Added some experiments.

oiltest.py

@@ -0,0 +1,52 @@
+import sys
+import glob
+import os
+import random
+
+import numpy as np
+import numpy.linalg as linalg
+import numpy.random as rnd
+from mpl_toolkits.mplot3d.axes3d import Axes3D
+
+import GPy
+
+sys.path.append('../tools/')
+sys.path.append('..')
+
+import tools.split_data as split_data
+
+import parallel_GPLVM
+
+Q = 7
+max_iters = 200
+N = 1000
+P = 6
+num_inducing = min(40, N / P) - 1
+
+# Get oildata and divide up
+oildata = GPy.util.datasets.oil()
+Y = oildata['X'][:N]
+path = './oildata/'
+split_data.clean_dir(path)
+split_data.split_data(Y, P, path, 'oil')
+
+# Run the Parallel GPLVM
+options = {}
+options['input'] = path + '/inputs/'
+options['embeddings'] = path + '/embeddings/'
+options['parallel'] = 'local'
+options['iterations'] = max_iters
+options['statistics'] = path + '/tmp'
+options['tmp'] = path + '/tmp'
+options['M'] = num_inducing
+options['Q'] = Q
+options['D'] = 12
+options['fixed_embeddings'] = False
+options['keep'] = True
+options['load'] = False
+options['fixed_beta'] = True
+options['init'] = 'PCA'
+options['optimiser'] = 'SCG_adapted'
+options['fixed_beta'] = False
+
+parallel_GPLVM.main(options)

run-airline-experiment.py

@@ -0,0 +1,52 @@
+import os
+import glob
+import shutil
+import time
+
+import numpy as np
+
+import tools.split_data as split_data
+import parallel_GPLVM
+
+P = 10
+Q = 7
+num_inducing = 40
+path = './flight/'
+dname = 'flight'
+
+# First delete all current inputs & embeddings
+split_data.clean_dir(path)
+
+# Load data
+Y = np.loadtxt(path + '/proc/flight', delimiter=',')
+split_data.split_data(Y, P, path, dname)
+
+# Prepare directories
+reqdirs = ['inputs', 'embeddings', 'tmp', 'proc']
+for dirname in reqdirs:
+    if not os.path.exists(path + '/' + dirname):
+        os.mkdir(path + '/' + dirname)
+
+# Run the Parallel GPLVM
+options = {}
+options['input'] = path + '/inputs/'
+options['embeddings'] = path + '/embeddings/'
+options['parallel'] = 'local'
+options['iterations'] = 1000
+options['statistics'] = path + '/tmp'
+options['tmp'] = path + '/tmp'
+options['M'] = num_inducing
+options['Q'] = Q
+options['D'] = 12
+options['fixed_embeddings'] = False
+options['keep'] = False
+options['load'] = False
+options['fixed_beta'] = True
+options['init'] = 'PCA'
+options['optimiser'] = 'SCG_adapted'
+options['fixed_beta'] = False
+
+parallel_GPLVM.main(options)
+
+# Copy output directory
+shutil.copytree(path, '/scratch/mv310/results/' + dname + str(time.time()))

run-mnist-experiment.py

@@ -0,0 +1,62 @@
+import os
+import glob
+import shutil
+import time
+import gzip
+import cPickle
+
+import numpy as np
+
+import tools.split_data as split_data
+import parallel_GPLVM
+
+P = 60
+Q = 20
+num_inducing = 100
+path = './mnist/'
+dname = 'mnist'
+
+# First delete all current inputs & embeddings
+split_data.clean_dir(path)
+
+# Load data
+f = gzip.open(path + '/proc/mnist.pkl.gz', 'rb')
+train_set, valid_set, test_set = cPickle.load(f)
+f.close()
+
+Y = train_set[0][:N, :]
+
+split_data.split_data(Y, P, path, dname)
+
+# Prepare directories
+reqdirs = ['inputs', 'embeddings', 'tmp', 'proc']
+for dirname in reqdirs:
+    if not os.path.exists(path + '/' + dirname):
+        os.mkdir(path + '/' + dirname)
+
+# Run the Parallel GPLVM
+options = {}
+options['input'] = path + '/inputs/'
+options['embeddings'] = path + '/embeddings/'
+options['parallel'] = 'local'
+options['iterations'] = 1000
+options['statistics'] = path + '/tmp'
+options['tmp'] = path + '/tmp'
+options['M'] = num_inducing
+options['Q'] = Q
+options['D'] = Y.shape[1]
+options['fixed_embeddings'] = False
+options['keep'] = False
+options['load'] = False
+options['fixed_beta'] = True
+options['init'] = 'PCA'
+options['optimiser'] = 'SCG_adapted'
+options['fixed_beta'] = False
+
+start_time = time.time()
+parallel_GPLVM.main(options)
+print('Elapsed time:')
+print(time.time() - start_time)
+
+# Copy output directory
+shutil.copytree(path, '/scratch/mv310/results/' + dname + str(time.time()))

tools/show_embeddings.py

@@ -105,7 +105,8 @@ def vprint(s):
 
     if (opt.plotard):
         fig = plt.figure()
-
+        print alpha
+        plt.bar(range(len(alpha)), alpha)
 
     # Plot the outputs as a func of the X
     # if (not opt.output_dimension is None):
@@ -125,7 +126,7 @@ def vprint(s):
 
     plt.show()
 
-    return parGPLVM_X
+    return (parGPLVM_X, alpha)
 
 if __name__ == '__main__':
     usage = "usage: %prog [options] data_dir"