Added mlp mapping (with possibility of having multiple layers).

GPy/kern/constructors.py

@@ -69,6 +69,40 @@ def mlp(input_dim,variance=1., weight_variance=None,bias_variance=100.,ARD=False
     part = parts.mlp.MLP(input_dim,variance,weight_variance,bias_variance,ARD)
     return kern(input_dim, [part])
 
+# def gibbs(input_dim,variance=1., mapping=None):
+#     """
+#     Gibbs and MacKay non-stationary covariance function.
+
+#     .. math::
+       
+#        r = sqrt((x_i - x_j)'*(x_i - x_j))
+       
+#        k(x_i, x_j) = \sigma^2*Z*exp(-r^2/(l(x)*l(x) + l(x')*l(x')))
+
+#        Z = \sqrt{2*l(x)*l(x')/(l(x)*l(x) + l(x')*l(x')}
+
+#        where :math:`l(x)` is a function giving the length scale as a function of space.
+#        This is the non stationary kernel proposed by Mark Gibbs in his 1997
+#         thesis. It is similar to an RBF but has a length scale that varies
+#         with input location. This leads to an additional term in front of
+#         the kernel.
+
+#         The parameters are :math:`\sigma^2`, the process variance, and the parameters of l(x) which is a function that can be specified by the user, by default an multi-layer peceptron is used is used.
+
+#         :param input_dim: the number of input dimensions
+#         :type input_dim: int 
+#         :param variance: the variance :math:`\sigma^2`
+#         :type variance: float
+#         :param mapping: the mapping that gives the lengthscale across the input space.
+#         :type mapping: GPy.core.Mapping
+#         :param ARD: Auto Relevance Determination. If equal to "False", the kernel is isotropic (ie. one weight variance parameter \sigma^2_w), otherwise there is one weight variance parameter per dimension.
+#         :type ARD: Boolean
+#         :rtype: Kernpart object
+
+#     """
+#     part = parts.gibbs.Gibbs(input_dim,variance,mapping)
+#     return kern(input_dim, [part])
+
 def poly(input_dim,variance=1., weight_variance=None,bias_variance=1.,degree=2, ARD=False):
     """
     Construct a polynomial kernel
@@ -312,10 +346,10 @@ def coregionalise(output_dim, rank=1, W=None, kappa=None):
 
     Used for computing covariance functions of the form
     .. math::
-       k_2(x, y)=B k(x, y)
+       k_2(x, y)=\mathbf{B} k(x, y)
     where
     .. math::
-       B = WW^\top + kappa I..
+       \mathbf{B} = \mathbf{W}\mathbf{W}^\top + kappa \mathbf{I}
 
     :param output_dim: the number of output dimensions
     :type output_dim: int

GPy/kern/parts/__init__.py

@@ -4,6 +4,7 @@ import coregionalise
 import exponential
 import finite_dimensional
 import fixed
+import gibbs
 import independent_outputs
 import linear
 import Matern32