README.txt

A simple convolutional neural network written in C++, using the boost::python library to make it accessible through Python. 

This implementation is rather slow compared with TensorFLow or PyTorch versions, and lacks modern features such as batch normalization, network shortcuts and learning rate optimizer. It is intended at showing how simple convolutional neural networks work on a (nearly) minimal example rather than real-world applications. 

The Python file CNN3_test.py shows a simple example. The notebook CNN3.ipynb shows the results of some numerical experiments.

Requirements: 
    * a C++ compiler (tested with gcc 7.5.0),
    * the boost library,
    * a Python 3 development environment (tested with Anaconda 4.8.5). 

The library CNN3.so can be compiled on Linux using the script make.sh. The string “~/miniconda3/include/python3.8” may need to be replaced by the path to your Python include libraries.