Introduction

• CNNumpy is a Convolutional Neural Network written in pure Numpy (educational purpose only).
• There are 2 implementation versions:
  • Slow: The naive version with nested for loops.
  • Fast: The im2col/col2im version.
• The slow implementation takes around 4 hours for 1 epoch where the fast implementation takes only 6 min for 1 epoch.
• For your information, with the same architecture using Pytorch, it will take around 1 min for 1 epoch.
• For more details, here are my blog posts explaining in depth what’s going on under the hood for each implementation (@machine-learning/blog-post-cnnumpy-slow">slow and @machine-learning/blog-post-cnnumpy-fast">fast).
• In the demo-cnnumpy-fast.ipynb notebook, the im2col/col2im implementation can achieve an accuracy up to 97.2% in 1 epoch (~6 min). Here are some results:

Installation

• Create a virtual environment in the root folder using virtualenv and activate it.

# On Linux terminal, using virtualenv.
virtualenv myenv
# Activate it.
source myenv/bin/activate

• Install requirements.txt.

pip install -r requirements.txt
# Tidy up the root folder.
python3 setup.py clean

Usage of demo notebooks

To play with the demo-notebooks/ files, you need to make sure jupyter notebook can select your virtual environnment as a kernel.

• Follow “Installation” instructions first and make sure your virtual environment is still activated.
• Run the following line in the terminal.

  python -m ipykernel install --user--name=myenv

• Run the notebook file only from demo_notebooks/ and then select Kernel > Switch Kernel > myenv. You are now ready to go !