Feature: Expr and GenExpr support NumPy binary functions like np.add

CHANGELOG.md

@@ -2,7 +2,8 @@
 
 ## Unreleased
 ### Added
-- `Expr` and `GenExpr` support NumPy unary functions (`np.sin`, `np.cos`, `np.sqrt`, `np.exp`, `np.log`, `np.absolute`)
+- `Expr` and `GenExpr` support NumPy unary functions (`np.sin`, `np.cos`, `np.sqrt`, `np.exp`, `np.log`, `np.absolute`, `np.negative`)
+- `Expr` and `GenExpr` support NumPy binary functions (`np.add`, `np.subtract`, `np.multiply`, `np.divide`, `np.true_divide`, `np.power`, `np.less_equal`, `np.greater_equal`, `np.equal`)
 - Added `getBase()` and `setBase()` methods to `LP` class for getting/setting basis status
 - Added `getMemUsed()`, `getMemTotal()`, and `getMemExternEstim()` methods
 ### Fixed

src/pyscipopt/expr.pxi

@@ -246,7 +246,32 @@ cdef class ExprLike:
             )
 
         if method == "__call__":
-            if ufunc is np.absolute:
+            if arrays := [a for a in args if type(a) is np.ndarray]:
+                if any(a.dtype.kind not in "fiub" for a in arrays):
+                    return NotImplemented
+                # If the np.ndarray is of numeric type, all arguments are converted to
+                # MatrixExpr or MatrixGenExpr and then the ufunc is applied.
+                return ufunc(*[_ensure_matrix(a) for a in args], **kwargs)
+
+            if ufunc is np.add:
+                return args[0] + args[1]
+            elif ufunc is np.subtract:
+                return args[0] - args[1]
+            elif ufunc is np.multiply:
+                return args[0] * args[1]
+            elif ufunc in {np.divide, np.true_divide}:
+                return args[0] / args[1]
+            elif ufunc is np.power:
+                return args[0] ** args[1]
+            elif ufunc is np.negative:
+                return -args[0]
+            elif ufunc is np.less_equal:
+                return args[0] <= args[1]
+            elif ufunc is np.greater_equal:
+                return args[0] >= args[1]
+            elif ufunc is np.equal:
+                return args[0] == args[1]
+            elif ufunc is np.absolute:
                 return args[0].__abs__()
             elif ufunc is np.exp:
                 return args[0].exp()
@@ -1031,6 +1056,12 @@ cdef inline object _wrap_ufunc(object x, object ufunc):
         return res.view(MatrixGenExpr) if isinstance(res, np.ndarray) else res
     return ufunc(_to_const(x))
 
+cdef inline object _ensure_matrix(object arg):
+    if type(arg) is np.ndarray:
+        return arg.view(MatrixExpr)
+    matrix = MatrixExpr if isinstance(arg, Expr) else MatrixGenExpr
+    return np.array(arg, dtype=object).view(matrix)
+
 
 def expr_to_nodes(expr):
     '''transforms tree to an array of nodes. each node is an operator and the position of the 

tests/test_expr.py

@@ -222,7 +222,7 @@ def test_getVal_with_GenExpr():
         m.getVal(1 / z)
 
 
-def test_unary(model):
+def test_unary_ufunc(model):
     m, x, y, z = model
 
     res = "abs(sum(0.0,prod(1.0,x)))"
@@ -276,6 +276,45 @@ def test_unary(model):
         # forbid modifying Variable/Expr/GenExpr in-place via out parameter
         np.sin(x, out=np.array([0]))
 
+    # test np.negative
+    assert str(np.negative(x)) == "Expr({Term(x): -1.0})"
+
+
+def test_binary_ufunc(model):
+    m, x, y, z = model
+
+    # test np.add
+    assert str(np.add(x, 1)) == "Expr({Term(x): 1.0, Term(): 1.0})"
+    assert str(np.add(1, x)) == "Expr({Term(x): 1.0, Term(): 1.0})"
+    a = np.array([1])
+    assert str(np.add(x, a)) == "[Expr({Term(x): 1.0, Term(): 1.0})]"
+    assert str(np.add(a, x)) == "[Expr({Term(x): 1.0, Term(): 1.0})]"
+
+    # test np.subtract
+    assert str(np.subtract(x, 1)) == "Expr({Term(x): 1.0, Term(): -1.0})"
+    assert str(np.subtract(1, x)) == "Expr({Term(x): -1.0, Term(): 1.0})"
+    assert str(np.subtract(x, a)) == "[Expr({Term(x): 1.0, Term(): -1.0})]"
+    assert str(np.subtract(a, x)) == "[Expr({Term(x): -1.0, Term(): 1.0})]"
+
+    # test np.multiply
+    a = np.array([2])
+    assert str(np.multiply(x, 2)) == "Expr({Term(x): 2.0})"
+    assert str(np.multiply(2, x)) == "Expr({Term(x): 2.0})"
+    assert str(np.multiply(x, a)) == "[Expr({Term(x): 2.0})]"
+    assert str(np.multiply(a, x)) == "[Expr({Term(x): 2.0})]"
+
+    # test np.divide
+    assert str(np.divide(x, 2)) == "Expr({Term(x): 0.5})"
+    assert str(np.divide(2, x)) == "prod(2.0,**(sum(0.0,prod(1.0,x)),-1))"
+    assert str(np.divide(x, a)) == "[Expr({Term(x): 0.5})]"
+    assert str(np.divide(a, x)) == "[prod(2.0,**(sum(0.0,prod(1.0,x)),-1))]"
+
+    # test np.power
+    assert str(np.power(x, 2)) == "Expr({Term(x, x): 1.0})"
+    assert str(np.power(2, x)) == "exp(prod(1.0,sum(0.0,prod(1.0,x)),log(2.0)))"
+    assert str(np.power(x, a)) == "[Expr({Term(x, x): 1.0})]"
+    assert str(np.power(a, x)) == "[exp(prod(1.0,sum(0.0,prod(1.0,x)),log(2.0)))]"
+
 
 def test_mul():
     m = Model()