Simulates and trains a network of neurons to learn the MNIST dataset. Also performs preprocessing of MNIST dataset.
Requirements:
- the
makebuild tool - gcc or any other C++ compiler
- openMP (included with gcc)
- Eigen (included in repo) ## TODO
- Boost (not included in repo)
Searches the hyperparameter space Requirements:
- python 3.7 or greater
- numpy
- nevergrad
- python-fire
matlab and python scripts for generating figures (TODO: convert matlab scripts to python and organize it all) Requirements:
- python 3.7 or greater
- numpy
- pandas
- matplotlib
- python-fire
Plotting the figures should only take a matter of seconds. To run a full training session on the MNIST dataset can take a while, up to an hour, but running on a subset can take about 10 minutes.
Compiling usually only takes a matter of seconds or a few minutes.
using python-fire
- to see the list of functions availible in a module, do
python file_name.py --help - to get help about a specific function/command within a module, do
python file_name.py COMMAND -- --help(the standalone--acts as a separator)
uses pdoc3, install using pip install pdoc3
to build documentation, run pdoc --html --force . -o docs
- small demo
- requirements
- all dependecies, including version numbers
- versions the software has been tested on
- installation guide
- makefile
- instructions
- typicall install time
- demo
- instructions to run
- expected output
- expected runtime
- general instructions for use
- requirements.txt for pip and python things
- list of where Eigen and whatever other C++ packages we use
- runtime estimates (done?)
- explain structure of the output directories (!)
- plotting scripts are already documented, refer to them in the readme
- instructions for the launching scripts
- liscence
- link to repo