def evaluate(epoch_number, train_dataset_path, labels_dataset_path,
sentences_max_length):
train_generator = DataGenerator(train_dataset_path,
sentences_max_length,
tensor=True)
eval_generator = DataGenerator(labels_dataset_path,
sentences_max_length,
tensor=False,
labels=True)
alphabet_size = train_generator.get_dictionary_size()
output_syms = eval_generator.get_dictionary_size()
chars_dict = eval_generator.create_chars_dict()
model = TokNet(alphabet_size, sentences_max_length, output_syms)
file_name = "state_{}.pt".format(epoch_number)
model.load_state_dict(torch.load(os.path.join("./models", file_name)))
model.eval()
t_it = train_generator.__iter__()
e_it = eval_generator.__iter__()
n_int = 0
OK = 0
for i in range(1000):
sentence = next(t_it)
out = model(sentence)
y = torch.tensor([
elem.argmax()
for elem in out.view(-1, eval_generator.get_dictionary_size())
])
gold = str(next(e_it))
s = ""
for n in y:
s += list(chars_dict.keys())[list(chars_dict.values()).index(
int(n))]
for i in range(min(len(s), len(gold))):
n_int += 1
if (s[i] == gold[i]):
OK += 1
return OK / n_int
return listOfKeys
train_generator = DataGenerator(train_dataset_path,
Params.sentences_max_length,
tensor=True,
monograms=Params.monograms)
eval_generator = DataGenerator(labels_dataset_path,
Params.sentences_max_length,
tensor=True,
monograms=1)
alphabet_size = train_generator.get_dictionary_size()
output_syms = eval_generator.get_dictionary_size()
chars_dict = eval_generator.create_chars_dict()
model = TokNet(alphabet_size, Params.sentences_max_length, output_syms)
file_name = "state_0.pt"
write = 0
if (len(sys.argv) == 2):
file_name = "state_" + str(sys.argv[1]) + ".pt"
if (len(sys.argv) == 3):
file_name = "state_" + str(sys.argv[1]) + ".pt"
write = int(sys.argv[2])
model.load_state_dict(torch.load(os.path.join("./models", file_name)))
model.eval()
n_int = 0
OK = 0
