Update fuzzer.py to be less stringent when validating approx. metrics

test/scripts/fuzzer.py

@@ -400,11 +400,20 @@ def compute_stats(df: pd.DataFrame, keep_nans: bool, keep_infs: bool) -> Dict:
 
 
 def compare_metric(
-    worker_id: int, q1: str, q2: str, metric: str, expected, found, keep_nans: bool, keep_infs: bool
+    worker_id: int,
+    q1: str,
+    q2: str,
+    metric: str,
+    expected,
+    found,
+    keep_nans: bool,
+    keep_infs: bool,
+    rtol: float = 1.0e-5,
+    atol: float = 1.0e-8,
 ) -> bool:
     do_numeric_comparison = expected is not None and found is not None
     if do_numeric_comparison:
-        if not np.isclose(expected, found, equal_nan=True):
+        if not np.isclose(expected, found, rtol=rtol, atol=atol, equal_nan=True):
             logging.warning(
                 "[%d] %s, %s (%s; keep_nans=%s; keep_infs=%s): FAIL! Expected %.16g, found %.16g",
                 worker_id,
@@ -437,8 +446,8 @@ def compare_metric(
 
 
 def compare_query_stats(worker_id: int, q1: str, q2: str, expected, sel: hictkpy.PixelSelector) -> int:
-    exact_metrics = ["nnz", "sum", "min", "max", "min", "skewness", "kurtosis"]
-    approx_metrics = ["variance"]
+    exact_metrics = ["nnz", "sum", "min", "max", "min"]
+    approx_metrics = ["variance", "skewness", "kurtosis"]
     num_failures = 0
     for keep_nans in [True, False]:
         for keep_infs in [True, False]:
@@ -454,7 +463,7 @@ def compare_query_stats(worker_id: int, q1: str, q2: str, expected, sel: hictkpy
             for metric in approx_metrics:
                 n1 = stats_expected[metric]
                 n2 = stats_found[metric]
-                if not compare_metric(worker_id, q1, q2, metric, n1, n2, keep_nans, keep_infs):
+                if not compare_metric(worker_id, q1, q2, metric, n1, n2, keep_nans, keep_infs, 1.0e-4, 1.0e-6):
                     num_failures += 1
 
     return int(num_failures != 0)