README.md

MIOFlow

Setup

Create conda environment

$ conda create -n mioflow python=3.10
$ conda activate mioflow

install pytorch according to instructions https://pytorch.org/get-started/

we used

$ conda install pytorch pytorch-cuda=11.6 -c pytorch -c nvidia

install requirements and MIOFlow

$ pip install -e .

Remove unused packages and caches.

$ conda clean --all

Install MIOFlow for developers and internal use:

$ cd path/to/this/repository
$ pip install -e MIOFlow

Add kernel to Jupyter Notebook

automatic conda kernels

For greater detail see the official docs for nb_conda_kernels. In short, install nb_conda_kernels in the environment from which you launch JupyterLab / Jupyter Notebooks from (e.g. base) via:

$ conda install -n <notebook_env> nb_conda_kernels

to add a new or exist conda environment to Jupyter simply install ipykernel into that conda environment e.g.

$ conda install -n <python_env> ipykernel

manual ipykernel

add to your Jupyter Notebook kernels via

$ python -m ipykernel install --user --name sklab-mioflow

It can be removed via:

$ jupyter kernelspec uninstall sklab-mioflow

list kernels found by Jupyter

kernels recognized by conda

$ python -m nb_conda_kernels list

check which kernels are discovered by Jupyter:

$ jupyter kernelspec list

How to use

This repository consists of our python library MIOFlow as well as a directory of scripts for running and using it.

Scripts

To recreate our results with MMD loss and density regulariazation you can run the following command:

python scripts/run.py -d petals -c mmd -n petal-mmd

This will generate the directory results/petals-mmd and save everything there.

For a full list of parameters try running:

python scripts/run.py --help

Python Package

One could simply import everything and use it piecemeal:

from MIOFlow.ode import *
from MIOFlow.losses import *
from MIOFlow.utils import *
from MIOFlow.models import *
from MIOFlow.plots import *
from MIOFlow.train import *
from MIOFlow.constants import *
from MIOFlow.datasets import *
from MIOFlow.exp import *
from MIOFlow.geo import *
from MIOFlow.eval import *

Tutorials

One can also consult or modify the tutorial notebooks for their uses: - EB Bodies tutorial - Dyngen tutorial - Petals tutorial