Upload files to ''

README.md

@@ -1,2 +1,10 @@
-# DeepGenePrior
-
+
+Dependencies
+
+    TensorFlow >= 0.8.0 (due to prettytensor, might work with older versions of prettytensor - not tested)
+    prettytensor
+    numpy
+    optionally matplotlib, seaborn for VAE images
+Usage
+
+    To train latent-feature model (M1) run ssl_vae.py. Parameters set in same file.

neuralnetworks.py

@@ -0,0 +1,37 @@
+###
+'''
+Similar to M1 from https://github.com/dpkingma/nips14-ssl
+Original Author: S. Saemundsson
+Edited by: Z. Rahaie
+'''
+###
+
+import tensorflow as tf
+import prettytensor as pt
+import numpy as np
+import utils
+
+class FullyConnected( object ):
+
+	def __init__( 	self,
+					dim_output,
+					hidden_layers,
+					nonlinearity = tf.nn.softplus,
+					l2loss = 0.0,
+					name = 'FullyConnected'	):
+
+		self.dim_output = dim_output
+		self.hidden_layers = hidden_layers
+		self.nonlinearity = nonlinearity
+		self.l2loss = l2loss
+
+	def output( self, inputs, phase = pt.Phase.train ):
+
+		inputs = pt.wrap( inputs )
+		with pt.defaults_scope( phase = phase, activation_fn = self.nonlinearity, l2loss = self.l2loss ):
+			for layer in self.hidden_layers:
+				inputs = inputs.fully_connected( layer )
+
+			return inputs.fully_connected( self.dim_output, activation_fn = None )
+
+	

ssl_vae.py

@@ -0,0 +1,58 @@
+###
+'''
+Similar to M1 from https://github.com/dpkingma/nips14-ssl
+Original Author: S. Saemundsson
+Edited by: Z. Rahaie
+'''
+###
+from vae import VariationalAutoencoder
+import numpy as np
+import data.asd as asd
+
+if __name__ == '__main__':
+
+    num_batches = 0      
+    dim_z = 0             
+    epochs = 0           
+    learning_rate = 1e-4    
+    l2_loss = 1e-5          
+    seed = 12345            
+
+    hidden_layers_px = [ sqrt(input_size), sqrt(hidden_layers_px[0]),  sqrt(hidden_layers_px[1])]  
+    hidden_layers_qz = [ sqrt(input_size), sqrt(hidden_layers_px[0]), sqrt(hidden_layers_px[2]) ]  
+
+    
+    asd_path = ['nds/AutDB_ASD_cnv_dataset.txt']
+    #Uses anglpy module exists kingma github (linked before) to divide the dataset
+    train_x, train_y, valid_x, valid_y, test_x, test_y = asd.load_numpy(asd_path, binarize_y=True)
+
+    x_train, y_train = train_x.T, train_y.T
+    x_valid, y_valid = valid_x.T, valid_y.T
+    x_test, y_test = test_x.T, test_y.T
+
+    dim_x = x_train.shape[1]
+    dim_y = y_train.shape[1]
+
+
+    VAE = VariationalAutoencoder(   dim_x = dim_x, 
+                                    dim_z = dim_z,
+                                    hidden_layers_px = hidden_layers_px,
+                                    hidden_layers_qz = hidden_layers_qz,
+                                    l2_loss = l2_loss )
+
+    #every n iterations (set to 0 to disable)
+
+    VAE.train(  x = x_train, x_valid = x_valid, epochs = epochs, num_batches = num_batches,
+                learning_rate = learning_rate, seed = seed, stop_iter = 30, print_every = 10, draw_img = 0 )
+    
+    weights_as_numpy = VAE.get_weights()
+    
+    output_weights_to_label = weights_as_numpy[6]
+    
+    scores = weights_as_numpy[0] * weights_as_numpy[1] * weights_as_numpy [2] * output_weights_to_label
+    
+    genes_index = [i for i,v in enumerate(scores) if v > 0]
+    
+    print(genes_index)
+    
+    

