search

mle-infrastructure / mle-toolbox

Lightweight Tool to Manage Distributed ML Experiments 🛠
https://mle-infrastructure.github.io/mle_toolbox/toolbox/
MIT License
3 stars 1 forks source link

Option to install environment/container at experiment startup #27

Open RobertTLange opened 3 years ago

RobertTLange commented 3 years ago

I would like to allow users to set an option in their experiment .yaml file so that in this case a new clean virtual environment is created and all dependencies are installed. E.g. something like install_venv_at_startup in the meta_job_args. At the beginning of the experiment the toolbox searches for a requirements.txt file in the root directory and creates either a clean venv or a conda environment before starting up the experiment. We also have to make sure that mle_toolbox is being installed but the user does not have to configure it since we only need the DeepLogger, get_configs_ready and set_random_seeds functionalities.

RobertTLange commented 3 years ago

Later on this should also be made compatible with docker/singularity containers. We only need to make sure that drive mounting and driver installations are somehow managed. Talk to @wllhf and Florian Blume about their slurm HPC setup.