#import system things
from tensorflow.examples.tutorials.mnist import input_data # for data
import tensorflow as tf
import numpy as np
import os
#import helpers
import inference
import visualize
# prepare data and tf.session
mnist = input_data.read_data_sets('MNIST_data', one_hot=False)
sess = tf.InteractiveSession()
# setup siamese network
siamese = inference.siamese();
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(siamese.loss)
saver = tf.train.Saver()
tf.initialize_all_variables().run()
# if you just want to load a previously trainmodel?
load = False
model_ckpt = './model.meta'
if os.path.isfile(model_ckpt):
input_var = None
while input_var not in ['yes', 'no']:
input_var = input("We found model files. Do you want to load it and continue training [yes/no]?")
if input_var == 'yes':
load = True
# start training
#net_type = 'metricAuto' ## metricOnly: only use metric learning, metricAuto: using autoencoder with metric learning
net_type = 'holistic'
# net_type = 'metricOnly'
batch_train_size = 30
batch_test_size = 300
test_interval = 500
display_interval = 5000
learning_rate = 0.1
momentum = 0.1
# #resulTxt = net_type + '_noTanh.txt'
# resulTxt = net_type + 'Appeend.txt'
resulTxt = net_type + '.txt'
logid = open(resulTxt, 'w')
runOption = 1
##Construct Network ###############
siamese = inference.siamese(net_type=net_type)
#model_path = './models/' + net_type + '_noTanh_bal0001'# The reconstruction loss without tanh for output, the weight for reconstruction loss is 0.001
model_path = os.path.join('./models', net_type, 'weight')
weights_file = './models/metricOnly-195000'
#train_step = tf.train.GradientDescentOptimizer(0.01).minimize(siamese.loss)
if net_type == 'metricOnly':
train_step = tf.train.MomentumOptimizer(learning_rate=learning_rate,
momentum=momentum).minimize(
siamese.loss)
train_step_2 = tf.train.MomentumOptimizer(learning_rate=learning_rate,
momentum=momentum).minimize(
siamese.loss_2)
train_step_1 = tf.train.MomentumOptimizer(learning_rate=learning_rate,
momentum=momentum).minimize(
siamese.loss_1)
elif net_type == 'holistic':
#mnist = tf.keras.datasets.mnist
#(train_images, train_labels),(test_images, test_labels) = mnist.load_data()
mnist = input_data.read_data_sets('MNIST_data', one_hot=False)
#create session
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.InteractiveSession()
# setup siamese network
##################################################################################
lossType = 'inverseWassersteinCL'
margin = 5.0
siamese = inference.siamese(sess=sess,
margin=margin,
batch_size=batch_size,
lamb=0.01,
lossType=lossType)
#normLabel = 1
learning_rate = tf.placeholder(tf.float32, [1])
train_step = tf.train.MomentumOptimizer(learning_rate=learning_rate[0],
momentum=momentum).minimize(
siamese.loss,
var_list=siamese.var_list) #
##############################################################################################################
saver = tf.train.Saver() # save results ?
init = tf.global_variables_initializer() # initialization
sess.run(init)
#writer = tf.summary.FileWriter('./sinkhorn_board', sess.graph)
logid.write("Training results for {:10}".format(args.net_type))
logid.close()
logid = open(args.logfile, 'a')
checkpoint_dir = os.path.join('../models', args.net_type)
if not os.path.isdir(checkpoint_dir):
os.makedirs(checkpoint_dir)
modelName = 'model'
ckpt_status = get_checkpoint(checkpoint_dir)
##################################################################################
siamese = inference.siamese(
net_type=args.net_type,
margin_out=args.margin_out,
margin_hid=args.margin_hid,
alpha=args.alpha) # net_type: chosing metric only or with holistic loss
# siamese = inference_old.siamese()
train_step = tf.train.MomentumOptimizer(learning_rate=args.learning_rate,
momentum=args.momentum).minimize(
siamese.loss)
# train_step_2 = tf.train.MomentumOptimizer(learning_rate=learning_rate, momentum=momentum).minimize(siamese.loss_2)
# train_step_1 = tf.train.MomentumOptimizer(learning_rate=learning_rate, momentum=momentum).minimize(siamese.loss_1)
##############################################################################################################
saver = tf.train.Saver(max_to_keep=10000)
init = tf.global_variables_initializer()
with tf.Session() as sess:
print('Initializing all variables')
