def load_contrast(n=13, k=.01, filter="box", contrast="mean", want_dense=False): X = TorontoFace.load(want_mean=False, want_dense=True) X = X[:100000] X_mean = X.mean(axis=1) X_std = (X.var(1) + 10)**.5 X = (X - X_mean[:, np.newaxis]) / X_std[:, np.newaxis] X = X.reshape(-1, 1, 48, 48) if filter == "box": filters = nn.ones((n, n)) elif filter == "gaussian": filters = nn.gaussian2D(n, k) else: raise Exception("unhandled case") X_cn = nn.zeros(X.shape) for i in xrange(10): if contrast == "mean": X_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrastMean( X[10000 * i:10000 * (i + 1), :, :, :], filters) elif contrast == "mean-var": X_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrast( X[10000 * i:10000 * (i + 1), :, :, :], filters) else: raise Exception("unhandled case") if want_dense: X_cn = X_cn.reshape(-1, 48 * 48) return X_cn
def load_contrast(size=32, n=19,k=.01,filter="gaussian",contrast="mean-var"): X,T,X_test,T_test,T_train_labels,T_labels = NORB.load(size=size,want_mean=False,want_dense=True) X = X[:20000] X_mean=X.mean(axis=1) X_std = (X.var(1)+10)**.5 X = (X-X_mean[:, np.newaxis])/X_std[:, np.newaxis] X = X.reshape(-1,1,32,32) if filter=="box": filters = nn.ones((n,n)) elif filter=="gaussian": filters = nn.gaussian2D(n,k) else: raise Exception("unhandled case") X_cn = nn.zeros(X.shape) for i in xrange(2): if contrast=="mean": X_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrastMean(X[10000*i:10000*(i+1),:,:,:], filters) elif contrast=="mean-var": X_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrast(X[10000*i:10000*(i+1),:,:,:], filters) else: raise Exception("unhandled case") X_cn = X_cn.reshape(-1,1,size,size) return X_cn,T,X_test,T_test,T_train_labels,T_labels
def load_contrast(n=13,k=.01,filter="box",contrast="mean",want_dense = False): X = TorontoFace.load(want_mean=False,want_dense=True) X = X[:100000] X_mean=X.mean(axis=1) X_std = (X.var(1)+10)**.5 X = (X-X_mean[:, np.newaxis])/X_std[:, np.newaxis] X = X.reshape(-1,1,48,48) if filter=="box": filters = nn.ones((n,n)) elif filter=="gaussian": filters = nn.gaussian2D(n,k) else: raise Exception("unhandled case") X_cn = nn.zeros(X.shape) for i in xrange(10): if contrast=="mean": X_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrastMean(X[10000*i:10000*(i+1),:,:,:], filters) elif contrast=="mean-var": X_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrast(X[10000*i:10000*(i+1),:,:,:], filters) else: raise Exception("unhandled case") if want_dense: X_cn = X_cn.reshape(-1,48*48) return X_cn
def load_contrast(size=32, n=19, k=.01, filter="gaussian", contrast="mean-var"): X, T, X_test, T_test, T_train_labels, T_labels = NORB.load( size=size, want_mean=False, want_dense=True) X = X[:20000] X_mean = X.mean(axis=1) X_std = (X.var(1) + 10)**.5 X = (X - X_mean[:, np.newaxis]) / X_std[:, np.newaxis] X = X.reshape(-1, 1, 32, 32) if filter == "box": filters = nn.ones((n, n)) elif filter == "gaussian": filters = nn.gaussian2D(n, k) else: raise Exception("unhandled case") X_cn = nn.zeros(X.shape) for i in xrange(2): if contrast == "mean": X_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrastMean( X[10000 * i:10000 * (i + 1), :, :, :], filters) elif contrast == "mean-var": X_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrast( X[10000 * i:10000 * (i + 1), :, :, :], filters) else: raise Exception("unhandled case") X_cn = X_cn.reshape(-1, 1, size, size) return X_cn, T, X_test, T_test, T_train_labels, T_labels
