/backprop

Backpropagation from scratch

Primary LanguagePython

toolkitPython

Python port of BYU CS 478 machine learning toolkit

Works with Python 2.7 or 3. Requires NumPy.

Usage

In order to use this toolkit, most commands will be similar to those given on the class website for the Java and C++ toolkits. With the assumption that you already have NumPy installed (see their website for installation instructions), usage is straight-forward.

As example, execute the following commands from the root directory of this repository.

mkdir datasets
wget http://axon.cs.byu.edu/~martinez/classes/478/stuff/iris.arff -P datasets/
python -m toolkit.manager -L baseline -A datasets/iris.arff -E training

Aside from needing to specify the module to run, commands follow the same syntax as the other toolkits.

For information on the expected syntax, run

python -m toolkit.manager --help

Creating Learners

See the baseline_learner.py and its BaselineLearner class for an example of the format of the learner. In particular, new learners will need to override the train() and predict() functions of the SupervisedLearner base class.

Backprop

FML