def init():
print("\tInitialising sentences")
print("\t\tLoading and cleaning json files")
json_of_convs = load_all_json_conv('./Dataset/messages')
print("\t\tLoading two person convs")
duo_conversations = get_chat_friend_and_me(json_of_convs)
print("\t\tMaking two person convs discussions")
discussions = get_discussions(duo_conversations)
print("\t\tCreating pairs for training")
pairs_of_sentences = make_pairs(discussions)
print(f"\t\t{len(pairs_of_sentences)} different pairs")
print("\t\tCreating Vocabulary")
voc = Voc()
print("\t\tPopulating Vocabulary")
voc.createVocFromPairs(pairs_of_sentences)
print(f"\t\tVocabulary of : {voc.num_words} differents words")
print('\tBuilding encoder and decoder ...')
embedding = nn.Embedding(voc.num_words, HIDDEN_SIZE)
encoder = EncoderRNN(HIDDEN_SIZE, embedding, ENCODER_N_LAYERS, DROPOUT)
decoder = LuongAttnDecoderRNN(ATTN_MODEL, embedding, HIDDEN_SIZE,
voc.num_words, DECODER_N_LAYERS, DROPOUT)
encoder_optimizer = optim.Adam(encoder.parameters(), lr=LEARNING_RATE)
decoder_optimizer = optim.Adam(decoder.parameters(),
lr=LEARNING_RATE * DECODER_LEARNING_RATIO)
checkpoint = None
if LOADFILENAME:
print("\t\tLoading last training")
checkpoint = torch.load(LOADFILENAME)
# If loading a model trained on GPU to CPU
# checkpoint=torch.load(loadFilename,map_location=torch.device('cpu'))
encoder_sd = checkpoint['en']
decoder_sd = checkpoint['de']
encoder_optimizer_sd = checkpoint['en_opt']
decoder_optimizer_sd = checkpoint['de_opt']
embedding_sd = checkpoint['embedding']
voc.__dict__ = checkpoint['voc_dict']
print("\t\tPopulating from last training")
embedding.load_state_dict(embedding_sd)
encoder.load_state_dict(encoder_sd)
decoder.load_state_dict(decoder_sd)
encoder_optimizer.load_state_dict(encoder_optimizer_sd)
decoder_optimizer.load_state_dict(decoder_optimizer_sd)
encoder = encoder.to(DEVICE)
decoder = decoder.to(DEVICE)
return (encoder, decoder, encoder_optimizer, decoder_optimizer, embedding,
voc, pairs_of_sentences, checkpoint)
def main():
input_lang = Lang('data/WORDMAP_en.json')
output_lang = Lang('data/WORDMAP_zh.json')
print("input_lang.n_words: " + str(input_lang.n_words))
print("output_lang.n_words: " + str(output_lang.n_words))
train_data = TranslationDataset('train')
val_data = TranslationDataset('valid')
# Initialize encoder & decoder models
encoder = EncoderRNN(input_lang.n_words, hidden_size, encoder_n_layers,
dropout)
decoder = LuongAttnDecoderRNN(attn_model, hidden_size, output_lang.n_words,
decoder_n_layers, dropout)
# Use appropriate device
encoder = encoder.to(device)
decoder = decoder.to(device)
# Initialize optimizers
print('Building optimizers ...')
encoder_optimizer = optim.Adam(encoder.parameters(), lr=learning_rate)
decoder_optimizer = optim.Adam(decoder.parameters(), lr=learning_rate)
# Initializations
print('Initializing ...')
train_batch_time = ExpoAverageMeter() # forward prop. + back prop. time
train_losses = ExpoAverageMeter() # loss (per word decoded)
val_batch_time = ExpoAverageMeter()
val_losses = ExpoAverageMeter()
best_loss = 100000
epochs_since_improvement = 0
# Epochs
for epoch in range(start_epoch, epochs):
# Decay learning rate if there is no improvement for 8 consecutive epochs, and terminate training after 20
if epochs_since_improvement == 20:
break
if epochs_since_improvement > 0 and epochs_since_improvement % 8 == 0:
adjust_learning_rate(decoder_optimizer, 0.8)
adjust_learning_rate(encoder_optimizer, 0.8)
# One epoch's training
# Ensure dropout layers are in train mode
encoder.train()
decoder.train()
start = time.time()
# Batches
for i_batch in range(len(train_data)):
input_variable, lengths, target_variable, mask, max_target_len = train_data[
i_batch]
train_loss = train(input_variable, lengths, target_variable, mask,
max_target_len, encoder, decoder,
encoder_optimizer, decoder_optimizer)
# Keep track of metrics
train_losses.update(train_loss)
train_batch_time.update(time.time() - start)
start = time.time()
# Print status
if i_batch % print_every == 0:
print(
'[{0}] Epoch: [{1}][{2}/{3}]\t'
'Batch Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'.format(
timestamp(),
epoch,
i_batch,
