####################
# mnist_path = 'mnist/mnist_28.pkl.gz'
# #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) # num_lab的数值来调控label与unlabel的比例
#
# x_lab, y_lab = x_l.T, y_l.T
# # tmp0 = x_lab[0].reshape(28, 28)
# # tmp1 = x_lab[10].reshape(28, 28)
# # tmp2 = x_lab[20].reshape(28, 28)
# 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
x_lab, y_lab, x_ulab, y_ulab, x_valid, y_valid, x_test, y_test = load_dataset_split(
"/home/yuan/Code/PycharmProjects/vae/ddi/train_dataset",
int(num_lab / 2))
################
''' Load VAE '''
################
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/M2/models/VAE_600-600-0.0003-50.cpkt' # github最开始的版本
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/checkpoints/model_VAE_0.0003-26_1526306640.2848828.cpkt' # 药物数据
#VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/M2/checkpoints/model_VAE_0.0003-50_1526353473.6963947.cpkt' #本地训练的mnist数据
VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/M2/checkpoints/model_VAE_0.0003-288_1526823669.3839986.cpkt' #本地训练的mnist数据
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, x_lab.shape[1], min_std)
def main_process(num_lab, num_batches, learning_rate, epochs, print_times=100):
classify_dataset_path = "/data/cdy/ykq/ddi_dataset/test_dataset"
# classify_dataset_path = "/home/yuan/Code/PycharmProjects/vae/ddi/test_dataset"
# train_data, train_labels, _, __, test_data, test_labels = load_dataset(classify_dataset_path, 3)
train_data, train_labels, _u, __u, _, __, test_data, test_labels = load_dataset_split(classify_dataset_path, num_lab//2, 3)
dim_x = train_data.shape[1]
dim_y = 2
train_size = train_data.shape[0]
test_size = test_data.shape[0]
train_data = np.reshape(train_data, [train_size, dim_x, 1, 1])
test_data = np.reshape(test_data, [test_size, dim_x, 1, 1])
assert train_size % num_batches == 0, '#TrainSize % #Batches != 0'
assert test_size % num_batches == 0, '#TestSize % #Batches != 0'
train_batch_size = int(train_size / num_batches)
test_batch_size = int(test_size / num_batches)
data_placeholder = tf.placeholder(tf.float32, [None, dim_x, 1, 1])
labels_placeholder = tf.placeholder(tf.float32, [None, dim_y])
result = lenet5(data_placeholder, labels_placeholder, dim_y)
accuracy = result.softmax.evaluate_classifier(labels_placeholder, phase=pt.Phase.test)
optimizer = tf.train.GradientDescentOptimizer(learning_rate)
train_op = pt.apply_optimizer(optimizer, losses=[result.loss])
#save_path = '/data/cdy/ykq/checkpoints/model_conv2d_{}-{}.cpkt'.format(
# learning_rate, time.strftime("%m-%d-%H%M%S", time.localtime()))
#print("model has been saved: " + save_path)
#runner = pt.train.Runner(save_path)
runner = pt.train.Runner()
best_accuracy = 0
best_epoch = 0
with tf.Session() as sess:
# print(epochs)
for epoch in range(epochs):
train_data, train_labels = permute_dataset((train_data, train_labels))
# 并没有保存最佳的model
runner.train_model(train_op, result.loss, num_batches,
feed_vars=(data_placeholder, labels_placeholder),
feed_data=pt.train.feed_numpy(train_batch_size, train_data, train_labels))
classification_accuracy = runner.evaluate_model(accuracy, num_batches,
feed_vars=(data_placeholder, labels_placeholder),
feed_data=pt.train.feed_numpy(test_batch_size, test_data, test_labels))
# runner.train_model(
# train_op,
# result.loss,
# EPOCH_SIZE,
# feed_vars=(image_placeholder, labels_placeholder),
# feed_data=pt.train.feed_numpy(BATCH_SIZE, train_images, train_labels),
# print_every=100)
# classification_accuracy = runner.evaluate_model(
# accuracy,
# TEST_SIZE,
# feed_vars=(image_placeholder, labels_placeholder),
# feed_data=pt.train.feed_numpy(BATCH_SIZE, test_images, test_labels))
if best_accuracy < classification_accuracy[0]:
best_accuracy = classification_accuracy
best_epoch = epoch
print('Accuracy after %d epoch %g%%' % (epoch + 1, classification_accuracy[0]*100))
print('Train size is {}.Best accuracy is {}%% at {} epoch.'.format(num_lab, best_accuracy, best_epoch))
print('==================================')
return best_accuracy, best_epoch, accuracy
####################
# mnist_path = 'mnist/mnist_28.pkl.gz'
# #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) # num_lab的数值来调控label与unlabel的比例
#
# x_lab, y_lab = x_l.T, y_l.T
# # tmp0 = x_lab[0].reshape(28, 28)
# # tmp1 = x_lab[10].reshape(28, 28)
# # tmp2 = x_lab[20].reshape(28, 28)
# 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
test_dataset_path = "/home/cdy/ykq/vae/ddi/test_dataset"
