configure('pc_runs/' + NAME + '_Reg' , flush_secs = 2)
## define training parameters
PRINT_EVERY = 1
ADJUST_EVERY = 1000
START = time.time()
#best_val_loss = 1.0
best_val_loss = .05
best_valrsq = .20
best_epoch = 0
# train and validate
try:
print("Training for %d epochs..." % NUM_EPOCHS)
for epoch in range(1, NUM_EPOCHS + 1):
# perform training and validation
train_loss, train_r_sq, train_accu, train_accu2, val_loss, val_r_sq, val_accu, val_accu2 = train_utils.train_and_validate(perf_model, criterion, perf_optimizer, aug_training_data, aug_validation_data, METRIC, MTYPE, CTYPE)
# adjut learning rate
# train_utils.adjust_learning_rate(perf_optimizer, epoch, ADJUST_EVERY)
# log data for visualization later
####
log_value('train_loss', train_loss, epoch)
log_value('val_loss', val_loss, epoch)
log_value('train_r_sq', train_r_sq, epoch)
log_value('val_r_sq', val_r_sq, epoch)
log_value('train_accu', train_accu, epoch)
log_value('val_accu', val_accu, epoch)
log_value('train_accu2', train_accu2, epoch)
log_value('val_accu2', val_accu2, epoch)
#####
## define training parameters
PRINT_EVERY = 1
ADJUST_EVERY = 1000
START = time.time()
best_val_loss = 1.0
# train and validate
try:
print("Training for %d epochs..." % NUM_EPOCHS)
log_parameters = train_utils.log_init()
for epoch in range(1, NUM_EPOCHS + 1):
# perform training and validation
train_loss, train_r_sq, train_accu, val_loss, val_r_sq, val_accu = train_utils.train_and_validate(
perf_model,
criterion,
perf_optimizer,
training_data,
validation_data,
METRIC
)
# adjut learning rate
train_utils.adjust_learning_rate(perf_optimizer, epoch, ADJUST_EVERY)
# log data for visualization later
log_parameters = train_utils.log_epoch_stats(
log_parameters,
epoch,
train_loss,
train_r_sq,
train_accu,
val_loss,
## define training parameters
PRINT_EVERY = 1
ADJUST_EVERY = 1000
START = time.time()
#best_val_loss = 1.0
best_loss_contrastive_val = float('inf')
best_valrsq = .20
best_epoch = 0
best_ce_loss_val = float('inf')
# train and validate
try:
print("Training Encoder for %d epochs..." % NUM_EPOCHS)
if not Skip_encoder:
for epoch in range(1, NUM_EPOCHS + 1):
# perform training and validation
train_loss, train_r_sq, train_accu, train_accu2, val_loss, val_r_sq, val_accu, val_accu2 = train_utils.train_and_validate(perf_model, criterion, perf_optimizer, aug_training_data, aug_validation_data, METRIC, MTYPE, CTYPE, contrastive=criterion_contrastive, strength=(MSE_LOSS_STR, CONTR_LOSS_STR))
if contrastive:
print('Evaluating')
#print(eval_utils.eval_acc_contrastive(perf_model, criterion_contrastive, vef, METRIC, MTYPE, CTYPE, criterion_CE=criterion))
loss_contrastive_val, acc_contrastive_val, ce_loss_val = eval_utils.eval_acc_contrastive(perf_model, criterion_contrastive, vef, METRIC, MTYPE, CTYPE)
loss_contrastive_train, acc_contrastive_train, ce_loss_train = eval_utils.eval_acc_contrastive(perf_model, criterion_contrastive, aug_training_data, METRIC, MTYPE, CTYPE)
# adjut learning rate
# train_utils.adjust_learning_rate(perf_optimizer, epoch, ADJUST_EVERY)
# log data for visualization later
'''
else:
####
log_value('train_loss', train_loss, epoch)
log_value('val_loss', val_loss, epoch)
log_value('train_r_sq', train_r_sq, epoch)
log_value('val_r_sq', val_r_sq, epoch)