#########################
class Params(object):
pass
p = Params()
p.embeddings = embeddings
p.conv_layer_n = 1
p.ks = [3]
p.fold = [1]
p.W = [W]
p.b = [b]
p.W_logreg = W_logreg
p.b_logreg = b_logreg
dcnn = DCNN(p)
##################### Testing ####################
from test_util import (assert_matrix_eq, assert_about_eq)
x = np.asarray(np.random.randint(vocab_size, size = (3, 6)),
dtype=np.int32
)
y = np.asarray(np.random.randint(5, size = 3),
dtype=np.int32
)
########### Embedding layer ##############
actual = f_el(x)
get_padding = lambda sent: [padding_token_index] * (max_len - len(sent))
padded_sents = [(sent + get_padding(sent)) for sent in sents]
return padded_sents
WORD2INDEX = load(open("data/twitter.pkl"))[3]
PADDING_INDEX = WORD2INDEX[u"<PADDING>"]
from param_util import load_dcnn_model_params
from dcnn import DCNN
params = load_dcnn_model_params(
"models/filter_widths=8,6,,batch_size=10,,ks=20,8,,fold=1,1,,conv_layer_n=2,,ebd_dm=48,,l2_regs=1e-06,1e-06,1e-06,0.0001,,dr=0.5,0.5,,nkerns=7,12.pkl"
)
MODEL = DCNN(params)
def sentiment_scores_of_sents(sents):
"""
Predict the sentiment positive scores for a bunch of sentences
>>> sentiment_scores_of_sents([u'simultaneously heart breaking and very funny , the last kiss is really all about performances .', u'( u ) stupid .'])
array([ 0.78528505, 0.0455901 ])
"""
word_indices = [
get_word_index_array(nltk.word_tokenize(sent), WORD2INDEX)
for sent in sents
]
x = np.asarray(pad_sents(word_indices, PADDING_INDEX), dtype=np.int32)
class Params(object):
pass
p = Params()
p.embeddings = embeddings
p.conv_layer_n = 1
p.ks = [3]
p.fold = [1]
p.W = [W]
p.b = [b]
p.W_logreg = W_logreg
p.b_logreg = b_logreg
dcnn = DCNN(p)
##################### Testing ####################
from test_util import (assert_matrix_eq, assert_about_eq)
x = np.asarray(np.random.randint(vocab_size, size=(3, 6)), dtype=np.int32)
y = np.asarray(np.random.randint(5, size=3), dtype=np.int32)
########### Embedding layer ##############
actual = f_el(x)
expected = dcnn.e_layer.output(x)
assert_matrix_eq(actual, expected, "Embedding")
########## Conv layer ###################
from dcnn import DCNN
from util import load_data
from param_util import load_dcnn_model_params
params = load_dcnn_model_params(
"models/filter_widths=8,6,,batch_size=10,,ks=20,8,,fold=1,1,,conv_layer_n=2,,ebd_dm=48,,l2_regs=1e-06,1e-06,1e-06,0.0001,,dr=0.5,0.5,,nkerns=7,12.pkl"
)
model = DCNN(params)
datasets = load_data("data/twitter.pkl")
dev_set_x, dev_set_y = datasets[1]
test_set_x, test_set_y = datasets[2]
dev_set_x, dev_set_y = dev_set_x.get_value(), dev_set_y.get_value()
test_set_x, test_set_y = test_set_x.get_value(), test_set_y.get_value()
print "dev error:", model._errors(dev_set_x, dev_set_y)
print "test error:", model._errors(test_set_x, test_set_y)
# Cuántas palabras tieen que identificar el modelo
VOCAB_SIZE = tokenizer.vocab_size
EMB_DIM = 200 # Todas y cada una de las palabras se mapearán a un espacio de dimensión 200
NB_FILTERS = 100
FFN_UNITS = 256
NB_CLASSES = 2 # len(set(train_labels))
DROPOUT_RATE = 0.2
BATCH_SIZE = 32
NB_EPOCHS = 5
# FASE DE ENTRENAMIENTO:
print('Entrenamiento')
Dcnn = DCNN(vocab_size=VOCAB_SIZE,
emb_dim=EMB_DIM,
nb_filters=NB_FILTERS,
FFN_units=FFN_UNITS,
nb_classes=NB_CLASSES,
dropout_rate=DROPOUT_RATE)
if NB_CLASSES == 2:
Dcnn.compile(loss="binary_crossentropy",
optimizer="adam",
metrics=["accuracy"])
else:
Dcnn.compile(loss="sparse_categorical_crossentropy",
optimizer="adam",
metrics=["sparse_categorical_accuracy"])
# Sistema que nos permite marcar checkpoints en
# el entrenamiento para que cada cierto tiempo
# se vaya guardando la información y podamos
from dcnn import DCNN
from util import load_data
from param_util import load_dcnn_model_params
params = load_dcnn_model_params("models/filter_widths=8,6,,batch_size=10,,ks=20,8,,fold=1,1,,conv_layer_n=2,,ebd_dm=48,,l2_regs=1e-06,1e-06,1e-06,0.0001,,dr=0.5,0.5,,nkerns=7,12.pkl")
model = DCNN(params)
datasets = load_data("data/twitter.pkl")
dev_set_x, dev_set_y = datasets[1]
test_set_x, test_set_y = datasets[2]
dev_set_x, dev_set_y = dev_set_x.get_value(), dev_set_y.get_value()
test_set_x, test_set_y = test_set_x.get_value(), test_set_y.get_value()
print "dev error:", model._errors(dev_set_x, dev_set_y)
print "test error:", model._errors(test_set_x, test_set_y)
