Supported Platforms

Environments

Platform	Architecture	Runtime Requirements	Supported OS Examples
Microsoft Windows	x86_64	v142	Windows 11 Windows Server 2022+
GNU/Linux	x86_64	glibc 2.35 libstdc++.so.6.0.28	Ubuntu 22.04 LTS Debian 12+
	arm64-v8a	Coming soon	Coming soon
Apple macOS	Coming soon

Prerequisites

All versions:

Technology	Minimum version
Miniconda/Anaconda	Latest
Python	3.9–3.12 (to run the installer; the conda environment will use Python 3.9)

Note

Conda 25+ users (first-time setup): Starting with Conda version 25, you will be prompted to accept the Anaconda Terms of Service the first time you run a conda command. This is normal and only needs to be done once. Simply follow the prompt in the command line to accept the terms and continue.

GPU version additional requirements:

Technology	Minimum version
NVIDIA GPU	(compute capability 3.5+)
NVIDIA driver	515.48 or newer
CUDA toolkit	11.7 (only if nvcc is needed; runtime libraries are installed in conda env)

Note

The GPU requirements above apply to both Windows and Linux. The NVIDIA driver (515.48+) is the hard requirement for GPU support on both platforms. The CUDA toolkit is only needed if your workflow requires nvcc or other toolkit binaries. On Linux, the installer automatically uses environment_gpu_linux.yml when available.

Important

Newer NVIDIA GPUs: The version of PyTorch bundled with HOOPS AI does not yet support GPUs newer than the NVIDIA RTX 3000 series (e.g., RTX 4000, RTX 5000, RTX PRO series). If you have one of these GPUs, HOOPS AI will still work in CPU mode. GPU support for newer architectures is planned for an upcoming release.

Package Description

Here is a quick description of the package :

hoops_ai
├───cadaccess
│   ├─── HOOPSLoader
│   ├─── HOOPSModel
│   ├─── HOOPSBrep
│   └─── HOOPSTools
├───cadencoder
│   └─── BrepEncoder
├───dataset
│   ├─── DatasetLoader
│   └─── DatasetExplorer
├───flowmanager
│   ├─── @flowtask Decorators
│   ├─── ParallelTask
│   ├─── SequentialTask
│   └─── ParallelExecutor
├───insights
│   ├─── DatasetViewer
│   ├─── CADViewer
│   └─── ColorPalette
├───ml
│   └───EXPERIMENTAL
│       ├─── FlowModel
│       ├─── FlowTrainer
│       └─── FlowInference
└───storage
     ├─── OptStorage
     ├─── MemoryStorage
     ├─── JsonStorageHandler
     ├─── SchemaBuilder
     └─── DatasetMerger

Module Documentation

The hoops_ai package is organized into specialized modules that work together to transform CAD files into machine learning-ready datasets. For detailed documentation on each module, see the Programming Guide:

Data Flow Management:

• CAD Data Access - CAD file loading with HOOPSLoader, HOOPSModel, HOOPSBrep, and HOOPSTools

• CAD Data Encoding - Feature extraction with BrepEncoder

• Datasets - ML-Ready Inputs - Schema definition with SchemaBuilder

• Data Storage - Data persistence with OptStorage, MemoryStorage, JsonStorageHandler, and DatasetMerger

• Data Flow Customisation - Pipeline orchestration with @flowtask decorators, ParallelTask, SequentialTask, and ParallelExecutor

Machine Learning:

• Dataset Exploration and Mining - Dataset exploration with DatasetExplorer and DatasetLoader

• Develop Your own ML Model - Model training with FlowModel, FlowTrainer, and FlowInference (EXPERIMENTAL)

• Parts Classification Model - Part classification examples

• CAD Feature Recognition Model - Feature recognition examples

Visualization:

• Data Visualization Experience - Interactive visualization with DatasetViewer, CADViewer, and ColorPalette

Next Steps

If you are already familiar with the Supported Formats page, you are now ready to discover how to Get Started with HOOPS AI.