def load_dataset(dataset_name, cv):
opt = fake_dataset_opt(dataset_name, cv)
# print(dataset_name)
# print(opt.__dict__)
# input()
dataset, val_dataset, tst_dataset = create_trn_val_tst_dataset(opt) # create a dataset given opt.dataset_mode and other options
return dataset, val_dataset, tst_dataset
def test(opt):
trn_dataset, val_dataset, tst_dataset = create_trn_val_tst_dataset(opt)
logger.info('The number of validation samples = %d' % len(val_dataset))
logger.info('The number of testing samples = %d' % len(tst_dataset))
model = create_model(
opt) # create a model given opt.model and other options
model.setup(opt)
model.cuda()
# test
logger.info('Loading model : epoch-%d' % opt.eval_epoch)
model.load_networks(opt.eval_epoch)
logger.info('Finish loading model')
acc = eval(model, val_dataset, tst_dataset, is_save=False, phase='val')
logger.info('Val result acc %.4f' % (acc))
acc = eval(model, val_dataset, tst_dataset, is_save=False, phase='test')
logger.info('Tst result acc %.4f' % (acc))
def train(opt):
trn_dataset, val_dataset, tst_dataset = create_trn_val_tst_dataset(opt)
dataset_size = len(trn_dataset)
logger.info('The number of training samples = %d' % dataset_size)
writer = SummaryWriter()
model = create_model(
opt) # create a model given opt.model and other options
model.setup(opt)
model.cuda()
best_eval_acc = 0 # record the best eval UAR
total_iters = 0 # the total number of training iterations
best_eval_epoch = -1 # record the best eval epoch
for epoch in range(
1, opt.niter + opt.niter_decay + 1
): # outer loop for different epochs; we save the model by <epoch_count>, <epoch_count>+<save_latest_freq>
epoch_start_time = time.time() # timer for entire epoch
iter_data_time = time.time() # timer for data loading per iteration
epoch_iter = 0 # the number of training iterations in current epoch, reset to 0 every epoch
for i, data in enumerate(trn_dataset): # inner loop within one epoch
iter_start_time = time.time(
) # timer for computation per iteration
total_iters += 1 # opt.batch_size
epoch_iter += opt.batch_size
model.set_input(
data) # unpack data from dataset and apply preprocessing
model.optimize_parameters(
epoch
) # calculate loss functions, get gradients, update network weights
if total_iters % opt.print_freq == 0: # print training losses and save logging information to the disk
losses = model.get_current_losses()
t_comp = (time.time() - iter_start_time) / opt.batch_size
logger.info('Cur epoch {}'.format(epoch) + ' loss ' + ' '.join(
map(lambda x: '{}:{{{}:.4f}}'.format(x, x),
model.loss_names)).format(**losses))
writer.add_scalars('training_loss', dict(losses), total_iters)
if total_iters % opt.save_latest_freq == 0: # cache our latest model every <save_latest_freq> iterations
logger.info(
'saving the latest model (epoch %d, total_iters %d)' %
(epoch, total_iters))
save_suffix = 'iter_%d' % total_iters if opt.save_by_iter else 'latest'
model.save_networks(save_suffix)
iter_data_time = time.time()
if epoch % opt.save_epoch_freq == 0: # cache our model every <save_epoch_freq> epochs
logger.info('saving the model at the end of epoch %d, iters %d' %
(epoch, total_iters))
model.save_networks('latest')
model.save_networks(epoch)
logger.info('End of training epoch %d / %d \t Time Taken: %d sec' %
(epoch, opt.niter + opt.niter_decay,
time.time() - epoch_start_time))
model.update_learning_rate(
logger) # update learning rates at the end of every epoch.
acc = eval(model, val_dataset, tst_dataset)
logger.info('Val result of epoch %d / %d acc %.4f ' %
(epoch, opt.niter + opt.niter_decay, acc))
if acc > best_eval_acc:
best_eval_epoch = epoch
best_eval_acc = acc
writer.close()
# print best eval result
logger.info('Best eval epoch %d found with acc %f' %
(best_eval_epoch, best_eval_acc))
# test
logger.info('Loading best model found on val set: epoch-%d' %
best_eval_epoch)
model.load_networks(best_eval_epoch)
acc = eval(model, val_dataset, tst_dataset, is_save=True, phase='test')
logger.info('Tst result acc %.4f' % (acc))
clean_chekpoints(opt.name, best_eval_epoch)
if __name__ == '__main__':
teacher_path = 'checkpoints/ef_AVL_Adnn512,256,128_Vlstm128_maxpool_Lcnn128_fusion256,128run1/'
# teacher_path = 'checkpoints/new_0416_simpleAE_ce_t1.0_f1.0_mse0.1_cycle0.1_run1'
opt_path = os.path.join(teacher_path, 'train_opt.conf')
opt = load_from_opt_record(opt_path)
opt.isTrain = False # teacher model should be in test mode
opt.gpu_ids = [0]
opt.serial_batches = True
opt.dataset_mode = 'iemocap_miss'
setattr(opt, 'miss_num', 'mix')
modality = 'L'
for cv in range(1, 11):
opt.cvNo = cv
teacher_path_cv = os.path.join(teacher_path, str(cv))
dataset, val_dataset, tst_dataset = create_trn_val_tst_dataset(
opt) # create a dataset given opt.dataset_mode and other options
# model = MultiFusionMultiModel(opt)
# model = NewTranslationModel(opt)
model = EarlyFusionMultiModel(opt)
model.cuda()
model.load_networks_cv(teacher_path_cv)
# extractor = MultiLayerFeatureExtractor(model, 'netC.module[4]')
# save_root = 'analysis/teacher_feats/{}/'.format(modality) + str(cv)
# if not os.path.exists(save_root):
# os.makedirs(save_root)
# extract(model, dataset, save_root, phase='trn', modality=modality)
# extract(model, val_dataset, save_root, phase='val', modality=modality)
# extract(model, tst_dataset, save_root, phase='tst', modality=modality)
save_root = '/data2/lrc/Iemocap_feature/early_fusion_reps_mix'