A solution for the package dependency issues 📦

[SuhasSrinivasan]

As of 2024-04-26, the main branch has issues with unmet package dependencies, some examples below:

NVTX functions not installed

Cannot load Graphbolt C++ library

Missing packages such as torchdata, pandas and pydantic

These issues have been documented at least in #105, #104, #99, #94, #93, #37, #36 and #33.
create_env.sh is a unified solution for dependencies in RF2na-linux.yml and SE3Transformer/requirements.txt.

---

Developed and tested on
OS: Ubuntu 22.04.4 LTS
Nvidia Driver: 550.54.15
CUDA: 12.4
conda: 24.3.0

---

README has been updated to reflect the simplified package installation.

Documented are the conda YAML and conda table of packages.

Hope this is helpful.

[Ivyfeather]

Consider merge this into main?
As I also encountered #36 #93 #99 while using RF2na-linux.yml in main, and this new config solves my problems.

[SuhasSrinivasan]

Thank you for confirming @Ivyfeather.
It would be helpful if @fdimaio can please consider merging this.

[GavinKarsters]

Upon using this new yml file to create my environment, the pipeline still fails at this step:

ModuleNotFoundError: No module named 'torchdata.datapipes'

At the "Running RoseTTAFold2NA to predict structures" predict.py part. Any ideas? All solutions I try lead into a rabbit hole of torch/torchdata/numpy/dgl/graphbolt dependency issues

[sikai123]

Upon using this new yml file to create my environment, the pipeline still fails at this step:

ModuleNotFoundError: No module named 'torchdata.datapipes'

At the "Running RoseTTAFold2NA to predict structures" predict.py part. Any ideas? All solutions I try lead into a rabbit hole of torch/torchdata/numpy/dgl/graphbolt dependency issues

extra install using pip:
torchdata==0.7.0
numpy==1.26.4
torch==2.1.0

works for me