Hi ShinyLobster84 ,
Plots are reported as plotly objects, so the immediate possible output is a response containing a list or a mapping of metric/variant to plotly JSON objects. If you feel there's a better output format, we can think about it. This is still not part of the Task methods, but we'll be happy to add it - we'll appreciate a GitHub issue with a description of your use-case and the required output.
Meanwhile, you can get the task's plots using the
APIClient (not the Task object), like so:
from clearml.backend_api.session.client import APIClient client = APIClient() res = client.events.get_task_plots(task="<task-id>")Note that the call returns the plots for a default number of last iterations (controlled by the
iters parameter) and may return a limited number of results (due to the total result size), in which case you should receive a scroll ID which you can use in subsequent calls using the
scroll_id parameter. For more info on this API call, see https://clear.ml/docs/latest/docs/references/api/events#post-eventsget_task_plots
Hi SuccessfulKoala55 CostlyOstrich36 ,
Thanks for your answers!
The code you write here was very helpful, I used it and succeeded to extract the relevant table from its output directly to my notebook:). There are cases that I want to investigate the outputs I saved in ClearML. In my case it was a table that I wanted to filter and get specific values from it, so this functionality helps me to analyze the results (I know I can download the json to my computer and then upload it back to my notebook but I looked for a direct solution. So I think an api is a good idea and can make it easier but this current solution works for me too:)