How does projector plugin work

If you upload meta.tsv and vec.tsv , You can visualize the embedding at online https://projector.tensorflow.org/

But if I use tensorboard with

projector_config.pbtxt as the following

embeddings {
    tensor_name: "embedding_name"
    metadata_path: "./assets/meta.tsv"
    tensor_path: "./assets/vecs.tsv"
}

it throws an error as No checkpoint has been saved yet. Please refresh the page periodically

Is it neccessary to have .ckpt file since we are able to visualize it using the normal two tensor files which are the meta.tsv data file and vec.tsv file which I uploaded at https://projector.tensorflow.org/ using load button step 1 and step 2

So I wanted to know why doesn’t it work in tensorboard --logdir . / localhost

cc @stephanwlee

Note : I am gsoc student for mlpack and I would be building custom backend for mlpack to log metrics and sclaers and other information so that tensorbaord could be used as a frontend , We would be supporting all the protobuf formats ,While working I stumble upon the projector plugin and couldn’t move ahead. Any help would be appreciated, I did query first on gitter and then was redirected to raise a issue to know in depth about the internals

Thanks

[TKassis]

Thanks! Still not working, unfortunately. I gave up. I’m justing using the one hosted by Tensorflow and uploading my TSV files.

[jeffin143]

@TKassis , for me the issue was I gave an absolute path instead of relative Also My config is as following

embeddings {
tensor_name: "embedding_name"
tensor_path: "logs/imdb-example/vecs.tsv"
metadata_path: "logs/imdb-example/metadata.tsv"
}

Only the above things are required

As far as I know --> Model checkpoint path is not needed

Also the path should be relative to log dir, What I suggest is keep projector_config.pbtxt in log dir and the give relative path ../../imdb-example/vecs.tsv