from dataHandler import DataHandler
from boW import EOS_token
RSEED = 50
SIZE = 10
NCLASS = 3
TFIDF = False
NORMALIZE = False
np.random.seed(RSEED)
d = DataHandler()
d.createDictionary()
#prepare data
data = d.readDataPreproc(pre=True)[1:]
x = [d.askDictionary.seq2tensor(xi[1], tfidf=TFIDF) for xi in data]
x = np.array(x)
if NORMALIZE:
x = preprocessing.normalize(x)
if NCLASS == 3:
y = [
0 if yi[2] == "alegria" else 1 if yi[2] == "neutro" else 2
for yi in data
]
else:
y = [0 if yi[2] == "alegria" else 1 for yi in data]
y = np.array(y)
output_sentence = ' '.join(output_words)
print('<', output_sentence)
print('')
def evaluateAndShowAttention(self, input_sentence):
output_words, attentions = self.evaluate(input_sentence)
print('input =', input_sentence)
print('output =', ' '.join(output_words))
showAttention(input_sentence, output_words, attentions)
if __name__ == '__main__':
d = DataHandler()
d.createDictionary()
pairs = d.readDataPreproc()
s = Seq2Seq(d.askDictionary, d.ansDictionary)
if isfile(PATH):
s = torch.load(PATH)
showPlot(s.plot_losses,
title="Loss",
axis=["Milhares de iterações", "loss"])
else:
s.trainIters(pairs, 10000, print_every=5000)
torch.save(s.state_dict(), "disc_" + PATH)
torch.save(s, PATH)
s.evaluateRandomly(pairs)