Tensorboard嵌入，仅在悬停时显示精灵

Question

我正在寻找一种在悬停时显示嵌入精灵的方法，与使用Tensorboard投影仪时的常量相反。下面的代码成功创建了一个嵌入式电路板，并将适当的元数据和精灵附加到每个数据点。但是，我尝试可视化超过10,000点的数据集，因此在单个屏幕上显示所有图像会使探索变得异常困难。相反，我正在寻找一种仅在悬停时显示精灵图像的方法，与常量相反。有没有人能够做到这一点？

from tensorflow.contrib.tensorboard.plugins import projector

with tf.Session() as sess:
    # Create summary writer.
    writer = tf.summary.FileWriter('./graphs/embedding_test', sess.graph)

    # Initialize embedding_var
    sess.run(embedding_var.initializer)

    # Create Projector config
    config = projector.ProjectorConfig()

    # Add embedding visualizer
    embedding = config.embeddings.add()

    # Attach the embedding name
    embedding.tensor_name = embedding_var.name

    # Data point meta data
    embedding.metadata_path = EMBEDDING_META_PATH

    #associate data point sprite
    embedding.sprite.image_path = EMBEDDING_SPRITE_PATH

    #specify dimensions of images within sprite
    embedding.sprite.single_image_dim.extend([imgSize, imgSize])

    # Add writer and config to Projector
    projector.visualize_embeddings(writer, config)

    # Save the model
    saver_embed = tf.train.Saver([embedding_var])
    saver_embed.save(sess, './graphs/embedding_test/embedding_test.ckpt', 1)



writer.close()

正如你所看到的，不断地显示所有图像是相当压倒性的......