Skip to content

FAQ

Error messages and their fixes are under Troubleshooting.

Do I install the environments again next time?

No. mtb.env.install keeps each environment on disk, and later sessions use it: mtb.run finds it, and mtb.env.install skips it. Only Google Colab starts empty, because Colab deletes the disk with the runtime.

Where are the environments?

In mtb.config.envs_dir, one folder per environment:

import multibench as mtb
print(mtb.config.envs_dir)
mtb.env.install(["scJoint"], dry_run=True)   # its environment, and "have" once installed

Method environments says which folder envs_dir is.

Can I use an environment outside multibench?

Yes. Each one is an ordinary conda environment. Activate it by its path, for example to run a method's own code or your own script with the same packages:

conda activate /path/to/envs_dir/scmb_scjoint
python my_script.py
conda deactivate

When envs_dir is your conda's envs folder, the name works too: conda activate scmb_scjoint.

Install extra packages in an environment of your own, not in these: a changed environment can change the methods' results.

Can several people share one set of environments?

Yes. Install them once into a shared folder, and give everyone the same setting before they import multibench:

export MULTIBENCH_ENVS_DIR=/shared/multibench_envs

The person who installs needs write access to the folder.

How do I remove an environment?

Delete its folder in envs_dir. The .layers folder there holds files that several PyTorch environments share; delete it only with the last of them.