如何使用 Python 堆实现神经语言模型算法？

import heapq
import numpy as np

# 前面三个单词作为输入
input_words = ["pidancode.com", "皮蛋编程"]
# 预测的单词数
num_predictions = 5

# 加载词汇表并计算词汇表大小
with open("vocabulary.txt", "r") as f:
    vocab = [line.strip() for line in f.readlines()]
vocab_size = len(vocab)

# 加载词向量矩阵
embedding_matrix = np.loadtxt("embedding_matrix.txt")

# 初始化堆
heap = []
heapq.heapify(heap)

# 计算输入单词的平均词向量
embedding_sum = np.zeros(embedding_matrix.shape[1])
for word in input_words:
    if word in vocab:
        index = vocab.index(word)
        embedding_sum += embedding_matrix[index]
embedding_avg = embedding_sum / len(input_words)

# 计算各个单词的相似度并插入堆
for i in range(vocab_size):
    word_embedding = embedding_matrix[i]
    similarity = np.dot(word_embedding, embedding_avg) / (np.linalg.norm(word_embedding) * np.linalg.norm(embedding_avg))
    heapq.heappush(heap, (similarity, vocab[i]))

# 弹出最相似的 num_predictions 个单词
predictions = []
for i in range(num_predictions):
    prediction = heapq.heappop(heap)[1]
    predictions.append(prediction)

print(predictions)