基于字符的RNN模型如何预测字符？

import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import LSTM, Dense

# 数据预处理
text = "hello world"
chars = sorted(set(text))
char_to_index = {c: i for i, c in enumerate(chars)}
index_to_char = {i: c for i, c in enumerate(chars)}

seq_length = 3
dataX = []
dataY = []

for i in range(0, len(text) - seq_length, 1):
    seq_in = text[i:i + seq_length]
    seq_out = text[i + seq_length]
    dataX.append([char_to_index[char] for char in seq_in])
    dataY.append(char_to_index[seq_out])

n_patterns = len(dataX)
n_chars = len(chars)

X = np.reshape(dataX, (n_patterns, seq_length, 1))
X = X / float(n_chars)
y = tf.keras.utils.to_categorical(dataY)

# 构建模型
model = Sequential()
model.add(LSTM(256, input_shape=(X.shape[1], X.shape[2])))
model.add(Dense(y.shape[1], activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam')

# 训练模型
model.fit(X, y, epochs=20, batch_size=128)

# 预测
start = np.random.randint(0, len(dataX) - 1)
pattern = dataX[start]
print("Seed:")
print("\"", ''.join([index_to_char[value] for value in pattern]), "\"")

for i in range(50):
    x = np.reshape(pattern, (1, len(pattern), 1))
    x = x / float(n_chars)

    prediction = model.predict(x, verbose=0)
    index = np.argmax(prediction)
    result = index_to_char[index]
    seq_in = [index_to_char[value] for value in pattern]
    print(result, end='')

    pattern.append(index)
    pattern = pattern[1:len(pattern)]