x_lab, y_lab, x_ulab, y_ulab, x_valid, y_valid, x_test, y_test = load_dataset_split(test_dataset_path, int(num_lab/2), 3)
print("test_dataset_path: ", test_dataset_path)
print("num_lab: ", num_lab)
################
''' Load VAE '''
################
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/M2/models/VAE_600-600-0.0003-50.cpkt'
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/checkpoints/model_VAE_0.0003-26_1526306640.2848828.cpkt'
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/M2/checkpoints/model_VAE_0.0003-50_1526353473.6963947.cpkt'
VAE_model_path = '/home/cdy/ykq/vae/M2/checkpoints/model_VAE_0.0003-260_1526433033.8844438.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, x_lab.shape[1], encode_hidden_layers_px, encode_hidden_layers_qz, min_std )
def ddi_classifier(VAE_model_path, min_std, num_lab, dim_z,
M1_hidden_layers_px, M1_hidden_layers_qz,
M2_hidden_layers_px, M2_hidden_layers_qz,
M2_hidden_layers_qy):
#############################
''' Experiment Parameters '''
#############################
# if len(sys.argv) > 1:
# print(sys.argv[1])
# num_lab = int("Number of total labelled examples: ", sys.argv[1])
# else:
# num_lab = 12000 #Number of labelled examples (total)
num_batches = 100 #Number of minibatches in a single epoch
# dim_z = 50 #Dimensionality of latent variable (z)
epochs = 1001 #Number of epochs through the full dataset
learning_rate = 3e-4 #Learning rate of ADAM
alpha = 0.1 #Discriminatory factor (see equation (9) of http://arxiv.org/pdf/1406.5298v2.pdf)
seed = 31415 #Seed for RNG
#Neural Networks parameterising p(x|z,y), q(z|x,y) and q(y|x)
# M1_hidden_layers_px = [600, 1000, 800, 500]
# M1_hidden_layers_qz = [600, 1000, 800, 500]
# M2_hidden_layers_px = [ 500 ]
# M2_hidden_layers_qz = [ 500 ]
# M2_hidden_layers_qy = [ 500 ]
print("ddi_classifier: M2_hidden_layers_px: ", M2_hidden_layers_px)
print("ddi_classifier: M2_hidden_layers_qz: ", M2_hidden_layers_qz)
print("ddi_classifier: M2_hidden_layers_qy: ", M2_hidden_layers_qy)
####################
''' Load Dataset '''
####################
test_dataset_path = "/home/cdy/ykq/vae/ddi/test_dataset"
x_lab, y_lab, x_ulab, y_ulab, x_valid, y_valid, x_test, y_test = load_dataset_split(
test_dataset_path, int(num_lab / 2), 3)
print("test_dataset_path: ", test_dataset_path)
print("num_lab: ", num_lab)
################
''' Load VAE '''
################
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/M2/models/VAE_600-600-0.0003-50.cpkt'
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/checkpoints/model_VAE_0.0003-26_1526306640.2848828.cpkt'
# VAE_model_path = '/home/yuan/Code/PycharmProjects/vae/M2/checkpoints/model_VAE_0.0003-50_1526353473.6963947.cpkt'
# VAE_model_path = '/home/cdy/ykq/vae/M2/checkpoints/model_VAE_0.0003-288_1526980680.8521073.cpkt'
# min_std = 0.1 #Dimensions with std < min_std are removed before training with GC
#################
'''Load VAE-M1'''
#################
def encode_dataset(
model_path,
facter_dim,
dim_z,
min_std=0.0,
):
# VAE = VariationalAutoencoder( dim_x = 28*28, dim_z = 50 ) #Should be consistent with model being loaded
VAE = VariationalAutoencoder(
dim_x=facter_dim,
dim_z=dim_z,
hidden_layers_px=M1_hidden_layers_px,
hidden_layers_qz=M1_hidden_layers_qz
) # Should be consistent with model being loaded
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
data_lab, data_ulab, data_valid, data_test = encode_dataset(
VAE_model_path, x_lab.shape[1], dim_z, 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]
print("dim_x", dim_x)
print("dim_y", dim_y)
###################################
''' Train Generative Classifier '''
###################################
GC = DdiGenerativeClassifier(dim_x,
dim_z,
dim_y,
num_examples,
num_lab,
num_batches,
hidden_layers_px=M2_hidden_layers_px,
hidden_layers_qz=M2_hidden_layers_qz,
hidden_layers_qy=M2_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 = DdiGenerativeClassifier(
dim_x,
dim_z,
dim_y,
num_examples,
num_lab,
num_batches,
hidden_layers_px=M2_hidden_layers_px,
hidden_layers_qz=M2_hidden_layers_qz,
hidden_layers_qy=M2_hidden_layers_qy,
)
with GC_eval.session:
GC_eval.saver.restore(GC_eval.session, GC.save_path)
GC_eval.predict_labels(data_test, y_test)