if model_type == 'DLSTM3':
model = models.DLSTM3(feature_size, hidden_size)
model.load_state_dict(checkpoint['state_dict'])
else:
raise ValueError("Model type not recognized")
else:
if model_type == 'DLSTM3':
model = models.DLSTM3(feature_size, hidden_size)
elif model_type == 'SingleLSTM':
model = models.SingleLSTM(feature_size, hidden_size)
else:
raise ValueError("Model type not recognized")
model = model.to(device)
print("This model has {} trainable parameters".format(
count_trainable_params(model)))
###############################################################################
# Training code
###############################################################################
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
try:
optimizer.load_state_dict(checkpoint['optimizer'])
except NameError:
print("Optimizer initializing")
criterion = nn.NLLLoss().to(device)
warm_up_text = open(args.train, encoding='utf-8').read()[0:args.bptt]
'''
Training Loop
Can interrupt with Ctrl + C
'''
with tf.Session(config=sess_config) as sess:
tf.global_variables_initializer().run()
model_name = 'split1_nonlocal'
# load_fn = slim.assign_from_checkpoint_fn(os.path.join(
# './logs/results/model', 'triple_anet'+'.model-60'),tf.global_variables(),ignore_missing_vars=True)
# load_fn(sess)
# print(model_name+' have been loaded')
mkdir_if_missing('./logs/' + model_name + '/')
mkdir_if_missing('./logs/' + model_name + '/saliency/')
mkdir_if_missing('./logs/' + model_name + '/model/')
saver = tf.train.Saver(tf.global_variables())
parameters = utils.count_trainable_params()
print("Total training params: %.1fM \r\n" % (parameters / 1e6))
infile = open('./logs/' + model_name + '/log_split1.txt', 'w')
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
try:
GGamma = {}
start_time = time.time()
learning_rate = 0.01
# learning_rate = 0.0
batch_count = np.int32(9688 / batch_size)
# batch_count = np.int32(600 / batch_size)
counter = 0
train_images, train_label = sess.run(
[train_image_batch, train_label_batch])
model = models.DLSTM3(feature_size, hidden_size)
model.load_state_dict(checkpoint['state_dict'])
else:
raise ValueError("Model type not recognized")
else:
if model_type == 'DLSTM3':
model = models.DLSTM3(feature_size, hidden_size)
elif model_type == 'SingleLSTM':
model = models.SingleLSTM(feature_size, hidden_size)
else:
raise ValueError("Model type not recognized")
model = model.to(device)
### Size of model
print ("This model has {} trainable parameters".format(count_trainable_params(model)))
###############################################################################
# Training code
###############################################################################
### Optimizer
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
try:
optimizer.load_state_dict(checkpoint['optimizer'])
except NameError:
print("Optimizer initializing")
### Loss function
criterion = nn.NLLLoss().to(device)