From our side everything work ok i would say if we resume from an epoch : https://demoapp.demo.clear.ml/projects/5678e22221984581b089fd110c8db1ea/compare-experiments;ids=da10aacccc97459ea13df27b9cd44561,8276746945f6424781309513cee21cf8/scalars/graph?scalars=graph

Let me check that, thanks !
I was also started to think about closing the task and call init. Let me detail again the use-case if it was unclear:
parent process |-- Task.init(...) |-- step 1 sub-process | |-- Task.init(...) | |-- task.connect |-- step 2 sub-process | |-- Task.init(...) | |-- task.connect |- end

Thanks AgitatedDove14 !

Hi AgitatedDove14 , yes, exactly

according to https://allegro.ai/clearml/docs/docs/concepts_fundamentals/concepts_fundamentals_storage.html?highlight=output_uri#configuring-clearml-for-artifact-storage there is no way to do that ?

https://github.com/allegroai/clearml/issues/328

actually, i'm using subprocess. Popen and step X is another python script file executed from a single main parent process.

Thanks! Yes, this could be great !
And if you could also update the docs with all env vars possible to set up it would awesome!

Hi AgitatedDove14 thanks for the answer.
I'm using python doit: https://pydoit.org/ to schedule multiple steps. Basically, it does subprocess.Popen for each step. In each step (=subprocess) I get the task using Task.current_task() .
t0: doit -> parent proc creates new Task t1: |_ child proc: current_task() and do smth t2: |_ child proc: current_task() and do smth
My question is what happens if I launch in parallel multiple doit commands that create new Tasks. I would l...

Thanks AgitatedDove14 !

Could you please explain in 2-3 words how current task fetches the good Task ?

Thanks AgitatedDove14 ! This is what was I was looking !

Hi AgitatedDove14
if there is no clearml.conf on the system, then the script with Task.init("a", "b") will create project "a" and task "b" on demoapp server.
User can also do the following:
Task.set_credentials(...) Task.init("a", "b")and now it will go to provided api_host etc, still without clearml.conf file.
Now, I would like to raise an issue if api_host is the default one (the one when without clearml.conf file) and thus tracking is going to public demoapp server.

But actua...

Thanks a lot !

=>
Task.init(project_name='main_project/sub_project', task_name='test')https://clear.ml/docs/latest/docs/fundamentals/task#task-creation

Yes, that make sense !

Could you please open a GitHub issue, so we remember to update the feature ?

sure !

AgitatedDove14 I tried and it works OK with
Task.current_task() and overall Task.init .
Sorry for confusion.

Thanks UptightCoyote42 !

I think it is OK for me, thanks !

Well, I would like "quickly" to compare durations of several experiments ...

Updated column also can contain some more recent values and not the time when experiment ended/failed etc ...

Patching sys.argv[0] = __file__ where __file__ is the file where Task.init happens helps with git repo discovery.

It's code is like that
` (env-name) /code# cat /opt/conda/envs/env-name/bin/doit `` #!/opt/conda/envs/env-name/bin/python

-- coding: utf-8 --

import re
import sys

from doit.main import main

if name == 'main':
sys.argv[0] = re.sub(r'(-script.pyw?|.exe)?$', '', sys.argv[0])
sys.exit(main()) `

If using the launcher as is -> Can not detect the repository:
launcher argsbut if execute the launcher as a module -> repository is detected
python -m launcher argsIt would be nice to avoid writing "python -m" everytime...

It is python doit: https://pydoit.org/

As they didn't all started at the same time, I have to do math to subtract the duration 🙂