Hi there,
I have a question regarding the automatic requirements collection of clearml-agent. Currently I have two pip packages installed in the base-docker image. When I run my experiment I use: task.set_packages(packages="requirements.txt") to override the automatic collection of required packages. The agent is able to pick this up and run the experiment. After the experiment, the requirements of the run changed from my requirements.txt to the automatically collected packages on the agent. Is this behavior avoidable?
This leads to errors when this experiment is cloned and run because (e.g. numpy==1.2.6 -> numpy @ file:///numpy ) and this crashed during installation?