Kohonen neural network implementation in Python.
- python >= 2.6
- numpy
- matplotlib
python main.py mnist_path valid_count test_count train_count network_width
- mnist_path: the cpickled mnist database
- valid_count: the size of the validation dataset (<= 10000)
- test_count: the size of the testing dataset (<= 10000)
- train_count: the size of the training dataset (<= 60000)
- network_width: the size of the neural network (between 4 and 30 is good)
For example:
python main.py data/mnist.pkl.gz 100 100 600 10
is a good starting point
More info on while1read.com.