Hello, I ran two sweeps between which I changed a single parameter which isn’t in the actual sweep parameters.
Would It be possible to add a tag or property to the sweeps so I can compare my different sweeps on the same graph to see the influence of this particular property ?
Thanks !
Hi @hugo-aoyagi - thanks for writing in your question.
I’d like to better understand your use case to provide further guidance, would you mind sharing some additional information.
Are you looking to compare the aggregate results of Runs from different Sweep and compare them?
If this is the case, you could use the Group function in the Project workspace and use the Sweep field to Group runs and you will be able to visualise Metrics grouped for all the Runs in each sweep
If instead, you are looking to compare Runs results based on a specific parameter, regardless of the Sweep they have been logged with, you can use the grouping function with that parameters (see the example below with learning_rate).
You could also Group by Sweep and a parameter and vice-versa.
Let me know if you have any further questions on this, or if you are happy to share the URL for the project you are looking at and provide additional guidance on your use case we can keep looking at a different approach for your use case.
Thanks you for this very helpful answer.
What I meant by my original post was :
Let’s say I perform a sweep on three different datasets or with 3 different models but I forgot to add the model type or dataset type as a property in the sweep.
Could I then still show a hyper-parameter graph with model or dataset type as a new column to see it’s impact on the accuracy.
No problem - yes, that is still possible.
As long as the Configuration value (the model, or dataset) has been logged as config with the Run (or added afterwards using the api, see here), you can edit the Parallel coordinate panel and add any metric / configuration as column - as in the recording below:
Hi @hugo-aoyagi, I wanted to follow up on this request. Please let us know if we can be of further assistance
Hi @hugo-aoyagi , since we have not heard back from you we are going to close this request. If you would like to re-open the conversation, please let us know!
