Installation Guide

Quick Install

Requirements: Python 3.12 or higher, CUDA-capable GPU

BLISS is pip installable:

pip install bliss-toolkit

Install poetry:

curl -sSL https://install.python-poetry.org | python3 -
poetry config virtualenvs.in-project true

Install the fftw library (used by GalSim):
```
sudo apt-get install libfftw3-dev
```
Install git-lfs if you haven’t already:
```
git-lfs install
```

Clone the repo and install dependencies:

git clone git@github.com:prob-ml/bliss.git
cd bliss
poetry install

BLISS requires a CUDA-capable GPU for training. Check your CUDA version with:

nvidia-smi

BLISS uses PyTorch with CUDA 12.8 support, which works on both older GPUs (RTX 2080 Ti) and newer architectures.

BLISS requires the following core packages:

PyTorch (>=2.7.0): Deep learning framework
PyTorch Lightning (>=2.3.3): Training infrastructure
Hydra (>=1.0.4): Configuration management
GalSim (>=2.4.10): Galaxy image simulation
Astropy (>=6.1.1): Astronomical utilities
NumPy, SciPy, Matplotlib: Scientific computing

For specific use cases, you may need additional packages:

Weak Lensing Analysis:

pip install healpy

Galaxy Clustering:

pip install h5py pyarrow

Photo-z Estimation:

pip install scikit-learn

Visualization:

pip install plotly seaborn

TensorBoard Logging:

pip install tensorboard

All optional dependencies are included in the development installation (poetry install).

GalSim installation fails:

Ensure FFTW is installed:

sudo apt-get install libfftw3-dev  # Ubuntu/Debian
brew install fftw                   # macOS

CUDA not found:

Verify your CUDA installation:

nvcc --version
nvidia-smi

Memory errors during training:

Reduce batch size in configuration:

bliss mode=train cached_simulator.batch_size=8