experiments: add sweep code of reduction on synthetic dataset

experiments/sweeper/sweeper.py

@@ -0,0 +1,40 @@
+import time
+from itertools import product
+
+import pandas as pd
+from joblib import Parallel, delayed
+
+from evaluation.reduction import scores_of_synthetic
+from reduction.runner import REDUCTION_METHODS, reduce_by_method
+
+
+def _run_reduction(method: str, data_param: dict, data: dict) -> dict:
+    normal_data = data["normal_data"]
+    abnormal_data = data["abnormal_data"]
+    ground_truth = data["ground_truth"]
+    time_series = pd.concat([normal_data, abnormal_data], axis=0, ignore_index=True)
+
+    time_start: float = time.perf_counter()
+
+    remained_metrics = reduce_by_method(method, time_series, ground_truth)
+
+    time_reduction: float = time.perf_counter() - time_start
+
+    scores = scores_of_synthetic(
+        pred_anomalous_metrics=remained_metrics,
+        true_root_fault_metrics=set(ground_truth["root_fault_nodes"]),
+        true_fault_propagated_metrics=set(ground_truth["fault_propagated_nodes"]),
+        total_metrics=set(time_series.columns.tolist()),
+    )
+    scores["time_reduction"] = time_reduction
+
+    return data_param | {"reduction_method": method} | scores
+
+
+def sweep_reduction_on_synthetic(dataset: list[tuple[dict, dict]], n_jobs: int = -1) -> pd.DataFrame:
+    ret = Parallel(n_jobs=n_jobs)(
+        delayed(_run_reduction)(method, data_param, data)
+        for method, (data_param, data) in product(REDUCTION_METHODS, dataset)
+    )
+    assert ret is not None
+    return pd.DataFrame(ret)