len(train_data),
batch_time=train_batch_time,
loss=train_losses))
# One epoch's validation
start = time.time()
# Batches
for i_batch in range(len(val_data)):
input_variable, lengths, target_variable, mask, max_target_len = val_data[
i_batch]
val_loss = valid(input_variable, lengths, target_variable, mask,
max_target_len, encoder, decoder)
# Keep track of metrics
val_losses.update(val_loss)
val_batch_time.update(time.time() - start)
start = time.time()
# Print status
if i_batch % print_every == 0:
print(
'Validation: [{0}/{1}]\t'
'Batch Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'.format(
i_batch,
len(val_data),
batch_time=val_batch_time,
loss=val_losses))
val_loss = val_losses.avg
print('\n * LOSS - {loss:.3f}\n'.format(loss=val_loss))
# Check if there was an improvement
is_best = val_loss < best_loss
best_loss = min(best_loss, val_loss)
if not is_best:
epochs_since_improvement += 1
print("\nEpochs since last improvement: %d\n" %
(epochs_since_improvement, ))
else:
epochs_since_improvement = 0
save_checkpoint(epoch, encoder, decoder, encoder_optimizer,
decoder_optimizer, input_lang, output_lang, val_loss,
is_best)
# Initialize search module
searcher = GreedySearchDecoder(encoder, decoder)
for input_sentence, target_sentence in pick_n_valid_sentences(
input_lang, output_lang, 10):
decoded_words = evaluate(searcher, input_sentence, input_lang,
output_lang)
print('> {}'.format(input_sentence))
print('= {}'.format(target_sentence))
print('< {}'.format(''.join(decoded_words)))
# Reshuffle train and valid samples
np.random.shuffle(train_data.samples)
np.random.shuffle(val_data.samples)
def main():
train_loader = ChatbotDataset('train')
val_loader = ChatbotDataset('valid')
# Initialize word embeddings
embedding = nn.Embedding(voc.num_words, hidden_size)
# Initialize encoder & decoder models
encoder = EncoderRNN(hidden_size, embedding, encoder_n_layers, dropout)
decoder = LuongAttnDecoderRNN(attn_model, embedding, hidden_size, voc.num_words, decoder_n_layers, dropout)
# Use appropriate device
encoder = encoder.to(device)
decoder = decoder.to(device)
# Initialize optimizers
print('Building optimizers ...')
encoder_optimizer = optim.Adam(encoder.parameters(), lr=learning_rate)
decoder_optimizer = optim.Adam(decoder.parameters(), lr=learning_rate)
# Initializations
print('Initializing ...')
batch_time = AverageMeter() # forward prop. + back prop. time
losses = AverageMeter() # loss (per word decoded)
# Epochs
for epoch in range(start_epoch, epochs):
# One epoch's training
# Ensure dropout layers are in train mode
encoder.train()
decoder.train()
start = time.time()
# Batches
for i in range(train_loader.__len__()):
input_variable, lengths, target_variable, mask, max_target_len = train_loader.__getitem__(i)
loss = train(input_variable, lengths, target_variable, mask, max_target_len, encoder, decoder,
encoder_optimizer, decoder_optimizer)
# Keep track of metrics
losses.update(loss, max_target_len)
batch_time.update(time.time() - start)
start = time.time()
if i % print_every == 0:
print('[{0}] Epoch: [{1}][{2}/{3}]\t'
'Batch Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'.format(timestamp(), epoch, i, len(train_loader),
batch_time=batch_time,
loss=losses))
# One epoch's validation
val_loss = validate(val_loader, encoder, decoder)
print('\n * LOSS - {loss:.3f}\n'.format(loss=val_loss))
# Initialize search module
searcher = GreedySearchDecoder(encoder, decoder)
for sentence in pick_n_valid_sentences(10):
decoded_words = evaluate(searcher, sentence)
print('Human: {}'.format(sentence))
print('Bot: {}'.format(''.join(decoded_words)))
# Save checkpoint
if epoch % save_every == 0:
directory = save_dir
if not os.path.exists(directory):
os.makedirs(directory)
torch.save({
'epoch': epoch,
'en': encoder.state_dict(),
'de': decoder.state_dict(),
'en_opt': encoder_optimizer.state_dict(),
'de_opt': decoder_optimizer.state_dict(),
'loss': loss,
'voc': voc.__dict__
}, os.path.join(directory, '{}_{}_{}.tar'.format('checkpoint', epoch, val_loss)))
'embedding': embedding.state_dict()
},
os.path.join(directory,
'{}_{}.tar'.format(iteration, 'checkpoint')))
print('Building encoder and decoder ...')
