Installation Guide¶
Quick Start¶
Create conda environment (recommended):
conda create --name vbi python=3.10
conda activate vbi
Install VBI:
pip install vbi # Light version (CPU only)
pip install vbi[light-gpu] # Light + Cupy
pip install vbi[inference] # With (sbi, PyTorch)
pip install vbi[all] # Full (sbi, PyTorch, Cupy)
Installation Options¶
Command |
Includes |
Best For |
|---|---|---|
|
CPU simulation, feature extraction, CDE-based inference |
Avoiding heavy dependencies |
|
Everything + Cupy |
GPU simulation |
|
Everything + PyTorch, SBI |
Parameter inference (CPU) |
|
Everything + GPU acceleration |
Full functionality with GPU |
Docker Usage¶
# Quick start with pre-built image
docker run --rm -it -p 8888:8888 ghcr.io/ins-amu/vbi:main
# With GPU support
docker run --gpus all --rm -it -p 8888:8888 ghcr.io/ins-amu/vbi:main
For Docker building and advanced usage, see Building Docker Image and VBI Docker Quick Start.
Installation From Source¶
git clone https://github.com/ins-amu/vbi.git
cd vbi
pip install .
For development:
pip install -e .[all]
Platform-Specific Instructions¶
Google Colab
Google Colab doesn’t have VBI or SBI pre-installed, and Docker is not supported in Colab due to security restrictions. For optimal C++ module compilation, install from source:
# In a Colab cell, run:
!mkdir -p src && cd src
!git clone --depth 1 https://github.com/ins-amu/vbi.git
%cd src/vbi
!pip install -e .
Alternative: Use Colab Pro+ with Custom Runtimes
If you have Colab Pro+ and need a containerized environment, consider:
Using Kaggle Notebooks (supports Docker-based custom environments)
Using Binder with our repository (though with limited resources)
Setting up a local Jupyter server with our Docker image and connecting via ngrok
Note: The environment will be reset when the Colab runtime shuts down. You’ll need to reinstall for each new session.
EBRAINS Collab
EBRAINS has dependency management restrictions. Here’s a script to create a dedicated VBI environment:
#!/bin/bash
# Save this as setup_vbi_ebrains.sh
set -eux
# Create fresh environment
rm -rf /tmp/vbi
python3 -m venv /tmp/vbi
unset PYTHONPATH
source /tmp/vbi/bin/activate
# Install core dependencies
pip install ipykernel scikit_learn matplotlib
# Install PyTorch (CPU version to save space)
pip install torch --index-url https://download.pytorch.org/whl/cpu
# Install SBI without dependencies to avoid reinstalling large packages
pip install sbi --no-deps
# Install SBI dependencies manually
pip install pyro-ppl tensorboard nflows pyknos zuko arviz pymc
# Install VBI from source
mkdir -p /tmp/src && pushd /tmp/src
git clone --depth 1 https://github.com/ins-amu/vbi.git
cd vbi
pip install -e .
popd
# Create Jupyter kernel
python -m ipykernel install --user --name VBI
echo "VBI environment created! Please reload your browser and select the 'VBI' kernel."
echo "Note: This environment will be lost when the lab server shuts down."
Make the script executable and run it:
chmod +x setup_vbi_ebrains.sh
./setup_vbi_ebrains.sh
Important Notes:
Both environments are temporary and will be reset when the respective platforms shut down
For EBRAINS, you’ll need to rerun the setup script for each new session
For Colab, you’ll need to reinstall VBI in each new runtime
Windows
Windows installation is automatic - C++ compilation is automatically skipped:
pip install vbi
Verification¶
import vbi
vbi.tests()
vbi.test_imports()
Troubleshooting¶
C++ Compilation Issues
Note: the package is configured with SKIP_CPP=0 by default (C++ extensions are enabled). If you want to skip compilation of C++ components, set SKIP_CPP=1 when installing from source or via pip, for example:
export SKIP_CPP=1
pip install vbi
Common Issues
ImportError: Check Python version (3.10+ recommended)
CUDA issues: Verify GPU drivers and CUDA compatibility
Memory errors: Try lighter installation variants
For detailed troubleshooting, platform guides, and advanced scenarios, see the complete Installation Guide.