fix feature map (#326)

gpjax/kernels/computations/basis_functions.py

@@ -30,8 +30,7 @@ def cross_covariance(
         """
         z1 = self.compute_features(x)
         z2 = self.compute_features(y)
-        z1 /= self.kernel.num_basis_fns
-        return self.kernel.base_kernel.variance * jnp.matmul(z1, z2.T)
+        return self.scaling * jnp.matmul(z1, z2.T)
 
     def gram(self, inputs: Float[Array, "N D"]) -> DenseLinearOperator:
         r"""Compute an approximate Gram matrix.
@@ -47,9 +46,7 @@ def gram(self, inputs: Float[Array, "N D"]) -> DenseLinearOperator:
                 $`N \times N`$ Gram matrix.
         """
         z1 = self.compute_features(inputs)
-        matrix = jnp.matmul(z1, z1.T)  # shape: (n_samples, n_samples)
-        matrix /= self.kernel.num_basis_fns
-        return DenseLinearOperator(self.kernel.base_kernel.variance * matrix)
+        return DenseLinearOperator(self.scaling * jnp.matmul(z1, z1.T))
 
     def compute_features(self, x: Float[Array, "N D"]) -> Float[Array, "N L"]:
         r"""Compute the features for the inputs.
@@ -66,3 +63,7 @@ def compute_features(self, x: Float[Array, "N D"]) -> Float[Array, "N L"]:
         z = jnp.matmul(x, (frequencies / scaling_factor).T)
         z = jnp.concatenate([jnp.cos(z), jnp.sin(z)], axis=-1)
         return z
+
+    @property
+    def scaling(self):
+        return self.kernel.base_kernel.variance / self.kernel.num_basis_fns