使用Tensorflow可视化多重嵌入

import tensorflow as tf
import numpy as np

# 示例词嵌入矩阵和词汇表
embedding_matrix = np.random.rand(1000, 128)  # 1000个词，每个词128维
vocab = {i: f"word_{i}" for i in range(1000)}

embedding_var = tf.Variable(embedding_matrix, name='embedding')

import os

log_dir = 'logs/embedding'
os.makedirs(log_dir, exist_ok=True)

# 保存嵌入变量
checkpoint = tf.train.Checkpoint(embedding=embedding_var)
checkpoint.save(os.path.join(log_dir, "embedding.ckpt"))

# 保存词汇表
with open(os.path.join(log_dir, "vocab.txt"), "w") as f:
    for word in vocab.values():
        f.write(f"{word}\n")

embeddings {
  tensor_name: "embedding"
  metadata_path: "vocab.txt"
}

tensorboard --logdir logs/embedding

embedding_var1 = tf.Variable(embedding_matrix1, name='embedding1')
embedding_var2 = tf.Variable(embedding_matrix2, name='embedding2')

checkpoint1 = tf.train.Checkpoint(embedding=embedding_var1)
checkpoint1.save(os.path.join(log_dir, "embedding1.ckpt"))

checkpoint2 = tf.train.Checkpoint(embedding=embedding_var2)
checkpoint2.save(os.path.join(log_dir, "embedding2.ckpt"))

with open(os.path.join(log_dir, "vocab1.txt"), "w") as f:
    for word in vocab1.values():
        f.write(f"{word}\n")

with open(os.path.join(log_dir, "vocab2.txt"), "w") as f:
    for word in vocab2.values():
        f.write(f"{word}\n")

embeddings {
  tensor_name: "embedding1"
  metadata_path: "vocab1.txt"
}

embeddings {
  tensor_name: "embedding2"
  metadata_path: "vocab2.txt"
}

tensorboard --logdir logs/embedding