def main():
# Get tokenizer
with open(os.path.join(DATA_PATH, 'special/tokenizer.p'), 'rb') as f:
tokenizer = pickle.load(f)
# Build reverse dictionary
reverse_word_map = dict(map(reversed, tokenizer.word_index.items()))
with open(os.path.join(DATA_PATH, 'special/embedding_matrix.npy'),
'rb') as f:
embeddings = np.load(f, allow_pickle=True)
print("Please enter task number")
task_num = int(input())
if task_num <= 0 or task_num > 20:
print("Wrong task number")
else:
model = load_model('lstm_glove_train.h5')
print(model.summary())
running = True
while running:
# Get input from user
print("Please enter context or type 'q' for quit the program: ")
context = input()
if context == 'q':
running = False
break
print("Please enter question based on the context: ")
question = input()
# Transform the context and answer input
context = transform(context,
max_len=MAX_CONTEXT,
tokenizer=tokenizer)
question = transform(question,
max_len=MAX_QUESTION,
tokenizer=tokenizer)
answer = model.predict([context, question])
correct_tag_id = np.argmax(
answer) # Turn one hot encoding to index
# TODO: this only gives index 10 which is bedroom
# pred_word_ind = model.predict_classes(pad_encoded,verbose=0)[0]
# This doesnt work somehow saying there is no predict_classes
word = reverse_word_map.get(correct_tag_id)
print(word)
EMBED_HIDDEN_SIZE = 300 # size of the vector space (word embedding)
###################################
# Loading dataset #
###################################
# Get tokenizer
with open(os.path.join(data_path, 'special/tokenizer.p'), 'rb') as f:
tokenizer = pickle.load(f)
vocab_size = len(tokenizer.word_index) + 1
# Restore the model and construct the encoder and decoder.
model = load_model('lstm_train{}.h5'.format(TASK_NBR))
# Get Test data
with open(os.path.join(data_path, 'Context_Test_{}.txt'.format(TASK_NBR)), 'r') as f:
context = f.read().strip().split('\n')
with open(os.path.join(data_path, 'Question_Test_{}.txt'.format(TASK_NBR)), 'r') as f:
question = f.read().strip().split('\n')
with open(os.path.join(data_path, 'Answer_Test_{}.txt'.format(TASK_NBR)), 'r') as f:
answer = f.read().strip().split('\n')
context = transform(context, max_len=MAX_CONTEXT, tokenizer=tokenizer)
question = transform(question, max_len=MAX_QUESTION, tokenizer=tokenizer)
answer = transform(answer, max_len=1, tokenizer=tokenizer)
encoded_answer = to_categorical(tf.squeeze(answer, axis=1), num_classes=vocab_size)
loss, acc = model.evaluate([context, question], encoded_answer, batch_size=BATCH_SIZE)
print("Test Task {}: loss/accuracy = {:.4f}, {:.4f}".format(TASK_NBR, loss, acc))
# Test Task 1: loss/accuracy = 3.9904, 0.4360
# Test Task 2: loss/accuracy = 3.6211, 0.2090
# Test Task 3: loss/accuracy = 3.1279, 0.1610