# README
ConvNet 
This is an example of a 3-layer ConvNet built with Gorgonia. It uses the MNIST data as an example. The MNIST data is not provided, and MUST be put in ../testdata
.
This example currently works but is very slow, pending a change in transpose algorithms.
A step by step tutorial is exposed on to the Gorgonia website: https://gorgonia.org/tutorials/mnist/
Disclaimer
This example comes with a basic implementation of a ConvNet. It does not contain:
- Model serialization (you cannot save your models)
- Checkpointing.
- Multithreaded batching (training is highly serial).