train_classifier.py

@@ -0,0 +1,104 @@
+###
+'''
+Similar to M1 from https://github.com/dpkingma/nips14-ssl
+Original Author: S. Saemundsson
+Edited by: Z. Rahaie
+'''
+###
+
+from genclass import GenerativeClassifier
+from vae import VariationalAutoencoder
+import numpy as np
+import data.asd as asd
+
+def encode_dataset( model_path, min_std = 0.0 ):
+
+    VAE = VariationalAutoencoder( dim_x, dim_z ) 
+    with VAE.session:
+        VAE.saver.restore( VAE.session, VAE_model_path )
+
+        enc_x_lab_mean, enc_x_lab_var = VAE.encode( x_lab )
+        enc_x_ulab_mean, enc_x_ulab_var = VAE.encode( x_ulab )
+        enc_x_valid_mean, enc_x_valid_var = VAE.encode( x_valid )
+        enc_x_test_mean, enc_x_test_var = VAE.encode( x_test )
+
+        id_x_keep = np.std( enc_x_ulab_mean, axis = 0 ) > min_std
+
+        enc_x_lab_mean, enc_x_lab_var = enc_x_lab_mean[ :, id_x_keep ], enc_x_lab_var[ :, id_x_keep ]
+        enc_x_ulab_mean, enc_x_ulab_var = enc_x_ulab_mean[ :, id_x_keep ], enc_x_ulab_var[ :, id_x_keep ]
+        enc_x_valid_mean, enc_x_valid_var = enc_x_valid_mean[ :, id_x_keep ], enc_x_valid_var[ :, id_x_keep ]
+        enc_x_test_mean, enc_x_test_var = enc_x_test_mean[ :, id_x_keep ], enc_x_test_var[ :, id_x_keep ]
+
+        data_lab = np.hstack( [ enc_x_lab_mean, enc_x_lab_var ] )
+        data_ulab = np.hstack( [ enc_x_ulab_mean, enc_x_ulab_var ] )
+        data_valid = np.hstack( [enc_x_valid_mean, enc_x_valid_var] )
+        data_test = np.hstack( [enc_x_test_mean, enc_x_test_var] )
+
+    return data_lab, data_ulab, data_valid, data_test
+
+if __name__ == '__main__':
+    
+    
+    num_batches = 100       #Number of minibatches in a single epoch
+    epochs = 1001           #Number of epochs through the full dataset
+    learning_rate = 1e-4    #Learning rate of ADAM
+    alpha = 0.1             #Discriminatory factor (see equation (9) of http://arxiv.org/pdf/1406.5298v2.pdf)
+    seed = 12345            #Seed for RNG
+
+
+    ####################
+    ''' Load Dataset '''
+    ####################
+
+    asd_path = ['nds/asd_case.txt', 'nds/asd_control.txt']
+    #Uses anglpy module from original paper (linked at top) to split the dataset for semi-supervised training
+    train_x, train_y, valid_x, valid_y, test_x, test_y = mnist.load_numpy_split(mnist_path, binarize_y=True) 
+    x_l, y_l, x_u, y_u = mnist.create_semisupervised(train_x, train_y, num_lab)
+
+    x_lab, y_lab = x_l.T, y_l.T
+    x_ulab, y_ulab = x_u.T, y_u.T
+    x_valid, y_valid = valid_x.T, valid_y.T
+    x_test, y_test = test_x.T, test_y.T
+
+    ################
+    ''' Load VAE '''
+    ################
+
+    VAE_model_path = 'temp/mid_training_0.cpkt'
+    min_std = 0.1 #Dimensions with std < min_std are removed before training with GC
+
+    data_lab, data_ulab, data_valid, data_test = encode_dataset( VAE_model_path, min_std )
+
+    dim_x = data_lab.shape[1] / 2
+    dim_y = y_lab.shape[1]
+    num_examples = data_lab.shape[0] + data_ulab.shape[0]
+
+    ###################################
+    ''' Train Generative Classifier '''
+    ###################################
+
+    GC = GenerativeClassifier(  dim_x, dim_z, dim_y,
+                                num_examples, num_lab, num_batches,
+                                hidden_layers_px    = hidden_layers_px, 
+                                hidden_layers_qz    = hidden_layers_qz, 
+                                hidden_layers_qy    = hidden_layers_qy,
+                                alpha               = alpha )
+
+    GC.train(   x_labelled      = data_lab, y = y_lab, x_unlabelled = data_ulab,
+                x_valid         = data_valid, y_valid = y_valid,
+                epochs          = epochs, 
+                learning_rate   = learning_rate,
+                seed            = seed,
+                print_every     = 10,
+                load_path       = None )
+
+
+    ############################
+    ''' Evaluate on Test Set '''
+    ############################
+
+    GC_eval = GenerativeClassifier(  dim_x, dim_z, dim_y, num_examples, num_lab, num_batches )
+
+    with GC_eval.session:
+        GC_eval.saver.restore( GC_eval.session, GC.save_path )
+        GC_eval.predict_labels( data_test, y_test )

