I believe pipeline_from_tasks.py used to be called pipeline_controller.py

Windows 10, 3.9, using conda but even pip list should output something

No just a dictionary from the pipeline examples:
https://github.com/allegroai/clearml/blob/b010f775bdd72ba6729f5e1e569626692d7b18af/examples/pipeline/step3_train_model.py#L13-L17

I believe this also impacts python 3.8 + windows btw

I'll be your n00b w1nd0ws beta tester lol

Could it be because some of the packages then point to files?
pyarrow==5.0.0
pandas @ file:///D:/bld/pandas_1624391191530/work

I am using Task.force_requirements_env_freeze() (before Task.init) with no requirements file specified (because I do not have one), and my experiments are showing this under packages:

Dang that's not good news for me

I'm trying to accomplish that but instead of a pip freeze I just get a "Packages will be installed from projects requirements file" even though there is no projects requirements file. Do I need to do something with development.detect_with_pip_freeze ?

Yes

` from clearml import Task

args = {'hello': 'world'}

Task.force_requirements_env_freeze()
task = Task.init(project_name='example', task_name="Test")
task.connect(args)
task.execute_remotely() `

Ah interesting, that makes sense thanks

Yes I just did pip list in the env and it worked fine

1.1.1

You can see installed packages with that script?

Basically the file URI might be different on a different machine (out of my control) but they point to the same artifact storage location

I think I'll just switch to docker and mount

ah that is perfect, thank you

Thanks

Okay, thanks for looking into it

I do see that once a task fails the installed package list is updated with a full list of the environment... I don't know if that helps at all

Okay, definitely feel better knowing I am not crazy 👍