def load_contrast(n=13,k=.01,want_dense=False,filter="gaussian",contrast="mean"): X,T,X_test,T_test,T_train_labels,T_labels = CIFAR10.load(want_mean=False,want_dense=True) # for i in xrange(3): # mean = X[:,i,:,:].mean() # std = X[:,i,:,:].std() # X[:,i,:,:] = (X[:,i,:,:] - mean)/std # X_test[:,i,:,:] = (X_test[:,i,:,:] - mean)/std X_mean=X.mean(axis=1) X_test_mean=X_test.mean(axis=1) X_std = X.std(1) X_test_std = X_test.std(1) X = (X-X_mean[:, np.newaxis])/X_std[:, np.newaxis] X_test = (X_test-X_test_mean[:, np.newaxis])/X_test_std[:, np.newaxis] X = X.reshape(50000,3,32,32) X_test = X_test.reshape(10000,3,32,32) if filter=="box": filters = nn.ones((n,n)) elif filter=="gaussian": filters = nn.gaussian2D(n,k) else: raise Exception("unhandled case") X_cn = nn.zeros(X.shape) X_test_cn = nn.zeros(X_test.shape) for i in xrange(5): if contrast=="mean": X_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrastMean(X[10000*i:10000*(i+1),:,:,:], filters) elif contrast=="mean-var": X_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrast(X[10000*i:10000*(i+1),:,:,:], filters) else: raise Exception("unhandled case") for i in xrange(1): if contrast=="mean": X_test_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrastMean(X_test[10000*i:10000*(i+1),:,:,:], filters) elif contrast=="mean-var": X_test_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrast(X_test[10000*i:10000*(i+1),:,:,:], filters) else: raise Exception("unhandled case") if want_dense: X_cn = X_cn.reshape(50000,3072) X_test_cn = X_test_cn.reshape(10000,3072) return X_cn,T,X_test_cn,T_test,T_train_labels,T_labels
def load_contrast_extra(n=21, k=.01): X, T, X_test, T_test, T_train_labels, T_labels = SVHN.load_extra( want_mean=False, want_dense=True) print "Contrast Normalization." X_mean = X.mean(axis=1) X_test_mean = X_test.mean(axis=1) X_std = X.std(1) X_test_std = X_test.std(1) X = (X - X_mean[:, np.newaxis]) / X_std[:, np.newaxis] X_test = (X_test - X_test_mean[:, np.newaxis]) / X_test_std[:, np.newaxis] X = X.reshape(600000, 3, 32, 32) X_test = X_test.reshape(20000, 3, 32, 32) filters = nn.gaussian2D(n, k) X_cn = np.zeros(X.shape) X_test_cn = np.zeros(X_test.shape) for i in xrange(60): if i % 10 == 0: print i, X_cn[i * 10000:(i + 1) * 10000, :, :, :] = nn.SpatialContrast( X[i * 10000:(i + 1) * 10000, :, :, :], filters) for i in xrange(2): X_test_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrast( X_test[10000 * i:10000 * (i + 1), :, :, :], filters) h5f = h5py.File( '/media/nas/PSI-Share-no-backup/Ali/Dataset/SVHN/extra_contrast21.01_new.h5', 'w') print "file created." h5f.create_dataset('X', data=X_cn) h5f.create_dataset('T', data=T) h5f.create_dataset('T_train_labels', data=T_train_labels) print "train done." h5f.create_dataset('X_test', data=X_test_cn) h5f.create_dataset('T_test', data=T_test) h5f.create_dataset('T_labels', data=T_labels) return X_cn, T, X_test_cn, T_test, T_train_labels, T_labels
def load_contrast(n=21, k=.01, want_dense=False): X, T, X_test, T_test, T_train_labels, T_labels = SVHN.load( want_mean=False, want_dense=True) # for i in xrange(3): # mean = X[:,i,:,:].mean() # std = X[:,i,:,:].std() # X[:,i,:,:] = (X[:,i,:,:] - mean)/std # X_test[:,i,:,:] = (X_test[:,i,:,:] - mean)/std X_mean = X.mean(axis=1) X_test_mean = X_test.mean(axis=1) X_std = X.std(1) X_test_std = X_test.std(1) X = (X - X_mean[:, np.newaxis]) / X_std[:, np.newaxis] X_test = (X_test - X_test_mean[:, np.newaxis]) / X_test_std[:, np.newaxis] X = X.reshape(70000, 3, 32, 32) X_test = X_test.reshape(20000, 3, 32, 32) filters = nn.gaussian2D(n, k) X_cn = nn.zeros(X.shape) X_test_cn = nn.zeros(X_test.shape) for i in xrange(7): X_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrast( X[10000 * i:10000 * (i + 1), :, :, :], filters) for i in xrange(2): X_test_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrast( X_test[10000 * i:10000 * (i + 1), :, :, :], filters) if want_dense: X_cn = X_cn.reshape(70000, 3072) X_test_cn = X_test_cn.reshape(20000, 3072) return X_cn, T, X_test_cn, T_test, T_train_labels, T_labels
