Enable summarization by subsets and groups

examples/evaluate_bluebench.sh

@@ -28,4 +28,4 @@ unitxt-evaluate \
     --batch_size 8 \
     --verbosity ERROR
 
-unitxt-summarize ./results/bluebench
+unitxt-summarize --folder ./results/bluebench

src/unitxt/evaluate_cli.py

@@ -7,7 +7,7 @@
 import platform
 import subprocess
 import sys
-from datetime import datetime
+from datetime import datetime, timezone
 from functools import partial
 from typing import Any, Dict, List, Optional, Tuple, Union
 
@@ -691,9 +691,8 @@ def _save_results_to_disk(
         "results": global_scores,
     }
 
-    # prepend to the results_path name the time in a wat like this: 2025-04-04T11:37:32
-
-    timestamp = datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
+    # prepend the timestamp in UTC (e.g., 2025-01-18T11-37-32) to the file names
+    timestamp = datetime.now().astimezone(timezone.utc).strftime("%Y-%m-%dT%H-%M-%S")
 
     results_path = prepend_timestamp_to_path(results_path, timestamp)
     samples_path = prepend_timestamp_to_path(samples_path, timestamp)
@@ -836,48 +835,129 @@ def main():
     logger.info("Unitxt Evaluation CLI finished successfully.")
 
 
-def extract_scores(directory):  # pragma: no cover
+def extract_scores(folder: str, subset: str, group: str):  # pragma: no cover
     import pandas as pd
 
-    data = []
+    def safe_score(d: dict, key="score"):
+        na = "N/A"
+        return d.get(key, na) if isinstance(d, dict) else na
 
-    for filename in sorted(os.listdir(directory)):
-        if filename.endswith("evaluation_results.json"):
-            file_path = os.path.join(directory, filename)
-            try:
-                with open(file_path, encoding="utf-8") as f:
-                    content = json.load(f)
+    def extract_subset(results: dict, subset: str, group: str):
+        subset_results = results.get(subset, {})
+        row = {subset: safe_score(subset_results)}
+
+        groups = subset_results.get("groups", {})
+
+        if not groups:
+            return row
+
+        group_results = groups.get(group) if group else next(iter(groups.values()), {})
 
-                    env_info = content.get("environment_info", {})
-                    timestamp = env_info.get("timestamp_utc", "N/A")
-                    model = env_info.get("parsed_arguments", {}).get("model", "N/A")
-                    results = content.get("results", {})
+        if not isinstance(group_results, dict):
+            return row
 
-                    row = {}
-                    row["Model"] = model
-                    row["Timestamp"] = timestamp
-                    row["Average"] = results.get("score", "N/A")
+        row.update(
+            {k: safe_score(v) for k, v in group_results.items() if isinstance(v, dict)}
+        )
+        return row
+
+    def extract_all(results: dict):
+        row = {"Average": safe_score(results)}
+        row.update(
+            {k: safe_score(v) for k, v in results.items() if isinstance(v, dict)}
+        )
+        return row
+
+    data = []
 
-                    for key in results.keys():
-                        if isinstance(results[key], dict):
-                            score = results[key].get("score", "N/A")
-                            row[key] = score
+    for filename in sorted(os.listdir(folder)):
+        if not filename.endswith("evaluation_results.json"):
+            continue
 
-                    data.append(row)
-            except Exception as e:
-                logger.error(f"Error parsing results file {filename}: {e}.")
+        file_path = os.path.join(folder, filename)
+        try:
+            with open(file_path, encoding="utf-8") as f:
+                content = json.load(f)
+
+                env_info = content.get("environment_info", {})
+                row = {
+                    "Model": safe_score(env_info.get("parsed_arguments", {}), "model"),
+                    "Timestamp": safe_score(env_info, "timestamp_utc"),
+                }
+
+                results = content.get("results", {})
+
+                extra = (
+                    extract_subset(results, subset, group)
+                    if subset
+                    else extract_all(results)
+                )
+                row.update(extra)
+                data.append(row)
+        except Exception as e:
+            logger.error(f"Error parsing results file {filename}: {e}.")
 
     return pd.DataFrame(data).sort_values(by="Timestamp", ascending=True)
 
 
+def setup_summarization_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.RawTextHelpFormatter,
+        description="CLI utility for summarizing evaluation results.",
+    )
+
+    parser.add_argument(
+        "--folder",
+        "-f",
+        dest="folder",
+        type=str,
+        default=".",
+        help="Directory containing evaluation results json files. Default: current folder.\n",
+    )
+
+    parser.add_argument(
+        "--subset",
+        "-s",
+        type=str,
+        dest="subset",
+        default=None,
+        help="Subset to filter results by. Default: none.",
+    )
+
+    parser.add_argument(
+        "--group",
+        "-g",
+        type=str,
+        dest="group",
+        default=None,
+        help="Group to filter results to. Requires specifying a subset. Default: first group.",
+    )
+
+    parser.add_argument(
+        "--output",
+        "-o",
+        type=str,
+        choices=["markdown", "csv"],
+        dest="output",
+        default="markdown",
+        help="Output format. Can be markdown or csv. Default: markdown",
+    )
+
+    return parser
+
+
 def summarize_cli():
-    if len(sys.argv) != 2:
-        logger.error("Usage: python summarize_cli_results.py <results-directory>")
-        sys.exit(1)
-    directory = sys.argv[1]
-    df = extract_scores(directory)
+    parser = setup_summarization_parser()
+    args = parser.parse_args()
 
-    logger.info(df.to_markdown(index=False))
+    df = extract_scores(args.folder, args.subset, args.group)
+
+    if args.output == "markdown":
+        logger.info(df.to_markdown(index=False))
+    elif args.output == "csv":
+        logger.info(df.to_csv(index=False))
+    else:
+        logger.error(f"Unsupported output format: {args.output}")
 
 
 if __name__ == "__main__":

tests/library/test_cli.py

@@ -751,8 +751,12 @@ def test_save_results_to_disk_summary_only(
         # --- Arrange ---
         # (Arrange section remains the same as previous version)
         mock_timestamp = "2025-04-14T10:00:00"
+        mock_timestamp_utc = "2025-04-14T08:00:00"
         mock_now = MagicMock()
         mock_now.strftime.return_value = mock_timestamp
+        mock_astimezone = MagicMock()
+        mock_astimezone.strftime.return_value = mock_timestamp_utc
+        mock_now.astimezone.return_value = mock_astimezone
         mock_datetime.now.return_value = mock_now
         mock_utcnow = MagicMock()
         mock_utcnow.isoformat.return_value = "2025-04-14T08:00:00"
@@ -784,7 +788,9 @@ def test_save_results_to_disk_summary_only(
         }
         base_results_path = "/out/results_prefix.json"
         base_samples_path = "/out/results_prefix_samples.json"
-        expected_timestamped_results_path = f"/out/{mock_timestamp}_results_prefix.json"
+        expected_timestamped_results_path = (
+            f"/out/{mock_timestamp_utc}_results_prefix.json"
+        )
 
         # --- Act ---
         cli._save_results_to_disk(
@@ -844,7 +850,7 @@ def test_save_results_to_disk_summary_only(
         )
         log_calls = [call[0][0] for call in mock_logger.info.call_args_list]
         expected_timestamped_samples_path = (
-            f"/out/{mock_timestamp}_results_prefix_samples.json"
+            f"/out/{mock_timestamp_utc}_results_prefix_samples.json"
         )
         self.assertNotIn(
             f"Saving detailed samples to: {expected_timestamped_samples_path}",