etm_model = etm_model.to(device)
etm_model.eval()
print('Visualizing model quality after training...')
etm_model.eval()
etm_model.print_words_per_topic(
words_amount=config_dict['evaluation_params']['num_words'],
vocab=vocab,
lang='en')
with torch.no_grad():
# get document completion perplexities
test_ppl = etm_model.evaluate(
source='test',
test_1_tokens=test_1_tokens,
test_1_counts=test_1_counts,
test_2_tokens=test_2_tokens,
test_2_counts=test_2_counts,
train_tokens=train_tokens,
vocab=vocab,
tc=config_dict['evaluation_params']['tc'],
td=config_dict['evaluation_params']['td'])
# get most used topics
indices = torch.tensor(range(config_dict['num_docs_train']))
indices = torch.split(indices, config_dict['batch_size'])
thetaAvg = torch.zeros(
1, config_dict['model_params']['num_topics']).to(device)
thetaWeightedAvg = torch.zeros(
1, config_dict['model_params']['num_topics']).to(device)
cnt = 0
for idx, ind in enumerate(indices):
data_batch = data.get_batch(train_tokens, train_counts, ind,
tracemalloc.start()
if args.mode == 'train':
## train model on data
best_epoch = 0
best_val_ppl = 1e9
all_val_ppls = []
print('\n')
print('Visualizing model quality before training...', args.epochs)
#model.visualize(args, vocabulary = vocab)
print('\n')
for epoch in range(0, args.epochs):
print("I am training for epoch", epoch)
model.train_for_epoch(epoch, args, training_set)
val_ppl = model.evaluate(args, 'val', training_set, vocab, test_1, test_2)
print("The validation scores", val_ppl)
if val_ppl < best_val_ppl:
with open(ckpt, 'wb') as f:
torch.save(model, f)
best_epoch = epoch
best_val_ppl = val_ppl
else:
## check whether to anneal lr
lr = optimizer.param_groups[0]['lr']
if args.anneal_lr and (len(all_val_ppls) > args.nonmono and val_ppl > min(all_val_ppls[:-args.nonmono]) and lr > 1e-5):
optimizer.param_groups[0]['lr'] /= args.lr_factor
if epoch % args.visualize_every == 0:
model.visualize(args, vocabulary = vocab)
all_val_ppls.append(val_ppl)
with open(ckpt, 'rb') as f: