Is there a way to create projects/namespaces for Datasets?
Is there a way to create projects/namespaces for Datasets?
Is there a way to create projects/namespaces for Datasets?
Yes, Phoenix supports creating projects or namespaces for datasets. You can create multiple instances of the Fixture class, where each instance represents a project with its own datasets, schemas, and related items. This allows you to organize datasets under different projects or namespaces. Additionally, the log_traces method on the Phoenix client allows you to specify a project_name when loading a TraceDataset, which helps in organizing datasets under specific projects.Would you like more detail on how to implement this? Sources:
- https://github.com/Arize-ai/phoenix/issues/2335
- https://github.com/Arize-ai/phoenix/issues/2517
💡 Hint: Mention RunLLM in the thread for followups.
If you tell me the right answer to this question, I'll make sure I don't make the same mistake twice!
Feedback
Type the right answer here...
↵ Please 'enter' to submit
Hey Samantha Z., we don't currently have affordances for namespacing datasets. Curious to know how you're hoping to use those namespaces?
If you tell me the right answer to this question, I'll make sure I don't make the same mistake twice!
Feedback
Type the right answer here...
↵ Please 'enter' to submit
Xander S. RunLLM seems to think I can use a Fixture, is that not correct? We're sharing an instance of Phoenix with another team, working on separate projects. I'd like our datasets to be more easily organized, and findable. They have a huge list of files right now and adding ours in there isn't going to make it any easier to tell what's what.
It also looks like you still only support CSV for datasets, is that correct? And do you require an expected output column? Our stuff is all json, so I'll have to convert it.
We're sharing an instance of Phoenix with another team, working on separate projects. I'd like our datasets to be more easily organized, and findable. They have a huge list of files right now and adding ours in there isn't going to make it any easier to tell what's what.
That makes a lot of sense. I've filed an issue here.
RunLLM seems to think I can use a Fixture, is that not correct?
This is a hallucination unfortunately.
Filed a ticket for JSON dataset upload: https://github.com/Arize-ai/phoenix/issues/7264
I thought you already had one from a while ago, but I can't remember now.
Pretty sure we asked about this when we first started using Phoenix.
Look like this one https://github.com/Arize-ai/phoenix/issues/6453
Yes, Hard to believe that was already a couple of months ago. Doesn't look like it made it onto your roadmap yet, though.
Definitely feel your pain on this one. We have the issue marked as part of a dataset and experiment enhancements milestone, but it hasn't been scheduled yet.