from keras.models import load_model
import helper
import numpy as np
import sys
SEQUENCE_LENGTH = 40
SEQUENCE_STEP = 3
PATH_TO_CORPUS = "pink_floyd_lyrics.txt"
EPOCHS = 20
DIVERSITY = 1.0
text = helper.read_corpus(PATH_TO_CORPUS)
chars = helper.extract_characters(text)
sequences, next_chars = helper.create_sequences(text, SEQUENCE_LENGTH,
SEQUENCE_STEP)
char_to_index, indices_char = helper.get_chars_index_dicts(chars)
# load json and create model
json_file = open('model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights("lyrical_lstm_weights.h5")
loaded_model.load_weights("new_weights.h5")
print("Loaded model from disk")
x = np.zeros((1, 40, 49))
print(loaded_model.predict(x, verbose=0)[0])
def preprocess(input_text):
PATH_TO_CORPUS = "leevi_corpus.txt"
EPOCHS = 5
DIVERSITY = 1.0
"""
Read the corpus and get unique characters from the corpus.
"""
text = helper.read_corpus(PATH_TO_CORPUS)
words = text.split()
unique_words = helper.extract_characters(words)
"""
Create sequences that will be used as the input to the network.
Create next_chars array that will serve as the labels during the training.
"""
word_sequences, next_words = helper.create_word_sequences(
words, WORD_SEQUENCE_LENGTH, WORD_SEQUENCE_STEP)
word_to_index, indices_word = helper.get_chars_index_dicts(unique_words)
# """
# The network is not able to work with characters and strings, we need to vectorise.
# """
X, y = helper.vectorize(word_sequences, WORD_SEQUENCE_LENGTH, unique_words,
word_to_index, next_words)
# """
# Define the structure of the model.
# """
model = helper.build_model(WORD_SEQUENCE_LENGTH, unique_words)
# """
# Train the model
# """