def evaluate(sentence, encoder, decoder, max_length = loader.MAX_LENGTH): with torch.no_grad(): input_tensor = train.tensorFromSentence(input_lang, sentence) input_length = input_tensor.size(0) encoder_hidden = (encoder.initHidden().to(device), encoder.initHidden().to(device)) encoder_outputs = torch.zeros(max_length, encoder.hidden_size).to(device) for index in range(input_length): encoder_output, encoder_hidden = encoder(input_tensor[index], encoder_hidden) encoder_outputs[index] = encoder_output[0, 0] decoder_input = torch.tensor([[loader.SOS_token]]).to(device) decoder_hidden = encoder_hidden decoded_words = [] for index in range(max_length): decoder_output, decoder_hidden = decoder(decoder_input, decoder_hidden) topv, topi = decoder_output.data.topk(1) if topi.item() == loader.EOS_token: break else: decoded_words.append(output_lang.index2word[topi.item()]) decoder_input = topi.squeeze().detach() return decoded_words
def evaluate(encoder, decoder, sentence, max_length):
with torch.no_grad():
word_tokenize = nlp.word_tokenize(sentence)
syntax_info = nlp.dependency_parse(sentence)
sentence = ' '.join(word_tokenize)
#print(syntax_info)
#print(len(word_tokenize))
tmp_dependency_tree = dependency_tree(sentence, len(word_tokenize),
syntax_info)
syntax_matrix = tmp_dependency_tree.get_syntax_matrix()
#print(syntax_matrix)
input_tensor = tensorFromSentence(input_lang, sentence)
input_length = input_tensor.size()[0]
encoder_hidden = encoder.initHidden()
encoder_outputs = torch.zeros(max_length,
encoder.hidden_size,
device=device)
for ei in range(input_length):
encoder_output, encoder_hidden = encoder(input_tensor[ei],
encoder_hidden)
encoder_outputs[ei] += encoder_output[0, 0]
decoder_input = torch.tensor([[SOS_token]], device=device) # SOS
decoder_hidden = encoder_hidden
h_t_tilde = decoder_hidden[0]
decoded_words = []
decoder_attentions = torch.zeros(max_length, max_length)
for di in range(max_length):
decoder_output, decoder_hidden, decoder_attention = decoder(
decoder_input, decoder_hidden, encoder_outputs,
len(syntax_matrix), syntax_matrix)
#decoder_output, decoder_hidden, h_t_tilde, decoder_attention = decoder(decoder_input, decoder_hidden, encoder_outputs, h_t_tilde, len(syntax_matrix), syntax_matrix)
decoder_attentions[di] = decoder_attention.data
topv, topi = decoder_output.data.topk(1)
if topi.item() == EOS_token:
decoded_words.append('<EOS>')
break
else:
decoded_words.append(output_lang.index2word[topi.item()])
decoder_input = topi.squeeze().detach()
return decoded_words, decoder_attentions[:di + 1]
def evaluate(sentence, encoder, decoder, max_length=loader.MAX_LENGTH):
with torch.no_grad():
input_tensor = train.tensorFromSentence(input_lang, sentence)
# |input_tensor| = (sentence_length, 1)
input_length = input_tensor.size(0)
encoder_hidden = (encoder.initHidden().to(device),
encoder.initHidden().to(device))
# |encoder_hidden|= (2, num_layers*num_directions, batch_size, hidden_size)
encoder_outputs = torch.zeros(max_length,
encoder.hidden_size).to(device)
# |encoder_outputs| = (max_length, hidden_size)
for ei in range(input_length):
encoder_output, encoder_hidden = encoder(input_tensor[ei],
encoder_hidden)
# |encoder_output| = (batch_size, sequence_length, num_directions*hidden_size))
# |encoder_hidden| = (2, num_layers*num_directions, batch_size, hidden_size)
# 2: respectively, hidden state and cell state.
encoder_outputs[ei] = encoder_output[0, 0]
decoder_input = torch.tensor([[loader.SOS_token]]).to(device)
# |decoder_input| = (1, 1)
decoder_hidden = encoder_hidden
# |decoder_hidden|= (2, num_layers*num_directions, batch_size, hidden_size)
# 2: respectively, hidden state and cell state
decoded_words = []
for di in range(max_length):
decoder_output, decoder_hidden = decoder(decoder_input,
decoder_hidden)