util.py

@@ -0,0 +1,66 @@
+###
+'''
+Similar to M1 from https://github.com/dpkingma/nips14-ssl
+Original Author: S. Saemundsson
+Edited by: Z. Rahaie
+'''
+###
+
+import prettytensor as pt
+import tensorflow as tf
+import numpy as np
+
+logc = np.log(2.*np.pi)
+c = - 0.5 * np.log(2*np.pi)
+
+def tf_normal_logpdf(x, mu, log_sigma_sq):
+
+	return ( - 0.5 * logc - log_sigma_sq / 2. - tf.div( tf.square( tf.sub( x, mu ) ), 2 * tf.exp( log_sigma_sq ) ) )
+
+def tf_stdnormal_logpdf(x):
+
+	return ( - 0.5 * ( logc + tf.square( x ) ) )
+
+def tf_gaussian_ent(log_sigma_sq):
+
+	return ( - 0.5 * ( logc + 1.0 + log_sigma_sq ) )
+
+def tf_gaussian_marg(mu, log_sigma_sq):
+
+	return ( - 0.5 * ( logc + ( tf.square( mu ) + tf.exp( log_sigma_sq ) ) ) )
+
+def tf_binary_xentropy(x, y, const = 1e-10):
+
+    return - ( x * tf.log ( tf.clip_by_value( y, const, 1.0 ) ) + \
+             (1.0 - x) * tf.log( tf.clip_by_value( 1.0 - y, const, 1.0 ) ) )
+
+def feed_numpy_semisupervised(num_lab_batch, num_ulab_batch, x_lab, y, x_ulab):
+
+	size = x_lab.shape[0] + x_ulab.shape[0]
+	batch_size = num_lab_batch + num_ulab_batch
+	count = int(size / batch_size)
+
+	dim = x_lab.shape[1]
+
+	for i in xrange(count):
+		start_lab = i * num_lab_batch
+		end_lab = start_lab + num_lab_batch
+		start_ulab = i * num_ulab_batch
+		end_ulab = start_ulab + num_ulab_batch
+
+		yield [	x_lab[start_lab:end_lab,:dim/2], x_lab[start_lab:end_lab,dim/2:dim], y[start_lab:end_lab],
+				x_ulab[start_ulab:end_ulab,:dim/2], x_ulab[start_ulab:end_ulab,dim/2:dim] ]
+
+def feed_numpy(batch_size, x):
+
+	size = x.shape[0]
+	count = int(size / batch_size)
+
+	dim = x.shape[1]
+
+	for i in xrange(count):
+		start = i * batch_size
+		end = start + batch_size
+
+		yield x[start:end]
+