Getting Started
Set up OpenNN and understand the basic workflow before moving into the core classes.
Building OpenNN
Install and build OpenNN from source for a local C++ environment.
OpenNN in 6 steps
Walk through the essential workflow for creating and training a model.
The software model of OpenNN
Understand how the core objects fit together in an OpenNN project.
Main Classes
Work with the main OpenNN objects used to define data, networks, training, validation, and testing.
The data set class
Load, inspect, and transform data before training.
The neural network class
Define layers, parameters, and model structure.
The training strategy class
Configure loss functions, optimization, and stopping criteria.
The model selection class
Select inputs, complexity, and model settings with validation.
The testing analysis class
Evaluate trained models and interpret performance.
Performance & Configuration
Tune APIs, precision, CUDA execution, and multi-GPU training workflows.
Automated Mixed Precision
Use mixed precision to accelerate training and reduce memory use.
Configuring the Device and Precision
Choose CPU/GPU devices and numeric precision.
CUDA Graphs Explained
Reduce GPU overhead with reusable CUDA execution graphs.
Multi-GPU Data Parallelization
Distribute batches across GPUs for faster training.
Multi-GPU Model Parallelization
Split large models across GPUs when one device is not enough.
Machine Learning Examples
Apply OpenNN to approximation, classification, forecasting, and text classification problems.
Approximation: Airfoil self-noise prediction
Build a regression model for engineering noise prediction.
Classification: Breast cancer diagnosis
Train a classification model for medical diagnosis data.
Forecasting: Airline passengers estimation
Estimate future values from time-ordered passenger data.
Text Classification: Amazon reviews classification
Classify review text with a neural network model.
Need the full C++ reference?
Use the API documentation when you need class-level details, methods, and implementation references.