def predict(self, input_data):
try:
title_1, title_2 = input_data['title_1'], input_data['title_2']
tokens1 = np.array([
shutils.padd_fn(shutils.get_tokens(title_1),
max_len=cnt.MAX_WORDS)
])
tokens2 = np.array([
shutils.padd_fn(shutils.get_tokens(title_2),
max_len=cnt.MAX_WORDS)
])
sent_data_1 = shutils.get_vectors(self.vector_model, tokens1,
cnt.VECTOR_DIM)
sent_data_2 = shutils.get_vectors(self.vector_model, tokens2,
cnt.VECTOR_DIM)
prediction, probability = self.network.predict(
[sent_data_1, sent_data_2], return_probability=True)
return {
"status": 1,
"is_match": int(prediction[0]),
"confidence": float(probability[0])
}
except Exception as err:
return {"status": 0, "message": str(err)}
def get_all_tokens_for_vector(items, char_tokens=False):
text_data = [str(item[0]) for item in items]
all_tokens = [
shutils.padd_fn(shutils.get_tokens(x, char_tokens=char_tokens),
max_len=cnt.MAX_WORDS) for x in text_data
]
return all_tokens
def get_all_tokens_for_vector(items, char_tokens=False):
wm_text = [str(item[3]) for item in items]
cm_text = [str(item[5]) for item in items]
all_titles = wm_text + cm_text
all_tokens = [shutils.padd_fn(shutils.get_tokens(title, char_tokens=char_tokens), max_len=cnt.MAX_WORDS) for title in all_titles]
return all_tokens
def get_tokens_indices(items, indices):
data_pairs = []
for i in indices:
wm_title = str(items[i][3])
cm_title = str(items[i][5])
# wm_desc = str(items[i][4])
# cm_desc = str(items[i][6])
wm_word_tokens = shutils.padd_fn(shutils.get_tokens(wm_title, char_tokens=False), max_len=cnt.MAX_WORDS)
cm_word_tokens = shutils.padd_fn(shutils.get_tokens(cm_title, char_tokens=False), max_len=cnt.MAX_WORDS)
wm_char_tokens = [shutils.padd_fn(shutils.get_tokens(token, char_tokens=True), max_len=cnt.MAX_CHARS) for token in wm_word_tokens]
cm_char_tokens = [shutils.padd_fn(shutils.get_tokens(token, char_tokens=True), max_len=cnt.MAX_CHARS) for token in cm_word_tokens]
label = int(items[i][-1])
data_pairs.append((wm_word_tokens, cm_word_tokens, wm_char_tokens, cm_char_tokens, label))
return data_pairs
def predict_batch(self, multi_input_data):
try:
tokens1, tokens2 = [], []
for input_data in multi_input_data:
title_1, title_2 = input_data['title_1'], input_data['title_2']
tokens1.append(
shutils.padd_fn(shutils.get_tokens(title_1),
max_len=cnt.MAX_WORDS))
tokens2.append(
shutils.padd_fn(shutils.get_tokens(title_2),
max_len=cnt.MAX_WORDS))
tokens1 = np.array(tokens1)
tokens2 = np.array(tokens2)
sent_data_1 = shutils.get_vectors(self.vector_model, tokens1,
cnt.VECTOR_DIM)
sent_data_2 = shutils.get_vectors(self.vector_model, tokens2,
cnt.VECTOR_DIM)
prediction, probability = self.network.predict(
[sent_data_1, sent_data_2], return_probability=True)
return {
"status":
1,
"response": [{
"status": 1,
"is_match": int(prediction[i]),
"confidence": float(probability[i])
} for i in range(len(multi_input_data))]
}
except Exception as err:
return {"status": 0, "message": str(err)}