Suppose I’d like to download the history for a group of runs and do some minor analysis. I can already do this using wandb.api. How can I then upload the parsed data as a wandb.Table (or csv) such that it can be visualized / shared as part of the original experiment or in a report?
Hi there, I’m not an expert in this, but the way I’d do it is probably to create a separate run that is tagged as an evaluation run, use the artifact of the training run in the evaluation so that there is lineage between the training and the evaluation. Then you can upload whatever tables you want to upload.
Hi @jeromeku-test! Thank you for writing in.
Currently, you have a couple of options of what can be done to add data to a run after it has been finished.
- The recommended way - Go ahead and resume your run, and upload data to the run after it has been resumed. Then finish the run again.
- Update the summary metrics via the API, and then display the data using a wandb table.
- Exactly what @rcasburn has suggested. Go ahead and create a separate run that is ragged as an eval run, and then group it together with the first run in order to have them together to display data together.
Warmly,
Artsiom
Hi JJ,
We wanted to follow up with you regarding your support request as we have not heard back from you. Please let us know if we can be of further assistance or if your issue has been resolved.
Best,
Weights & Biases
Hi, since we have not heard back from you, we are going to close this request. If you would like to reopen the conversation, please let us know! Unfortunately, at the moment, we do not receive notifications if a thread reopens on Discourse. So, please feel free to create a new ticket regarding your concern if you’d like to continue the conversation.