Merge pull request #132 from tdixon97/main

Allow table evaluations to return np.ndarray also without numexpr

Author: gipert

pyproject.toml

@@ -76,6 +76,7 @@ test = [
     "pylegendtestdata",
     "pytest>=6.0",
     "pytest-cov",
+    "dbetto",
 ]
 
 [project.scripts]

src/lgdo/types/table.py

@@ -351,31 +351,39 @@ def eval(
         msg = f"evaluating {expr!r} with locals={(self_unwrap | parameters)} and {has_ak=}"
         log.debug(msg)
 
-        # use numexpr if we are only dealing with numpy data types (and no global dictionary)
-        if not has_ak and modules is None:
-            out_data = ne.evaluate(
-                expr,
-                local_dict=(self_unwrap | parameters),
-            )
-
-            msg = f"...the result is {out_data!r}"
-            log.debug(msg)
-
-            # need to convert back to LGDO
-            # np.evaluate should always return a numpy thing?
-            if out_data.ndim == 0:
-                return Scalar(out_data.item())
-            if out_data.ndim == 1:
-                return Array(out_data)
-            if out_data.ndim == 2:
-                return ArrayOfEqualSizedArrays(nda=out_data)
+        def _make_lgdo(data):
+            if data.ndim == 0:
+                return Scalar(data.item())
+            if data.ndim == 1:
+                return Array(data)
+            if data.ndim == 2:
+                return ArrayOfEqualSizedArrays(nda=data)
 
             msg = (
-                f"evaluation resulted in {out_data.ndim}-dimensional data, "
+                f"evaluation resulted in {data.ndim}-dimensional data, "
                 "I don't know which LGDO this corresponds to"
             )
             raise RuntimeError(msg)
 
+        # use numexpr if we are only dealing with numpy data types (and no global dictionary)
+        if not has_ak and modules is None:
+            try:
+                out_data = ne.evaluate(
+                    expr,
+                    local_dict=(self_unwrap | parameters),
+                )
+
+                msg = f"...the result is {out_data!r}"
+                log.debug(msg)
+
+                # need to convert back to LGDO
+                # np.evaluate should always return a numpy thing?
+                return _make_lgdo(out_data)
+
+            except Exception:
+                msg = f"Warning {expr} could not be evaluated with numexpr probably due to some not allowed characters, trying with eval()."
+                log.debug(msg)
+
         # resort to good ol' eval()
         globs = {"ak": ak, "np": np}
         if modules is not None:
@@ -392,6 +400,10 @@ def eval(
                 return Array(out_data.to_numpy())
             return VectorOfVectors(out_data)
 
+        # modules can still produce numpy array
+        if isinstance(out_data, np.ndarray):
+            return _make_lgdo(out_data)
+
         if np.isscalar(out_data):
             return Scalar(out_data)
 

tests/types/test_table_eval.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import dbetto
 import hist
 import numpy as np
 import pytest
@@ -85,6 +86,19 @@ def test_eval_dependency():
     res = obj.eval("lgdo.Array([1,2,3])", {}, modules={"lgdo": lgdo})
     assert res == lgdo.Array([1, 2, 3])
 
+    # test with module returning np.array
+    assert obj.eval("np.sum(a)", {}, modules={"np": np}).value == np.int64(10)
+
     # check bad type
     with pytest.raises(RuntimeError):
         obj.eval("hist.Hist()", modules={"hist": hist})
+
+    # check impossible numexpr can still run
+    assert np.allclose(
+        obj.eval(
+            "a*args.value",
+            {"args": dbetto.AttrsDict({"value": 2})},
+            modules={"lgdo": lgdo},
+        ).view_as("np"),
+        [2, 4, 6, 8],
+    )