# |decoder_output| = (sequence_length, output_lang.n_words)
# |decoder_hidden| = (2, num_layers*num_directions, batch_size, hidden_size)
# 2: respectively, hidden state and cell state.
topv, topi = decoder_output.data.topk(1) # top-1 value, index
# |topv|, |topi| = (1, 1)
if topi.item() == loader.EOS_token:
# decoded_words.append('<EOS>')
break
else:
decoded_words.append(output_lang.index2word[topi.item()])
decoder_input = topi.squeeze().detach()
return decoded_words
def evaluate(context, classification, device, sentence, inputs_dict,
target_dict):
with torch.no_grad():
input_tensor = tensorFromSentence(inputs_dict, sentence, device)
input_length = input_tensor.size()[0]
context_hidden = context.init_hidden()
for ei in range(input_length):
context_hidden = context(input_tensor[ei], context_hidden)
classification_inputs = context_hidden[0]
result = classification(classification_inputs)
result = classFromTensor(result, target_dict)
return result
def evaluate(encoder, decoder, sentence, max_length=loader.MAX_LENGTH):
with torch.no_grad():
input_tensor = train.tensorFromSentence(input_lang, sentence)
# |input_tensor| = (sentence_length, 1)
input_length = input_tensor.size(0)
encoder_hidden = (encoder.initHidden().to(device),
encoder.initHidden().to(device))
# |encoder_hidden[0]|, |encoder_hidden[1]| = (2, 1, hidden_size/2)
encoder_outputs = torch.zeros(max_length,
encoder.hidden_size,
device=device)
# |encoder_outputs| = (max_length, hidden_size)
for ei in range(input_length):
encoder_output, encoder_hidden = encoder(input_tensor[ei],
encoder_hidden)
# |encoder_output| = (1, 1, hidden_size)
# |encoder_hidden[0]|, |encoder_hidden[1]| = (2, 1, hidden_size/2)
encoder_outputs[ei] += encoder_output[0, 0]
decoder_input = torch.tensor([[loader.SOS_token]], device=device)
decoder_hidden = train.merge_encoder_hiddens(encoder_hidden)
# |decoder_input| = (1, 1)
# |decoder_hidden[0]|, |decoder_hidden[1]| = (1, 1, hidden_size)
decoded_words = []
for di in range(max_length):
decoder_output, decoder_hidden = decoder(decoder_input,
decoder_hidden)
# |decoder_output| = (1, output_lang.n_words)
# |decoder_hidden[0]|, |decoder_hidden[1]| = (1, 1, hidden_size)
topv, topi = decoder_output.data.topk(
1) # decoder_output.data == decoder_output
# |topv| = (1, 1)
# |topi| = (1, 1)
if topi.item() == loader.EOS_token:
decoded_words.append('<EOS>')
break
else:
decoded_words.append(output_lang.index2word[topi.item()])
decoder_input = topi.squeeze().detach()
return decoded_words
def evaluate(input_lang,
output_lang,
device,
encoder,
decoder,
sentence,
max_length=MAX_LENGTH):
with torch.no_grad():
input_tensor = tensorFromSentence(input_lang, sentence, device)
input_length = input_tensor.size()[0]
encoder_hidden = encoder.initHidden()
encoder_outputs = torch.zeros(max_length,
encoder.hidden_size,
device=device)
for ei in range(input_length):
encoder_output, encoder_hidden = encoder(input_tensor[ei],
encoder_hidden)
encoder_outputs[ei] += encoder_output[0, 0]
decoder_input = torch.tensor([[SOS_token]], device=device) # SOS
decoder_hidden = encoder_hidden
decoded_words = []
decoder_attentions = torch.zeros(max_length, max_length)
for di in range(max_length):
decoder_output, decoder_hidden, decoder_attention = decoder(
decoder_input, decoder_hidden, encoder_outputs)
decoder_attentions[di] = decoder_attention.data
topv, topi = decoder_output.data.topk(1)
if topi.item() == EOS_token:
decoded_words.append('<EOS>')
break
else:
decoded_words.append(output_lang.index2word[topi.item()])
decoder_input = topi.squeeze().detach()
return decoded_words, decoder_attentions[:di + 1]
