Tensorboard: "Cannot read property 'toString' of undefined" in embeddings projector

[wchargin]

Migrated from https://github.com/tensorflow/tensorflow/issues/9094.

Some nice repro files and tracebacks over there. :-)

[atishsawant]

Was there a fix for this?

[wchargin]

Not as far as I recall. cc @dsmilkov ?

[CermakM]

:+1: Stumbled upon this issue as well. However, I believe my problem is a bit different, since I don't even get to the 3D label enabling part.
The labels seem to be properly loaded, however (See the image bellow).

Any progress with this? Thanks!

[jart]

Marking contributions welcome because it looks like a potentially easy fix that might simply be related to the type of data being fed into the projector.

[Camuslu]

I met the same issue today. Would be eager to see if this gets fixed.

[jart]

Searching the codebase, this is probably being triggered by metadata labels, e.g. point.metadata[this.selectedLabelOption].toString();. I know @francoisluus made a good contribution not too long ago for Interactive Supervision which includes label editing. Perhaps he could take a look?

[francoisluus]

I can take a look, thanks for the notification.

@CermakM @Camuslu if it is possible at all, please share a (sanitized) zipped version of your log folder used with TensorBoard, so that I can recreate the scenario, thanks.

[Camuslu]

@francoisluus It turns out mine wasn't really a bug in Tensorboard but something I did wrong. It's working now on my end.

[CermakM]

@francoisluus, probably something similar to @Camuslu ... also made it work a while ago, I think it was a problem with the embeddings being different size than the labels (not 100% sure) .. the error is indeed very uninformative, though.

[MathieuLAllier]

Encoding the labels into utf-8 resolved the problem for me.

[AmitMY]

Solution kinda works .encode('utf8')

No more error, but now the string is:

[MathieuLAllier]

On the Labels itself.

see here
https://github.com/tensorflow/tensorboard/issues/71#issuecomment-461739767

On Mon, 4 Mar 2019 at 07:48, bono1567 [e-mail] wrote:

Encoding the labels into utf-8 resolved the problem for me.

Where did you make a change??

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/tensorflow/tensorboard/issues/71#issuecomment-469240878,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AQhtTx05rAI7n-ZVSgsaH23eJIBO5tHpks5vTRYegaJpZM4N833a
.

--
Mathieu L'Allier
514 213 3767
M.[email protected]
http://ca.linkedin.com/in/mathieulallier

[bono1567]

It throws the same error if the number of tensors are not equal to the number of labels in the tsv file

[reyhan1994]

how can i Encod my labels into utf-8 ?

[Proteusiq]

look at your .tsv do you have. You might be having 1 letter per line instead of word or N-gram of words per line. I discovered data processing was wrong :)

[wakameog]

Me too.

I was able to plot it well in GoogleColaboratory. However, I get an error in the local environment.
Is the problem with plotting in Japanese?

[reyhan1994]

I was able to plot it well in GoogleColaboratory two month ago, but now I can't, I don't know what is the problem can you help me? I really need this plotting for my final project. I am plotting in Iran. I attach the photos of my errors.

May you send me the version of your TensorFlow and tensorBoard?and the part of related code please, thanks a lot.