Calculate aggregate statistics for all metrics, split by category.
(all_metrics: List[Dict[str, float]], all_categories: List[int])
| 165 | return metrics |
| 166 | |
| 167 | def aggregate_metrics(all_metrics: List[Dict[str, float]], all_categories: List[int]) -> Dict[str, Dict[str, Union[float, Dict[str, float]]]]: |
| 168 | """Calculate aggregate statistics for all metrics, split by category.""" |
| 169 | if not all_metrics: |
| 170 | return {} |
| 171 | |
| 172 | # Initialize aggregates for overall and per-category metrics |
| 173 | aggregates = defaultdict(list) |
| 174 | category_aggregates = defaultdict(lambda: defaultdict(list)) |
| 175 | |
| 176 | # Collect all values for each metric, both overall and per category |
| 177 | for metrics, category in zip(all_metrics, all_categories): |
| 178 | for metric_name, value in metrics.items(): |
| 179 | aggregates[metric_name].append(value) |
| 180 | category_aggregates[category][metric_name].append(value) |
| 181 | |
| 182 | # Calculate statistics for overall metrics |
| 183 | results = { |
| 184 | "overall": {} |
| 185 | } |
| 186 | |
| 187 | for metric_name, values in aggregates.items(): |
| 188 | results["overall"][metric_name] = { |
| 189 | 'mean': statistics.mean(values), |
| 190 | 'std': statistics.stdev(values) if len(values) > 1 else 0.0, |
| 191 | 'median': statistics.median(values), |
| 192 | 'min': min(values), |
| 193 | 'max': max(values), |
| 194 | 'count': len(values) |
| 195 | } |
| 196 | |
| 197 | # Calculate statistics for each category |
| 198 | for category in sorted(category_aggregates.keys()): |
| 199 | results[f"category_{category}"] = {} |
| 200 | for metric_name, values in category_aggregates[category].items(): |
| 201 | if values: # Only calculate if we have values for this category |
| 202 | results[f"category_{category}"][metric_name] = { |
| 203 | 'mean': statistics.mean(values), |
| 204 | 'std': statistics.stdev(values) if len(values) > 1 else 0.0, |
| 205 | 'median': statistics.median(values), |
| 206 | 'min': min(values), |
| 207 | 'max': max(values), |
| 208 | 'count': len(values) |
| 209 | } |
| 210 | |
| 211 | return results |
no test coverage detected