drcn

check_results.py

@@ -5,13 +5,15 @@
 src = 'svhn'
 tgt = 'mnist'
 datapair = src+'-'+tgt
-model = 'drcn'
+# model = 'convnet'
+model = 'drcn-st'
 dropout_rate = 0.5
-is_aug = 0
-denoising = 0.4
+is_aug = 1
+denoising = 0.5
 
 RESFILE = 'results/'+datapair+'_'+model+'_results_drop%.1f_aug%d_denoise%.1f.pkl.gz' % (dropout_rate, is_aug, denoising)
-# RESFILE = 'results/'+datapair+'_'+model+'_results_drop%.1f_aug%d.pkl.gz' % (dropout_rate, is_aug)
+# RESFILE = 'results/'+datapair+'_'+model+'_results_drop%.1f_aug%d_h300.pkl.gz' % (dropout_rate, is_aug)
+
 print(RESFILE)
 res = pickle.load(gzip.open(RESFILE,'rb'))
 

loader.py

@@ -166,15 +166,18 @@ def load_mnist(mode=0):
     train_set_x = train_set_x.reshape(train_set_x.shape[0], 1, 28, 28).astype('float32')
     valid_set_x = valid_set_x.reshape(valid_set_x.shape[0], 1, 28, 28).astype('float32')
     test_set_x = test_set_x.reshape(test_set_x.shape[0], 1, 28, 28).astype('float32')
+    train_set_y = train_set_y.astype('uint8')
+    valid_set_y = valid_set_y.astype('uint8')
+    test_set_y = test_set_y.astype('uint8')
 
-    rval = [(train_set_x, train_set_y), (valid_set_x, valid_set_y),
-        (test_set_x, test_set_y)]
+    rval = (train_set_x, train_set_y), (valid_set_x, valid_set_y), (test_set_x, test_set_y)
     return rval
 
 def load_mnist32x32(mode=0):
     # dataset='I:\Data\PhD Life\Tutorial\Python\data\mnist.pkl.gz' # linux
     # dataset='/u/students/gif/Desktop/PhD/Tutorial/dataset/MNIST/mnist32x32.pkl.gz' # linux
     dataset = '/local/scratch/gif/dataset/MNIST/mnist32x32.pkl.gz' #the-villa
+    # dataset = 'I:\Data\PhD Life\Tutorial\dataset\MNIST\mnist32x32.pkl.gz' # laptop
     f = gzip.open(dataset,'rb')
     train_set, valid_set, test_set = pickle.load(f)
     f.close()
@@ -222,12 +225,13 @@ def load_usps(mode=0):
     train_set_x = train_set_x.reshape(train_set_x.shape[0], 1, 28, 28).astype('float32')
     test_set_x = test_set_x.reshape(test_set_x.shape[0], 1, 28, 28).astype('float32')
 
-    rval = [(train_set_x, train_set_y), (test_set_x, test_set_y)]
+    rval = (train_set_x, train_set_y), (test_set_x, test_set_y)
     return rval
 
 def load_svhn():
     # dataset = '/u/students/gif/Desktop/PhD/Tutorial/dataset/SVHN/svhn_gray.pkl.gz' #linux
     dataset = '/local/scratch/gif/dataset/SVHN/svhn_gray.pkl.gz' #the-villa
+    # dataset = 'I:\Data\PhD Life\Tutorial\dataset\SVHN\svhn_gray.pkl.gz' # laptop
     f = gzip.open(dataset,'rb')
     (X_train, y_train), (X_test, y_test) = pickle.load(f)
     f.close()

main_convnet.py

@@ -8,27 +8,38 @@
 
 # hyperparameters
 learning_rate = 1e-4
-batch_size = 128
+batch_size = 32
 nb_epoch = 50
 augmentation = True
 dropout = 0.5
 dense_dim = 300
 
 
-src = 'svhn'
+src = 'usps'
 tgt = 'mnist'
 
 RESFILE = 'results/'+src+'-'+tgt+'_convnet_results_drop%.1f_aug%d_h%d.pkl.gz' % (dropout, augmentation, dense_dim)
 PARAMFILE = 'results/'+src+'-'+tgt+'_convnet_weights_drop%.1f_aug%d_h%d.pkl.gz' % (dropout, augmentation, dense_dim)
 print(PARAMFILE)
 