# word embedding
embedding = nn.Embedding(VOC.num_words, hp.hidden_size)
encoder = EncoderRNN(hp.hidden_size, embedding, hp.n_layers, hp.dropout)
decoder = LuongAttnDecoderRNN(hp.attn_model, embedding, hp.hidden_size,
VOC.num_words, hp.n_layers, hp.dropout)
encoder = encoder.to(device)
decoder = decoder.to(device)
print('Models built and ready to go!')
encoder.train()
decoder.train()
print('Building optimizers ...')
encoder_optimizer = optim.Adam(encoder.parameters(), lr=hp.lr)
decoder_optimizer = optim.Adam(decoder.parameters(),
lr=hp.lr * hp.decoder_learning_ratio)
encoder_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
encoder_optimizer, 5)
decoder_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
decoder_optimizer, 5)
if loadFilename:
encoder_optimizer.load_state_dict(encoder_optimizer_sd)
def main():
corpus_name = "cornell movie-dialogs corpus"
corpus = os.path.join("data", corpus_name)
printLines(os.path.join(corpus, "movie_lines.txt"))
# Define path to new file
datafile = os.path.join(corpus, "formatted_movie_lines.txt")
linefile = os.path.join(corpus, "movie_lines.txt")
conversationfile = os.path.join(corpus, "movie_conversations.txt")
# Initialize lines dict, conversations list, and field ids
MOVIE_LINES_FIELDS = ["lineID", "characterID", "movieID", "character", "text"]
MOVIE_CONVERSATIONS_FIELDS = ["character1ID", "character2ID", "movieID", "utteranceIDs"]
# Load lines and process conversations
preprocess = Preprocess(datafile, linefile, conversationfile, MOVIE_LINES_FIELDS, MOVIE_CONVERSATIONS_FIELDS)
preprocess.loadLines()
preprocess.loadConversations()
preprocess.writeCSV()
# Load/Assemble voc and pairs
save_dir = os.path.join("data", "save")
dataset = Dataset(corpus, corpus_name, datafile)
voc, pairs = dataset.loadPrepareData()
# # Print some pairs to validate
# print("\npairs:")
# for pair in pairs[:10]:
# print(pair)
# Trim voc and pairs
pairs = dataset.trimRareWords(voc, pairs, MIN_COUNT)
# Example for validation
small_batch_size = 5
batches = dataset.batch2TrainData(voc, [random.choice(pairs) for _ in range(small_batch_size)])
input_variable, lengths, target_variable, mask, max_target_len = batches
print("input_variable:", input_variable)
print("lengths:", lengths)
print("target_variable:", target_variable)
print("mask:", mask)
print("max_target_len:", max_target_len)
# Configure models
model_name = 'cb_model'
attn_model = 'dot'
#attn_model = 'general'
#attn_model = 'concat'
hidden_size = 500
encoder_n_layers = 2
decoder_n_layers = 2
dropout = 0.1
batch_size = 64
# Set checkpoint to load from; set to None if starting from scratch
loadFilename = None
checkpoint_iter = 4000
#loadFilename = os.path.join(save_dir, model_name, corpus_name,
# '{}-{}_{}'.format(encoder_n_layers, decoder_n_layers, hidden_size),
# '{}_checkpoint.tar'.format(checkpoint_iter))
if loadFilename:
# If loading on same machine the model was trained on
checkpoint = torch.load(loadFilename)
# If loading a model trained on GPU to CPU
#checkpoint = torch.load(loadFilename, map_location=torch.device('cpu'))
encoder_sd = checkpoint['en']
decoder_sd = checkpoint['de']
encoder_optimizer_sd = checkpoint['en_opt']
decoder_optimizer_sd = checkpoint['de_opt']
embedding_sd = checkpoint['embedding']
voc.__dict__ = checkpoint['voc_dict']
print('Building encoder and decoder ...')
# Initialize word embeddings
embedding = nn.Embedding(voc.num_words, hidden_size)
if loadFilename:
embedding.load_state_dict(embedding_sd)
# Initialize encoder & decoder models
encoder = EncoderRNN(hidden_size, embedding, encoder_n_layers, dropout)
decoder = LuongAttnDecoderRNN(attn_model, embedding, hidden_size, voc.num_words, decoder_n_layers, dropout)
if loadFilename:
encoder.load_state_dict(encoder_sd)
decoder.load_state_dict(decoder_sd)
# Use appropriate device
encoder = encoder.to(device)
decoder = decoder.to(device)
print('Models built and ready to go!')
# Configure training/optimization
clip = 50.0
teacher_forcing_ratio = 1.0
learning_rate = 0.0001
decoder_learning_ratio = 5.0
n_iteration = 4000
print_every = 1
save_every = 500
# Ensure dropout layers are in train mode
encoder.train()
decoder.train()
# Initialize optimizers
print('Building optimizers ...')
encoder_optimizer = optim.Adam(encoder.parameters(), lr=learning_rate)
decoder_optimizer = optim.Adam(decoder.parameters(), lr=learning_rate * decoder_learning_ratio)
if loadFilename:
encoder_optimizer.load_state_dict(encoder_optimizer_sd)
decoder_optimizer.load_state_dict(decoder_optimizer_sd)
# Run training iterations
print("Starting Training!")
model = Model(dataset.batch2TrainData, teacher_forcing_ratio)
model.trainIters(model_name, voc, pairs, encoder, decoder, encoder_optimizer, decoder_optimizer,
embedding, encoder_n_layers, decoder_n_layers, save_dir, n_iteration, batch_size,
print_every, save_every, clip, corpus_name, loadFilename)
# Set dropout layers to eval mode
encoder.eval()
decoder.eval()
# Initialize search module
searcher = GreedySearchDecoder(encoder, decoder)