How to use the projector?

[FSet89]

During training, the projector shows some points in the PCA space. Every point has the same color. What do they represent? What can I infer about the training process from this graph?

[omoindrot]

Normally you should be able to color by label if you provided the labels.

For MNIST, you will have one color for each digit.

Maybe this post can help your understanding? https://stackoverflow.com/questions/40849116/how-to-use-tensorboard-embedding-projector

[FSet89]

Thank you. Can you point me out where in your code you add the images for the projector summary?

[batrlatom]

It is in the visualize_embeddings.py
I think it is here

embedding.sprite.image_path = pathlib.Path(args.sprite_filename).name
    embedding.sprite.single_image_dim.extend([28, 28])

[omoindrot]

So it's in the metadata tsv file that you need to save:

tensorflow-triplet-loss/visualize_embeddings.py

Lines 83 to 90 in fc69836

	# Specify where you find the metadata
	# Save the metadata file needed for Tensorboard projector
	metadata_filename = "mnist_metadata.tsv"
	with open(os.path.join(eval_dir, metadata_filename), 'w') as f:
	for i in range(params.eval_size):
	c = labels[i]
	f.write('{}\n'.format(c))
	embedding.metadata_path = metadata_filename

    # Specify where you find the metadata
    # Save the metadata file needed for Tensorboard projector
    metadata_filename = "mnist_metadata.tsv"
    with open(os.path.join(eval_dir, metadata_filename), 'w') as f:
        for i in range(params.eval_size):
            c = labels[i]
            f.write('{}\n'.format(c))
    embedding.metadata_path = metadata_filename

If you want to visualize other colors, you can add it in the tsv file as a new column like this:

    metadata_filename = "mnist_metadata.tsv"
    with open(os.path.join(eval_dir, metadata_filename), 'w') as f:
        f.write("label\tother")
        for i in range(params.eval_size):
            c = labels[I]
            other = c % 2
            f.write('{}\t{}\n'.format(c, other))
    embedding.metadata_path = metadata_filename