A MLP implementation on multicore processors with Cuda.

The classic IRIS dataset is used to test and verify the MLP algorithm.

Dataset is downloaded from: https://archive.ics.uci.edu/ml/datasets/iris

All files are unzipped to _data directory.

• Computational intensive
• Efficient training and fitting
• Easily portable and expandable
• Lightweight and friendly program interface

• CUDA or CPU parallel?
• API design
• Algebra calculation module
• Data feeding module
• MLP training module
• Unit test

• Reinforcement learning application (e.g. DQN)
• Optimize IO with coroutine in C++20.

This code is released under the Apache-2.0 License.

Copyright (C) 2020
Xintao Hu <[email protected]>