 # Load data
+
 if src == 'svhn':
 	(X_train, Y_train), (X_test, Y_test) = load_svhn()
 	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist32x32()
 elif src == 'mnist':
-	(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
-	(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
+	if tgt == 'svhn':
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
+	else:
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_usps()
+
+elif src == 'usps':
+	(X_train, Y_train), (X_test, Y_test) = load_usps()
+	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist()
+
+
 
 
 print('Preprocess data ...')

main_drcn.py

@@ -4,24 +4,25 @@
 
 net_config = {
 	'lr': 1e-4,
-	'batch_size': 128,
+	'batch_size': 32,
 	'nb_epoch': 50,
-	'augmentation': False,
+	'augmentation': True,
 	'dropout': 0.5,
-	'shuffle': False,
+	'shuffle': True,
 	'dense_dim': 300,
 	'loss': 'categorical_crossentropy'
 }
 
 ae_config = {
 	'lr': 1e-4,
-	'batch_size': 128,
+	'batch_size': 32,
 	'denoising': 0.4,
-	'shuffle': False,
-	'loss': 'squared_error'
+	'shuffle': True,
+	'loss': 'squared_error',
+	'input': 't'
 }
 
-src = 'svhn'
+src = 'usps'
 tgt = 'mnist'
 
 RESFILE = 'results/'+src+'-'+tgt+'_drcn_results_drop%.1f_aug%d_denoise%.1f.pkl.gz' % (net_config['dropout'], net_config['augmentation'], ae_config['denoising'])
@@ -34,8 +35,17 @@
 	(X_train, Y_train), (X_test, Y_test) = load_svhn()
 	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist32x32()
 elif src == 'mnist':
-	(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
-	(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
+	if tgt == 'svhn':
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
+	else:
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_usps()
+
+elif src == 'usps':
+	(X_train, Y_train), (X_test, Y_test) = load_usps()
+	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist()
+
 
 
 print('Preprocess data ...')

main_drcn_s.py

@@ -8,7 +8,7 @@
 	'nb_epoch': 50,
 	'augmentation': True,
 	'dropout': 0.5,
-	'shuffle': False,
+	'shuffle': True,
 	'dense_dim': 300,
 	'loss': 'categorical_crossentropy'
 }
@@ -17,26 +17,36 @@
 	'lr': 1e-4,
 	'batch_size': 128,
 	'denoising': 0.5,
-	'shuffle': False,
+	'shuffle': True,
 	'loss': 'squared_error',
-	'input': 'src'
+	'input': 'st'
 }
 
 src = 'svhn'
 tgt = 'mnist'
-model = 'drcn-s'
+model = 'drcn-'+ae_config['input']
 
 RESFILE = 'results/'+src+'-'+tgt+'_'+model+'_results_drop%.1f_aug%d_denoise%.1f.pkl.gz' % (net_config['dropout'], net_config['augmentation'], ae_config['denoising'])
 PARAMFILE = 'results/'+src+'-'+tgt+'_'+model+'_weights_drop%.1f_aug%d_denoise%.1f.pkl.gz' % (net_config['dropout'], net_config['augmentation'], ae_config['denoising'])
 PREDICTPREFIX = src+'-'+tgt+'_'+model+'_'
+print(PARAMFILE)
+
 
 # Load data
 if src == 'svhn':
 	(X_train, Y_train), (X_test, Y_test) = load_svhn()
 	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist32x32()
 elif src == 'mnist':
-	(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
-	(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
+	if tgt == 'svhn':
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
+	else:
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_usps()
+
+elif src == 'usps':
+	(X_train, Y_train), (X_test, Y_test) = load_usps()
+	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist()
 
 
 print('Preprocess data ...')

models.py

@@ -508,14 +508,23 @@ def train(self, X, Y,
 
 		total_batches_ae = X_tgt.shape[0] / self.ae_config['batch_size']
 
+		if self.ae_config['input'] == 's':
+			X_ae = np.copy(X)
+		elif self.ae_config['input'] == 't':
+			X_ae = np.copy(X_tgt)
+		elif self.ae_config['input'] == 'st':
+			X_ae = np.concatenate([X, X_tgt])
+
 
 		for epoch in range(self.net_config['nb_epoch']):
 