def load_contrast_extra(n=21,k=.01): X,T,X_test,T_test,T_train_labels,T_labels = SVHN.load_extra(want_mean=False,want_dense=True) print "Contrast Normalization." X_mean=X.mean(axis=1) X_test_mean=X_test.mean(axis=1) X_std = X.std(1) X_test_std = X_test.std(1) X = (X-X_mean[:, np.newaxis])/X_std[:, np.newaxis] X_test = (X_test-X_test_mean[:, np.newaxis])/X_test_std[:, np.newaxis] X = X.reshape(600000,3,32,32) X_test = X_test.reshape(20000,3,32,32) filters = nn.gaussian2D(n,k) X_cn = np.zeros(X.shape) X_test_cn = np.zeros(X_test.shape) for i in xrange(60): if i%10==0: print i, X_cn[i*10000:(i+1)*10000,:,:,:] = nn.SpatialContrast(X[i*10000:(i+1)*10000,:,:,:], filters) for i in xrange(2): X_test_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrast(X_test[10000*i:10000*(i+1),:,:,:], filters) h5f = h5py.File('/media/nas/PSI-Share-no-backup/Ali/Dataset/SVHN/extra_contrast21.01_new.h5', 'w') print "file created." h5f.create_dataset('X', data=X_cn) h5f.create_dataset('T', data=T) h5f.create_dataset('T_train_labels', data=T_train_labels) print "train done." h5f.create_dataset('X_test', data=X_test_cn) h5f.create_dataset('T_test', data=T_test) h5f.create_dataset('T_labels', data=T_labels) return X_cn,T,X_test_cn,T_test,T_train_labels,T_labels
def load_contrast(n=21,k=.01,want_dense=False): X,T,X_test,T_test,T_train_labels,T_labels = SVHN.load(want_mean=False,want_dense=True) # for i in xrange(3): # mean = X[:,i,:,:].mean() # std = X[:,i,:,:].std() # X[:,i,:,:] = (X[:,i,:,:] - mean)/std # X_test[:,i,:,:] = (X_test[:,i,:,:] - mean)/std X_mean=X.mean(axis=1) X_test_mean=X_test.mean(axis=1) X_std = X.std(1) X_test_std = X_test.std(1) X = (X-X_mean[:, np.newaxis])/X_std[:, np.newaxis] X_test = (X_test-X_test_mean[:, np.newaxis])/X_test_std[:, np.newaxis] X = X.reshape(70000,3,32,32) X_test = X_test.reshape(20000,3,32,32) filters = nn.gaussian2D(n,k) X_cn = nn.zeros(X.shape) X_test_cn = nn.zeros(X_test.shape) for i in xrange(7): X_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrast(X[10000*i:10000*(i+1),:,:,:], filters) for i in xrange(2): X_test_cn[10000*i:10000*(i+1),:,:,:] = nn.SpatialContrast(X_test[10000*i:10000*(i+1),:,:,:], filters) if want_dense: X_cn = X_cn.reshape(70000,3072) X_test_cn = X_test_cn.reshape(20000,3072) return X_cn,T,X_test_cn,T_test,T_train_labels,T_labels
def load_contrast(n=13, k=.01, want_dense=False, filter="gaussian", contrast="mean"): X, T, X_test, T_test, T_train_labels, T_labels = CIFAR10.load( want_mean=False, want_dense=True) # for i in xrange(3): # mean = X[:,i,:,:].mean() # std = X[:,i,:,:].std() # X[:,i,:,:] = (X[:,i,:,:] - mean)/std # X_test[:,i,:,:] = (X_test[:,i,:,:] - mean)/std X_mean = X.mean(axis=1) X_test_mean = X_test.mean(axis=1) X_std = X.std(1) X_test_std = X_test.std(1) X = (X - X_mean[:, np.newaxis]) / X_std[:, np.newaxis] X_test = (X_test - X_test_mean[:, np.newaxis]) / X_test_std[:, np.newaxis] X = X.reshape(50000, 3, 32, 32) X_test = X_test.reshape(10000, 3, 32, 32) if filter == "box": filters = nn.ones((n, n)) elif filter == "gaussian": filters = nn.gaussian2D(n, k) else: raise Exception("unhandled case") X_cn = nn.zeros(X.shape) X_test_cn = nn.zeros(X_test.shape) for i in xrange(5): if contrast == "mean": X_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrastMean( X[10000 * i:10000 * (i + 1), :, :, :], filters) elif contrast == "mean-var": X_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrast( X[10000 * i:10000 * (i + 1), :, :, :], filters) else: raise Exception("unhandled case") for i in xrange(1): if contrast == "mean": X_test_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrastMean( X_test[10000 * i:10000 * (i + 1), :, :, :], filters) elif contrast == "mean-var": X_test_cn[10000 * i:10000 * (i + 1), :, :, :] = nn.SpatialContrast( X_test[10000 * i:10000 * (i + 1), :, :, :], filters) else: raise Exception("unhandled case") if want_dense: X_cn = X_cn.reshape(50000, 3072) X_test_cn = X_test_cn.reshape(10000, 3072) return X_cn, T, X_test_cn, T_test, T_train_labels, T_labels