Skip to content

installation.md

Latest commit

 

History

History
120 lines (83 loc) · 3.32 KB

installation.md

File metadata and controls

120 lines (83 loc) · 3.32 KB
Raw

Installation

CheMLFlow is usually installed from a local source checkout. We recommend using a fresh conda or mamba environment so compiled scientific packages, RDKit, and optional deep-learning dependencies do not collide with other projects.

The commands below assume Python 3.12 and should be run from the CheMLFlow/ repository root.

Option 1. Standard source install

Clone the repository and create an environment:

git clone https://github.com/nijamudheen/CheMLFlow.git
cd CheMLFlow

conda create -n chemlflow_env python=3.12
conda activate chemlflow_env

Install compiled dependencies from conda-forge first:

conda install -c conda-forge \
  numpy scipy scikit-learn matplotlib-base seaborn \
  lightgbm xgboost catboost rdkit shap numba llvmlite

Then install the Python requirements without asking pip to re-resolve the compiled stack:

pip install -r requirements.txt --no-deps

This is the most reliable path on macOS and works well on Linux clusters where compiled packages are easier to manage through conda.

Option 2. Editable developer install

Use this when you plan to modify CheMLFlow code and import it as a package:

git clone https://github.com/nijamudheen/CheMLFlow.git
cd CheMLFlow

conda create -n chemlflow_dev python=3.12
conda activate chemlflow_dev

conda install -c conda-forge \
  numpy scipy scikit-learn matplotlib-base seaborn \
  lightgbm xgboost catboost rdkit shap numba llvmlite

pip install -r requirements.txt --no-deps
pip install -e . --no-deps

For tests:

pip install pytest

Optional. Deep-learning models

Install PyTorch if you want to run neural network model families such as dl_simple, dl_deep, dl_gru, dl_resmlp, dl_tabtransformer, or dl_aereg.

For CUDA systems, choose the PyTorch index URL that matches your driver/CUDA stack. For example:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install pytorch-lightning

For CPU-only use:

pip install torch torchvision torchaudio
pip install pytorch-lightning

For Apple Silicon or other CPU-only systems, install the standard PyTorch wheels:

pip install torch torchvision torchaudio
pip install pytorch-lightning

Optional. TDC datasets

Install pytdc only if you need Therapeutic Data Commons datasets, such as the PGP benchmark export workflow:

pip install pytdc
python -c "from tdc.benchmark_group import admet_group; print('pytdc ok')"

If pytdc pulls in conflicting dependencies, use a small separate export environment and write the dataset to CSV before running CheMLFlow in the main environment.

Check the install

From the repository root:

python -m MLModels.training.cli --help
python scripts/generate_doe.py --help

For a short train-and-predict workflow, continue with the Quickstart.

Troubleshooting

  • Run commands from the repo root so relative paths resolve.
  • If conda activate is unavailable, run conda init zsh and restart the shell.
  • If NumPy, RDKit, SHAP, numba, or llvmlite report binary compatibility errors, reinstall those packages from conda-forge in the active environment.
  • CatBoost is stable on Python 3.12; Python 3.13 is not recommended for this project yet.
  • pytdc is optional and is not installed by requirements.txt.