 			# ========== CONVAE =======
 			start_time = time.time()
 			loss_ae = 0
 			nbatch = 0
-			for X_batch, Y_batch in gdatagen.flow(X_tgt, np.copy(X_tgt), batch_size=self.ae_config['batch_size'], shuffle=self.ae_config['shuffle']):
+
+			# print('AE training : ',X_ae.shape)
+			for X_batch, Y_batch in gdatagen.flow(X_ae, np.copy(X_ae), batch_size=self.ae_config['batch_size'], shuffle=self.ae_config['shuffle']):
 				if self.ae_config['denoising'] > 0.:
 					X_batch = get_corrupted_output(X_batch, corruption_level=self.ae_config['denoising']).astype('float32')
 				else:
@@ -601,7 +610,7 @@ def train(self, X, Y,
 			if RESFILE is not None:
 				pickle.dump(self.res, gzip.open(RESFILE,'wb'))
 
-			if epoch % 5 == 0:
+			if epoch % 10 == 0:
 				print('=== > Save weights !')
 				self.save_weights(PARAMFILE)
 		# end epoch

test_drcn.py

@@ -10,24 +10,34 @@
 
 src = 'svhn'
 tgt = 'mnist'
-
-RESFILE = 'results/'+src+'-'+tgt+'_drcn_results_drop0.5_aug0_denoise0.4.pkl.gz'
-PARAMFILE = 'results/'+src+'-'+tgt+'_drcn_weights_drop0.5_aug0_denoise0.4.pkl.gz'
+model = 'drcn-st'
+RESFILE = 'results/'+src+'-'+tgt+'_'+model+'_results_drop0.5_aug1_denoise0.5.pkl.gz'
+PARAMFILE = 'results/'+src+'-'+tgt+'_'+model+'_weights_drop0.5_aug1_denoise0.5.pkl.gz'
 
 
 print('Load data...')
 inds = pickle.load(gzip.open('data_indices100.pkl.gz','rb'))
 
+
 if src == 'svhn':
 	(X_train, Y_train), (X_test, Y_test) = load_svhn()
 	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist32x32()
 	idx_src = inds['svhn_train']
 	idx_tgt = inds['mnist_test']
+
 elif src == 'mnist':
-	(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
-	(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
-	idx_src = inds['mnist_train']
-	idx_tgt = inds['svhn_test']
+	if tgt == 'svhn':
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist32x32()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_svhn()
+		idx_src = inds['mnist_train']
+		idx_tgt = inds['svhn_test']
+	else:
+		(X_train, Y_train), (_, _), (X_test, Y_test) = load_mnist()
+		(_, _), (X_tgt_test, Y_tgt_test) = load_usps()
+
+elif src == 'usps':
+	(X_train, Y_train), (X_test, Y_test) = load_usps()
+	(_, _), (_, _), (X_tgt_test, Y_tgt_test) = load_mnist()
 
 print('Preprocess data ...')
 # X_train, scaler = remove_mean(X_train)
@@ -46,7 +56,7 @@
 model.load_weights(PARAMFILE)
 
 
-show_filter(X_train[idx_src], grayscale=True, filename='viz/'+src+'_'+tgt+'_drcn-fixed_X100-src-orig_denoise0.4.png')
-show_filter(model.convae_.predict(X_train[idx_src]), grayscale=True, filename='viz/'+src+'_'+tgt+'_drcn-fixed_X100-src-pred_denoise0.4.png')
+show_filter(X_train[idx_src], grayscale=True, filename='viz/'+src+'_'+tgt+'_'+model+'_X100-src-orig.png')
+show_filter(model.convae_.predict(X_train[idx_src]), grayscale=True, filename='viz/'+src+'_'+tgt+'_'+model+'_X100-src-pred.png')
 # show_filter(X_tgt_test[idx_tgt], grayscale=True, filename='viz/'+src+'_'+tgt+'_drcn-fixed_X100-tgt-orig.png')
 # show_filter(model.convae_.predict(X_tgt_test[idx_tgt]), grayscale=True, filename='viz/'+src+'_'+tgt+'_drcn-fixed_X100-tgt-pred.png')