LightGBM Installation & Usage

Install and configure LightGBM, a high-performance gradient boosting framework that supports CPU, GPU, CUDA, and distributed training.

Overview

LightGBM is a gradient boosting framework that uses tree-based learning algorithms. This skill guides you through installation options ranging from simple PyPI installs to custom builds with GPU, CUDA, or MPI support.

Installation Options

1. Standard Installation (PyPI)

Install the standard CPU/GPU-ready version:

```bash

pip install lightgbm

```

**Platform-specific requirements:**

2. Installation with Optional Dependencies

Install with framework-specific extras:

```bash

For Dask distributed computing (Linux only)

pip install 'lightgbm[dask]'

For pandas integration

pip install 'lightgbm[pandas]'

For scikit-learn integration

pip install 'lightgbm[scikit-learn]'

```

3. Build from Source

For custom configurations or unsupported platforms:

```bash

pip install --no-binary lightgbm lightgbm

```

**macOS users:** Choose Apple Clang (install OpenMP first) or gcc (set compilers: `export CXX=g++-14 CC=gcc-14`)

**Windows users:** Requires Visual Studio or VS Build Tools

4. GPU Version

Build with OpenCL GPU support:

```bash

pip install lightgbm \

--config-settings=cmake.define.USE_GPU=ON \

--config-settings=cmake.define.OpenCL_INCLUDE_DIR="/usr/local/cuda/include/" \

--config-settings=cmake.define.OpenCL_LIBRARY="/usr/local/cuda/lib64/libOpenCL.so"

```

**Requirements:**

5. CUDA Version

Build with CUDA support for NVIDIA GPUs:

```bash

pip install lightgbm --config-settings=cmake.define.USE_CUDA=ON

```

**Usage:** Pass `{"device": "cuda"}` in parameters when training

6. MPI Version (Distributed Training)

Build with MPI for multi-node training:

```bash

pip install lightgbm --config-settings=cmake.define.USE_MPI=ON

```

**Requirements:** MPI libraries (OpenMPI, MPICH, or MS-MPI)

7. Conda Installation

Install precompiled packages from conda-forge:

```bash

Standard installation

conda install -c conda-forge lightgbm

Auto-select CUDA build on CUDA-enabled systems (v4.4.0+)

conda install -c conda-forge 'lightgbm>=4.4.0'

```

**Best for:** Systems without wheels (PowerPC), GPU users, conda environment management

8. GitHub Development Version

Install bleeding-edge version:

```bash

git clone --recursive https://github.com/microsoft/LightGBM.git

cd LightGBM

sh ./build-python.sh install

Variations:

sh ./build-python.sh install --nomp # Disable OpenMP

sh ./build-python.sh install --mpi # Enable MPI

sh ./build-python.sh install --gpu # Enable GPU

sh ./build-python.sh install --cuda # Enable CUDA

sh ./build-python.sh install --mingw # Use MinGW on Windows

```

Verification

Test your installation:

```python

import lightgbm as lgb

print(lgb.__version__)

Check GPU availability (if built with GPU support)

try:

lgb.LGBMClassifier(device='gpu')

print("GPU support available")

except:

print("GPU support not available")

```

Common Issues

1. **32-bit Python**: Not supported. Use 64-bit Python.

2. **Missing OpenMP**: Install `libgomp1` (Linux), `libomp` (macOS), or VC runtime (Windows)

3. **GPU not detected**: Install OpenCL runtime or check drivers

4. **Windows compilation**: Use Visual Studio (better multithreading) over MinGW-w64

Advanced CMake Options

Available via `--config-settings=cmake.define.{option}`:

Constraints

Next Steps

After installation, explore: