_row(row_data, model, api_key, provider, semaphore)
for row_data in rows_data
]
_model = model.split("/")[-1]
# Run all tasks concurrently with progress bar
results = []
for task in async_tqdm.as_completed(tasks, total=len(tasks)):
result = await task
results.append(result)
if len(results) % 10 == 0:
df_pred = pd.DataFrame(results)
file_name = DATA_PATH / f"df_pred_{provider}_{_model}.csv"
df_pred.to_csv(file_name, index=False)
# Final save
df_pred = pd.DataFrame(results)
file_name = DATA_PATH / f"df_pred_{provider}_{_model}.csv"
df_pred.to_csv(file_name, index=False)
print(f"\nPrediction successful, dataset saved to {file_name}")
return df_pred
def run_inference_on_dataframe(
df: pd.DataFrame,
model: str = "accounts/fireworks/models/qwen2p5-vl-72b-instruct",
api_key: Optional[str] = None,
provider: str = "Fireworks",
max_concurrent_requests: int = 10,
) -> pd.DataFrame:
"""
Run VLM inference on entire dataframe of images (sync wrapper for async function)
Args:
df: DataFrame containing images
model: Model to use for inference
api_key: API key for the provider
provider: Provider to use (Fireworks or OpenAI)
max_concurrent_requests: Maximum number of concurrent API requests (default: 10)
Returns:
pd.DataFrame: Results with columns:
- id: Image ID
- pred_masterCategory: Predicted master category
- pred_gender: Predicted gender
- pred_subCategory: Predicted sub-category
- pred_description: Predicted description
"""
return asyncio.run(
run_inference_on_dataframe_async(df, model, api_key, provider, max_concurrent_requests)
)
def calculate_metrics(
df_ground_truth: pd.DataFrame,
df_predictions: pd.DataFrame,
column: str,
id_col: str = "id",
average: str = "weighted",
) -> dict:
"""
Calculate precision, recall, and accuracy for a specific column
Args:
df_ground_truth: DataFrame with ground truth labels
df_predictions: DataFrame with predictions
column: Column name to evaluate (e.g., 'master_category', 'gender', 'sub_category')
id_col: Column name for joining the dataframes
average: Averaging method for multi-class metrics ('weighted', 'macro', 'micro')
Returns:
dict: Dictionary containing:
- accuracy: Overall accuracy
- precision: Precision score
- recall: Recall score
- classification_report: Detailed classification report
"""
# Merge ground truth and predictions on ID
merged = df_ground_truth[[id_col, column]].merge(
df_predictions[[id_col, f"pred_{column}"]], on=id_col, how="inner"
)
# Remove any rows with None predictions (failed inferences)
merged = merged.dropna(subset=[f"pred_{column}"])
if len(merged) == 0:
raise ValueError("No valid predictions found after merging and removing nulls")
# Get ground truth and predictions
y_true = merged[column]
y_pred = merged[f"pred_{column}"]
# Calculate metrics
accuracy = accuracy_score(y_true, y_pred)
precision = precision_score(y_true, y_pred, average=average, zero_division=0)
recall = recall_score(y_true, y_pred, average=average, zero_division=0)
# Get detailed classification report
report = classification_report(y_true, y_pred, zero_division=0)
return {
"accuracy": accuracy,
"precision": precision,
"recall": recall,
"classification_report": report,
"num_samples": len(merged),
}
def evaluate_all_categories(
df_ground_truth: pd.DataFrame,
df_predictions: pd.DataFrame,
id_col: str = "id",
categories: list[str] = None,
) -> dict:
"""
Evaluate predictions for all category types
Args:
df_ground_truth: DataFrame with ground truth labels
df_predictions: DataFrame with predictions
id_col: Column name for joining
categories: List of category columns to evaluate
Returns:
dict: Dictionary with metrics for each category
"""
if categories is None:
categories = ["masterCategory", "gender", "subCategory"]
results = {}
for category in categories:
print(f"\n{'='*60}")
print(f"Evaluating: {category}")
print(f"{'='*60}")
try:
metrics = calculate_metrics(
df_ground_truth=df_ground_truth,
df_predictions=df_predictions,
column=category,
id_col=id_col,
)
results[category] = metrics
# Print summary
print(f"Accuracy: {metrics['accuracy']:.4f}")
print(f"Precision: {metrics['precision']:.4f}")
print(f"Recall: {metrics['recall']:.4f}")
print(f"Samples: {metrics['num_samples']}")
print(f"\nClassification Report:\n{metrics['classification_report']}")
except Exception as e:
print(f"Error evaluating {category}: {e}")
results[category] = {"error": str(e)}
return results
def create_evaluation_summary(results: dict) -> pd.DataFrame:
"""
Create a summary DataFrame from evaluation results
Args:
results: Dictionary of evaluation results from evaluate_all_categories
Returns:
pd.DataFrame: Summary table with metrics for each category
"""
summary_data = []
for category, metrics in results.items():
if "error" not in metrics:
summary_data.append(
{
"Category": category,
"Accuracy": f"{metrics['accuracy']:.4f}",
"Precision": f"{metrics['precision']:.4f}",
"Recall": f"{metrics['recall']:.4f}",
"Samples": metrics["num_samples"],
}
)
else:
summary_data.append(
{
"Category": category,
"Accuracy": "Error",
"Precision": "Error",
"Recall": "Error",
"Samples": 0,
}
)
return pd.DataFrame(summary_data)
def extract_metrics(results_dict, model_name):
"""
Extract accuracy, precision, and recall for each category.
Args:
results_dict: Dictionary containing evaluation metrics
model_name: Name of the model for identification
Returns:
List of dictionaries with metrics per category
"""
metrics_list = []
for category, metrics in results_dict.items():
metrics_list.append({
'model': model_name,
'category': category,
'accuracy': metrics['accuracy'],
'precision': metrics['precision'],
'recall': metrics['recall'],
'num_samples': metrics['num_samples']
})
return metrics_list |