How to Display Data Values for 3D Embeddings in Visualization

·Apr 05, 2025 03:01 AM

My goal is to be able to hover over the individual embeddings in the 3D view and be able to see the actual data. How to make these "unknowns" show as actual values in my dataset. I want to visualize just my dataset, and not run any inference or pass in any prompts.

10 comments

· Sorted by Oldest

RunLLM
APP
·
To view actual data when hovering over individual embeddings in the 3D view, Phoenix provides tooltips that display relevant information about each data point. This feature was enhanced to improve the positioning and visibility of tooltips, as described in a <https://github.com/Arize-ai/phoenix/pull/890|GitHub pull request>. Ensure that your dataset is properly loaded with the necessary metadata to be displayed in the tooltips. For automatically assigning colors to different clusters, Phoenix allows you to color the UMAP point-cloud by various dimensions such as features, tags, predictions, and actuals. This can be done using the color by dimension feature, which maps points to distinct groups based on categorical or numeric values, as detailed in a <https://github.com/Arize-ai/phoenix/pull/423|GitHub pull request>. Would you like more detailed instructions on how to implement these features? Sources: - <https://github.com/Arize-ai/phoenix/pull/890|GitHub pull request on tooltips> - <https://github.com/Arize-ai/phoenix/pull/423|GitHub pull request>
💡 Hint: Mention RunLLM in the thread for followups.
Dustin N.
·
hi Lily S. thanks for using Phoenix. Lately we've been focusing on the LLM-observability functionality of Phoenix and embeddings visualization hasn't been prioritized lately. Unfortunately I'm not super familiar with this part of the product and can't be much help to you Xander S. do you know if we support the functionality Lily is looking for?
Xander S.
·
Hey Lily S., can you verify that your image URLs are accessible from your browser? It may also help to inspect the failed fetch requests in your brwoser's network tools. Here's an example notebook: https://colab.research.google.com/github/Arize-ai/phoenix/blob/main/tutorials/image_classification_tutorial.ipynb
Xander S.
·
I guess that is assuming you are trying to visualize images. Maybe you are dealing with text?
Lily S.
·
I have no images in my dataset.
Xander S.
·
It sounds like you're working with text data Lily S.?
Lily S.
·
Yes, just text data.
Xander S.
·
Might be worth taking a look at this tutorial if you haven't already: https://colab.research.google.com/github/Arize-ai/phoenix/blob/main/tutorials/sentiment_classification_tutorial.ipynb
Lily S.
·
Thank you for your help!
Xander S.
·
Of course, let us know if you have further questions!

RunLLM
APP
·
To view actual data when hovering over individual embeddings in the 3D view, Phoenix provides tooltips that display relevant information about each data point. This feature was enhanced to improve the positioning and visibility of tooltips, as described in a <https://github.com/Arize-ai/phoenix/pull/890|GitHub pull request>. Ensure that your dataset is properly loaded with the necessary metadata to be displayed in the tooltips. For automatically assigning colors to different clusters, Phoenix allows you to color the UMAP point-cloud by various dimensions such as features, tags, predictions, and actuals. This can be done using the color by dimension feature, which maps points to distinct groups based on categorical or numeric values, as detailed in a <https://github.com/Arize-ai/phoenix/pull/423|GitHub pull request>. Would you like more detailed instructions on how to implement these features? Sources: - <https://github.com/Arize-ai/phoenix/pull/890|GitHub pull request on tooltips> - <https://github.com/Arize-ai/phoenix/pull/423|GitHub pull request>
💡 Hint: Mention RunLLM in the thread for followups.
Dustin N.
·
hi Lily S. thanks for using Phoenix. Lately we've been focusing on the LLM-observability functionality of Phoenix and embeddings visualization hasn't been prioritized lately. Unfortunately I'm not super familiar with this part of the product and can't be much help to you Xander S. do you know if we support the functionality Lily is looking for?
Xander S.
·
Hey Lily S., can you verify that your image URLs are accessible from your browser? It may also help to inspect the failed fetch requests in your brwoser's network tools. Here's an example notebook: https://colab.research.google.com/github/Arize-ai/phoenix/blob/main/tutorials/image_classification_tutorial.ipynb
Xander S.
·
I guess that is assuming you are trying to visualize images. Maybe you are dealing with text?
Lily S.
·
I have no images in my dataset.
Xander S.
·
It sounds like you're working with text data Lily S.?
Lily S.
·
Yes, just text data.
Xander S.
·
Might be worth taking a look at this tutorial if you haven't already: https://colab.research.google.com/github/Arize-ai/phoenix/blob/main/tutorials/sentiment_classification_tutorial.ipynb
Lily S.
·
Thank you for your help!
Xander S.
·
Of course, let us know if you have further questions!