Add projector plugin colab #3423
Conversation
Adds a colab demo demonstrating how to use the projector plugin locally.
|
Check out this pull request on You'll be able to see Jupyter notebook diff and discuss changes. Powered by ReviewNB. |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:18Z Should be 2020. hfiller commented on 2020-03-31T21:19:13Z Done |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:19Z The table here isn't rendering correctly. |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:20Z I think "highly dimensional embeddings" -> "high dimensional embeddings".
Also, "visuallizing" typo
An image here might be nice :). |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:20Z I don't think we need to say that we are using TF 2.
'tensorboard' -> 'TensorBoard' (also there is an extra '`')
It seems like we are not actually using the MNIST dataset below. |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:21Z Maybe start with "We will use a dataset of ..." just to make it read a bit more smoothly. |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:22Z Should we refer to it as just "embedding" or "embedding layer"? I think the layer is what provides the embeddings.
I think instead of "vector" we should explicitly say "embedding"
It might be worth linking to additional resources so people can learn about embeddings. E.g. https://www.tensorflow.org/tutorials/text/word_embeddings?hl=en |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:22Z "The layer we are embedding": This makes it seem like embedding is a verb we are applying to the layer. I think instead it is "The embedding layer". |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:24Z Can this be done by the Keras TensorBoard callback?
Is this the only way of saving embeddings? hfiller commented on 2020-03-30T08:07:13Z Currently yes, I believe the Keras TensorBoard callback is broken, I'm currently working on fixing this. |
|
View / edit / reply to this conversation on ReviewNB GalOshri commented on 2020-03-26T02:35:24Z We should add an image here otherwise the tutorial page won't have it (see how it is done as an example at https://github.com/tensorflow/tensorboard/blob/90227dcf7e7f900ba77b5b7ced1b14378a7ec98f/docs/hyperparameter_tuning_with_hparams.ipynb
I think it would be useful to add some explanation of why this is interesting/valuable. What does the user get from doing this? Is there some insight? hfiller commented on 2020-03-31T21:20:30Z Added some analysis including a comparison between Alfred Hitchcock and Wes Anderson. Let me know what you think |
|
Thank you for adding this! This tutorial will be super valuable for helping people learn about and get started with the projector plugin. Can you also add it to https://github.com/tensorflow/tensorboard/blob/master/docs/_book.yaml? This enables it to show up in the left pane on the docs website. Let's put it under the hyperparameter tuning tutorial. |
| @@ -0,0 +1,337 @@ | |||
| { | |||
There was a problem hiding this comment.
"Tensorboard" -> "TensorBoard"
"from your tensorflow projects from the logs in the specified directory log_dir " is a bit confusing. Perhaps something like "from the logs in the specifiedlog_dir directory "?
Reply via ReviewNB
| @@ -0,0 +1,337 @@ | |||
| { | |||
There was a problem hiding this comment.
It might be worth including a sentence on how you identified the more closely associated words (search for a particular word, the neighbors are highlighted. These are words that are close to it in the embedding space).
Reply via ReviewNB
docs/_book.yaml
Outdated
| @@ -23,6 +23,8 @@ upper_tabs: | |||
| path: /tensorboard/graphs | |||
| - title: "Hyperparameter tuning" | |||
| path: /tensorboard/hyperparameter_tuning_with_hparams | |||
| - title: "Projector plugin" | |||
There was a problem hiding this comment.
We want to avoid using "plugin" in user documentation. How about "Embedding projector" or "Projector dashboard"?
* Add projector plugin colab * Adds a colab demo demonstrating how to use the projector plugin locally. * Add embedding_projector image for colab * Add colab to _book.yaml * Add analysis to projector plugin * Adding images for embedding projector colab * Rename Projector plugin to Embedding projector
* Add projector plugin colab * Adds a colab demo demonstrating how to use the projector plugin locally. * Add embedding_projector image for colab * Add colab to _book.yaml * Add analysis to projector plugin * Adding images for embedding projector colab * Rename Projector plugin to Embedding projector
Adds a colab demo demonstrating how to use the projector plugin locally.