weightslab 1.3.1__tar.gz → 1.3.2__tar.gz

weightslab-1.3.2/CHANGELOG.md

@@ -0,0 +1 @@
+# Changelog - 2026-07-07 v1.3.2

{weightslab-1.3.1/weightslab.egg-info → weightslab-1.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: weightslab
-Version: 1.3.1
+Version: 1.3.2
 Summary: Paving the way between black-box and white-box modeling.
 Author-email: Alexandru-Andrei Rotaru <alexandru@graybx.com>, Guillaue Pelluet <guillaue@graybx.com>
 License: BSD 2-clause

{weightslab-1.3.1 → weightslab-1.3.2}/tests/gRPC/test_grpc_histogram_categorical.py

@@ -172,6 +172,66 @@ class TestGetHistogramMixedNaN(unittest.TestCase):
         self.assertGreater(len(resp.bins), 0)
 
 
+class TestGetHistogramObjectDtypeNumeric(unittest.TestCase):
+    """A numeric column stored as object dtype (floats mixed with None/NaN)
+    must stay numeric — None/NaN are missing data, not an 'unset' category."""
+
+    def test_object_dtype_floats_with_none_stays_numeric(self):
+        # Mixing Python floats with None yields an object-dtype column; this is
+        # how a per-sample loss column looks before every sample has a value.
+        df = pd.DataFrame({
+            "loss": pd.Series(
+                [0.1, None, 0.3, np.nan, 0.5, None, 0.7, 0.2, None, 0.9],
+                dtype=object,
+            ),
+            "origin": ["train"] * 5 + ["eval"] * 5,
+            "discarded": [False] * 10,
+        })
+        self.assertEqual(df["loss"].dtype, object)
+        svc = _make_service_with_df(df)
+        resp = svc.GetHistogram(_req("loss"), MockContext())
+        self.assertTrue(resp.success, resp.message)
+        self.assertFalse(resp.is_categorical)
+        self.assertGreater(len(resp.bins), 0)
+        self.assertEqual(len(resp.categorical_bars), 0)
+
+    def test_none_nan_excluded_from_numeric_counts(self):
+        # Only the 6 finite values should be counted across all bins; the 4
+        # None/NaN entries must not appear as an "unset" category or a value.
+        df = pd.DataFrame({
+            "loss": pd.Series([1.0, None, 2.0, np.nan, 3.0, None, 4.0, 5.0, None, 6.0],
+                              dtype=object),
+            "origin": ["train"] * 10,
+            "discarded": [False] * 10,
+        })
+        svc = _make_service_with_df(df)
+        resp = svc.GetHistogram(_req("loss"), MockContext())
+        self.assertFalse(resp.is_categorical)
+        total_counted = sum(b.count for b in resp.bins)
+        self.assertEqual(total_counted, 6)
+
+
+class TestGetHistogramAllNaNFloatColumn(unittest.TestCase):
+    """A genuine float dtype column that is entirely NaN (e.g. a loss column
+    no sample has filled yet) must render as an empty numeric histogram — not
+    a categorical 'unset' bar."""
+
+    def test_all_nan_float_column_stays_numeric(self):
+        df = pd.DataFrame({
+            "loss": pd.Series([np.nan] * 8, dtype=float),
+            "origin": ["train"] * 8,
+            "discarded": [False] * 8,
+        })
+        self.assertTrue(pd.api.types.is_float_dtype(df["loss"]))
+        svc = _make_service_with_df(df)
+        resp = svc.GetHistogram(_req("loss"), MockContext())
+        self.assertTrue(resp.success, resp.message)
+        self.assertFalse(resp.is_categorical)
+        self.assertEqual(len(resp.categorical_bars), 0)
+        # No finite values → every bin is empty, none surfaced as "unset".
+        self.assertEqual(sum(b.count for b in resp.bins), 0)
+
+
 class TestGetHistogramEdgeCases(unittest.TestCase):
     def test_column_not_in_df(self):
         df = pd.DataFrame({"a": [1, 2, 3], "origin": ["train"] * 3,

{weightslab-1.3.1 → weightslab-1.3.2}/weightslab/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.3.1'
-__version_tuple__ = version_tuple = (1, 3, 1)
+__version__ = version = '1.3.2'
+__version_tuple__ = version_tuple = (1, 3, 2)
 
-__commit_id__ = commit_id = 'g3898cb715'
+__commit_id__ = commit_id = 'g8eab0cc6d'

{weightslab-1.3.1 → weightslab-1.3.2}/weightslab/trainer/services/data_service.py

@@ -4339,18 +4339,29 @@ class DataService:
                     else np.zeros(n, bool))
 
             # Detect whether column is categorical (string/object) or numeric.
+            # A column is numeric if ANY value coerces to a finite number — even
+            # when the pandas dtype is ``object`` (e.g. a loss column holding
+            # floats mixed with None/NaN for samples that have no value yet).
+            # None/NaN are MISSING data, not a category, so they must never turn
+            # a numeric column into a categorical one (which would surface them
+            # as a spurious "unset" bar). We therefore treat as categorical only
+            # a genuine pandas ``category`` dtype, or a column whose values do
+            # not coerce to any numeric value at all (pure strings).
             col_series = df[column]
             numeric_vals = pd.to_numeric(col_series, errors="coerce")
-            is_categorical = numeric_vals.isna().all() or (
-                col_series.dtype == object or
-                str(col_series.dtype) == "category" or
-                hasattr(col_series, "cat")
+            is_category_dtype = (
+                str(col_series.dtype) == "category" or hasattr(col_series, "cat")
+            )
+            # A genuine numeric dtype (float/int/bool) is ALWAYS numeric — even
+            # when every value is NaN (e.g. a loss column no sample has filled
+            # yet). Such a column yields an empty numeric histogram, never a
+            # spurious "unset" categorical bar.
+            is_numeric_dtype = (
+                pd.api.types.is_numeric_dtype(col_series) and not is_category_dtype
+            )
+            is_categorical = (not is_numeric_dtype) and (
+                is_category_dtype or not bool(numeric_vals.notna().any())
             )
-            # A column that has a few non-NaN numeric values mixed with mostly
-            # NaN still counts as numeric; only treat as categorical when
-            # pd.to_numeric fails on all values.
-            if not is_categorical and numeric_vals.notna().sum() == 0:
-                is_categorical = True
 
             if is_categorical:
                 # --- Categorical path ---

{weightslab-1.3.1 → weightslab-1.3.2/weightslab.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: weightslab
-Version: 1.3.1
+Version: 1.3.2
 Summary: Paving the way between black-box and white-box modeling.
 Author-email: Alexandru-Andrei Rotaru <alexandru@graybx.com>, Guillaue Pelluet <guillaue@graybx.com>
 License: BSD 2-clause

{weightslab-1.3.1 → weightslab-1.3.2}/weightslab.egg-info/scm_version.json

@@ -1,7 +1,7 @@
 {
-  "tag": "1.3.1",
+  "tag": "1.3.2",
   "distance": 0,
-  "node": "g3898cb71521e5b2e15b709a84c5ed095706ddd95",
+  "node": "g8eab0cc6dde4dbee84302aad461e8acdc26d3b9e",
   "dirty": false,
   "branch": "HEAD",
   "node_date": "2026-07-07"

weightslab-1.3.1/CHANGELOG.md

@@ -1 +0,0 @@
-# Changelog - 2026-07-07 v1.3.1