Merge branch 'naniny' into devel

GPy/models.py

@@ -4,20 +4,20 @@ Created on 14 Nov 2013
 @author: maxz
 '''
 
-from _models.bayesian_gplvm import BayesianGPLVM
-from _models.gp_regression import GPRegression
-from _models.gp_classification import GPClassification#; _gp_classification = gp_classification ; del gp_classification 
-from _models.sparse_gp_regression import SparseGPRegression#; _sparse_gp_regression = sparse_gp_regression ; del sparse_gp_regression 
-from _models.svigp_regression import SVIGPRegression#; _svigp_regression = svigp_regression ; del svigp_regression 
-from _models.sparse_gp_classification import SparseGPClassification#; _sparse_gp_classification = sparse_gp_classification ; del sparse_gp_classification 
-from _models.fitc_classification import FITCClassification#; _fitc_classification = fitc_classification ; del fitc_classification 
-from _models.gplvm import GPLVM#; _gplvm = gplvm ; del gplvm 
-from _models.bcgplvm import BCGPLVM#; _bcgplvm = bcgplvm; del bcgplvm
-from _models.sparse_gplvm import SparseGPLVM#; _sparse_gplvm = sparse_gplvm ; del sparse_gplvm 
-from _models.warped_gp import WarpedGP#; _warped_gp = warped_gp ; del warped_gp 
-from _models.bayesian_gplvm import BayesianGPLVM#; _bayesian_gplvm = bayesian_gplvm ; del bayesian_gplvm 
-from _models.mrd import MRD#; _mrd = mrd; del mrd 
-from _models.gradient_checker import GradientChecker#; _gradient_checker = gradient_checker ; del gradient_checker 
-from _models.gp_multioutput_regression import GPMultioutputRegression#; _gp_multioutput_regression = gp_multioutput_regression ; del gp_multioutput_regression 
-from _models.sparse_gp_multioutput_regression import SparseGPMultioutputRegression#; _sparse_gp_multioutput_regression = sparse_gp_multioutput_regression ; del sparse_gp_multioutput_regression 
-from _models.gradient_checker import GradientChecker
+from models_modules.bayesian_gplvm import BayesianGPLVM
+from models_modules.gp_regression import GPRegression
+from models_modules.gp_classification import GPClassification#; _gp_classification = gp_classification ; del gp_classification 
+from models_modules.sparse_gp_regression import SparseGPRegression#; _sparse_gp_regression = sparse_gp_regression ; del sparse_gp_regression 
+from models_modules.svigp_regression import SVIGPRegression#; _svigp_regression = svigp_regression ; del svigp_regression 
+from models_modules.sparse_gp_classification import SparseGPClassification#; _sparse_gp_classification = sparse_gp_classification ; del sparse_gp_classification 
+from models_modules.fitc_classification import FITCClassification#; _fitc_classification = fitc_classification ; del fitc_classification 
+from models_modules.gplvm import GPLVM#; _gplvm = gplvm ; del gplvm 
+from models_modules.bcgplvm import BCGPLVM#; _bcgplvm = bcgplvm; del bcgplvm
+from models_modules.sparse_gplvm import SparseGPLVM#; _sparse_gplvm = sparse_gplvm ; del sparse_gplvm 
+from models_modules.warped_gp import WarpedGP#; _warped_gp = warped_gp ; del warped_gp 
+from models_modules.bayesian_gplvm import BayesianGPLVM#; _bayesian_gplvm = bayesian_gplvm ; del bayesian_gplvm 
+from models_modules.mrd import MRD#; _mrd = mrd; del mrd 
+from models_modules.gradient_checker import GradientChecker#; _gradient_checker = gradient_checker ; del gradient_checker 
+from models_modules.gp_multioutput_regression import GPMultioutputRegression#; _gp_multioutput_regression = gp_multioutput_regression ; del gp_multioutput_regression 
+from models_modules.sparse_gp_multioutput_regression import SparseGPMultioutputRegression#; _sparse_gp_multioutput_regression = sparse_gp_multioutput_regression ; del sparse_gp_multioutput_regression 
+from models_modules.gradient_checker import GradientChecker

GPy/models_modules/bayesian_gplvm.py

@@ -12,6 +12,7 @@ from GPy.util import plot_latent, linalg
 from .gplvm import GPLVM
 from GPy.util.plot_latent import most_significant_input_dimensions
 from matplotlib import pyplot
+from GPy.core.model import Model
 
 class BayesianGPLVM(SparseGP, GPLVM):
     """
@@ -285,6 +286,57 @@ class BayesianGPLVM(SparseGP, GPLVM):
         self.init = state.pop()
         SparseGP.setstate(self, state)
 
+class BayesianGPLVMWithMissingData(Model):
+    """
+    Bayesian Gaussian Process Latent Variable Model with missing data support.
+    NOTE: Missing data is assumed to be missing at random!
+    
+    This extension comes with a large memory and computing time deficiency.
+    Use only if fraction of missing data at random is higher than 60%.
+    Otherwise, try filtering data before using this extension.
+    
+    Y can hold missing data as given by `missing`, standard is :class:`~numpy.nan`.
+    
+    If likelihood is given for Y, this likelihood will be discarded, but the parameters
+    of the likelihood will be taken. Also every effort of creating the same likelihood
+    will be done.
+     
+    :param likelihood_or_Y: observed data (np.ndarray) or GPy.likelihood
+    :type likelihood_or_Y: :class:`~numpy.ndarray` | :class:`~GPy.likelihoods.likelihood.likelihood` instance
+    :param int input_dim: latent dimensionality
+    :param init: initialisation method for the latent space
+    :type init: 'PCA' | 'random'
+    """
+    def __init__(self, likelihood_or_Y, input_dim, X=None, X_variance=None, init='PCA', num_inducing=10,
+                 Z=None, kernel=None, missing=np.nan, **kwargs):
+        if type(likelihood_or_Y) is np.ndarray:
+            likelihood = Gaussian(likelihood_or_Y)
+        else:
+            likelihood = likelihood_or_Y
+
+        if X == None:
+            X = self.initialise_latent(init, input_dim, likelihood.Y)
+        self.init = init
+
+        if X_variance is None:
+            X_variance = np.clip((np.ones_like(X) * 0.5) + .01 * np.random.randn(*X.shape), 0.001, 1)
+
+        if Z is None:
+            Z = np.random.permutation(X.copy())[:num_inducing]
+        assert Z.shape[1] == X.shape[1]
+
+        if kernel is None:
+            kernel = kern.rbf(input_dim) # + kern.white(input_dim)
+
+        SparseGP.__init__(self, X, likelihood, kernel, Z=Z, X_variance=X_variance, **kwargs)
+        self.ensure_default_constraints()
+
+    def _get_param_names(self):
+        X_names = sum([['X_%i_%i' % (n, q) for q in range(self.input_dim)] for n in range(self.num_data)], [])
+        S_names = sum([['X_variance_%i_%i' % (n, q) for q in range(self.input_dim)] for n in range(self.num_data)], [])
+        return (X_names + S_names + SparseGP._get_param_names(self))
+
+    pass
 
 def latent_cost_and_grad(mu_S, kern, Z, dL_dpsi0, dL_dpsi1, dL_dpsi2):
     """