Fix problem with lambdas wrapping numba functions get recompiled

sgkit/stats/aggregation.py

@@ -680,7 +680,7 @@ def variant_stats(
     --------
     :func:`count_variant_genotypes`
     """
-    from .aggregation_numba_fns import count_hom
+    from .aggregation_numba_fns import count_hom_new_axis
 
     variables.validate(ds, {call_genotype: variables.call_genotype_spec})
     mixed_ploidy = ds[call_genotype].attrs.get("mixed_ploidy", False)
@@ -697,7 +697,7 @@ def variant_stats(
     G = da.asarray(ds[call_genotype].data)
     H = xr.DataArray(
         da.map_blocks(
-            lambda *args: count_hom(*args)[:, np.newaxis, :],
+            count_hom_new_axis,
             G,
             np.zeros(3, np.uint64),
             drop_axis=2,
@@ -796,7 +796,7 @@ def sample_stats(
     ValueError
         If the dataset contains mixed-ploidy genotype calls.
     """
-    from .aggregation_numba_fns import count_hom
+    from .aggregation_numba_fns import count_hom_new_axis
 
     variables.validate(ds, {call_genotype: variables.call_genotype_spec})
     mixed_ploidy = ds[call_genotype].attrs.get("mixed_ploidy", False)
@@ -805,7 +805,7 @@ def sample_stats(
     GT = da.asarray(ds[call_genotype].transpose("samples", "variants", "ploidy").data)
     H = xr.DataArray(
         da.map_blocks(
-            lambda *args: count_hom(*args)[:, np.newaxis, :],
+            count_hom_new_axis,
             GT,
             np.zeros(3, np.uint64),
             drop_axis=2,

sgkit/stats/aggregation_numba_fns.py

@@ -2,6 +2,8 @@
 # in a separate file here, and imported dynamically to avoid
 # initial compilation overhead.
 
+import numpy as np
+
 from sgkit.accelerate import numba_guvectorize, numba_jit
 from sgkit.typing import ArrayLike
 
@@ -102,3 +104,7 @@ def count_hom(
         index = _classify_hom(genotypes[i])
         if index >= 0:
             out[index] += 1
+
+
+def count_hom_new_axis(genotypes: ArrayLike, _: ArrayLike) -> ArrayLike:
+    return count_hom(genotypes, _)[:, np.newaxis, :]

sgkit/stats/popgen.py

@@ -595,9 +595,7 @@ def pbs(
     cohorts = cohorts or list(itertools.combinations(range(n_cohorts), 3))  # type: ignore
     ct = _cohorts_to_array(cohorts, ds.indexes.get("cohorts_0", None))
 
-    p = da.map_blocks(
-        lambda t: _pbs_cohorts(t, ct), t, chunks=shape, new_axis=3, dtype=np.float64
-    )
+    p = da.map_blocks(_pbs_cohorts, t, ct, chunks=shape, new_axis=3, dtype=np.float64)
     assert_array_shape(p, n_windows, n_cohorts, n_cohorts, n_cohorts)
 
     new_ds = create_dataset(