Use random input instead of NumPy empty values in JAX test

`np.empty` was producing values at the edge of numerical zero. These values would apparently cause `jax` to truncate something during and intermediate step of a comparison operation like `array > 0`, which would produce a result different from standard NumPy's.

Use random input instead of NumPy empty values in JAX test
d5b9940d · Brandon T. Willard · 9ea670ca · d5b9940d
--- a/tests/sandbox/test_jax.py
+++ b/tests/sandbox/test_jax.py
@@ -354,7 +354,7 @@ def test_jax_Subtensors():


 def test_jax_IncSubtensor():
-    x_np = np.empty((3, 4, 5), dtype=tt.config.floatX)
+    x_np = np.random.uniform(-1, 1, size=(3, 4, 5)).astype(tt.config.floatX)
    x_tt = tt.arange(3 * 4 * 5).reshape((3, 4, 5)).astype(tt.config.floatX)

    # "Set" basic indices