fix: cast numbers to float64 in descriptive statistics to avoid integer summation overflow errors

src/awkward/operations/ak_corr.py

@@ -107,8 +107,33 @@ def _impl(x, y, weight, axis, keepdims, mask_identity, highlevel, behavior, attr
         )
 
     x = ctx.wrap(x_layout)
+    x = ak.operations.ak_values_astype._impl(
+        x,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     y = ctx.wrap(y_layout)
+    y = ak.operations.ak_values_astype._impl(
+        y,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     weight = ctx.wrap(weight_layout, allow_other=True)
+    if weight is not None:
+        weight = ak.operations.ak_values_astype._impl(
+            weight,
+            np.float64,
+            including_unknown=False,
+            highlevel=True,
+            behavior=ctx.behavior,
+            attrs=ctx.attrs,
+        )
 
     with np.errstate(invalid="ignore", divide="ignore"):
         xmean = ak.operations.ak_mean._impl(

src/awkward/operations/ak_covar.py

@@ -103,8 +103,33 @@ def _impl(x, y, weight, axis, keepdims, mask_identity, highlevel, behavior, attr
         )
 
     x = ctx.wrap(x_layout)
+    x = ak.operations.ak_values_astype._impl(
+        x,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     y = ctx.wrap(y_layout)
+    y = ak.operations.ak_values_astype._impl(
+        y,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     weight = ctx.wrap(weight_layout, allow_other=True)
+    if weight is not None:
+        weight = ak.operations.ak_values_astype._impl(
+            weight,
+            np.float64,
+            including_unknown=False,
+            highlevel=True,
+            behavior=ctx.behavior,
+            attrs=ctx.attrs,
+        )
 
     with np.errstate(invalid="ignore", divide="ignore"):
         xmean = ak.operations.ak_mean._impl(

src/awkward/operations/ak_linear_fit.py

@@ -108,8 +108,33 @@ def _impl(x, y, weight, axis, keepdims, mask_identity, highlevel, behavior, attr
         )
 
     x = ctx.wrap(x_layout)
+    x = ak.operations.ak_values_astype._impl(
+        x,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     y = ctx.wrap(y_layout)
+    y = ak.operations.ak_values_astype._impl(
+        y,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     weight = ctx.wrap(weight_layout, allow_other=True)
+    if weight is not None:
+        weight = ak.operations.ak_values_astype._impl(
+            weight,
+            np.float64,
+            including_unknown=False,
+            highlevel=True,
+            behavior=ctx.behavior,
+            attrs=ctx.attrs,
+        )
 
     with np.errstate(invalid="ignore", divide="ignore"):
         if weight is None:

src/awkward/operations/ak_mean.py

@@ -193,7 +193,24 @@ def _impl(x, weight, axis, keepdims, mask_identity, highlevel, behavior, attrs):
         )
 
     x = ctx.wrap(x_layout)
+    x = ak.operations.ak_values_astype._impl(
+        x,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     weight = ctx.wrap(weight_layout, allow_other=True)
+    if weight is not None:
+        weight = ak.operations.ak_values_astype._impl(
+            weight,
+            np.float64,
+            including_unknown=False,
+            highlevel=True,
+            behavior=ctx.behavior,
+            attrs=ctx.attrs,
+        )
 
     # Handle named axis
     named_axis = _get_named_axis(ctx)

src/awkward/operations/ak_moment.py

@@ -115,7 +115,24 @@ def _impl(
         )
 
     x = ctx.wrap(x_layout)
+    x = ak.operations.ak_values_astype._impl(
+        x,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     weight = ctx.wrap(weight_layout, allow_other=True)
+    if weight is not None:
+        weight = ak.operations.ak_values_astype._impl(
+            weight,
+            np.float64,
+            including_unknown=False,
+            highlevel=True,
+            behavior=ctx.behavior,
+            attrs=ctx.attrs,
+        )
 
     with np.errstate(invalid="ignore", divide="ignore"):
         if weight is None:

src/awkward/operations/ak_var.py

@@ -189,7 +189,24 @@ def _impl(x, weight, ddof, axis, keepdims, mask_identity, highlevel, behavior, a
         )
 
     x = ctx.wrap(x_layout)
+    x = ak.operations.ak_values_astype._impl(
+        x,
+        np.float64,
+        including_unknown=False,
+        highlevel=True,
+        behavior=ctx.behavior,
+        attrs=ctx.attrs,
+    )
     weight = ctx.wrap(weight_layout, allow_other=True)
+    if weight is not None:
+        weight = ak.operations.ak_values_astype._impl(
+            weight,
+            np.float64,
+            including_unknown=False,
+            highlevel=True,
+            behavior=ctx.behavior,
+            attrs=ctx.attrs,
+        )
 
     # Handle named axis
     named_axis = _get_named_axis(ctx)

tests/test_3527_descriptive_statistics_float64_casting.py

@@ -0,0 +1,33 @@
+# BSD 3-Clause License; see https://github.com/scikit-hep/awkward/blob/main/LICENSE
+
+from __future__ import annotations
+
+import numpy as np
+
+import awkward as ak
+
+
+def test_int32_overflow():
+    np.random.seed(42)
+    x = np.random.randint(2**21, 2**22, size=1000, dtype=np.int32)
+    y = np.random.randint(2**21, 2**22, size=1000, dtype=np.int32)
+
+    np.testing.assert_allclose(np.sum(x), ak.sum(x))
+    np.testing.assert_allclose(np.mean(x), ak.mean(x))
+    np.testing.assert_allclose(np.var(x), ak.var(x))
+    np.testing.assert_allclose(np.std(x), ak.std(x))
+    np.testing.assert_allclose(np.cov(x, y, ddof=0)[0][1], ak.covar(x, y))
+    np.testing.assert_allclose(np.corrcoef(x, y)[0][1], ak.corr(x, y))
+
+
+def test_int64_overflow():
+    np.random.seed(42)
+    x = np.random.randint(2**61, 2**62, size=1000, dtype=np.int64)
+    y = np.random.randint(2**61, 2**62, size=1000, dtype=np.int64)
+
+    np.testing.assert_allclose(np.sum(x), ak.sum(x))
+    np.testing.assert_allclose(np.mean(x), ak.mean(x))
+    np.testing.assert_allclose(np.var(x), ak.var(x))
+    np.testing.assert_allclose(np.std(x), ak.std(x))
+    np.testing.assert_allclose(np.cov(x, y, ddof=0)[0][1], ak.covar(x, y))
+    np.testing.assert_allclose(np.corrcoef(x, y)[0][1], ak.corr(x, y))