HTM for MNIST
These examples shows how to use an HTM network to classify images from the MNIST dataset.
Download and extract the MNIST data:
python -m nupic.vision.mnist.data.extract
This will download, extract, and preprocess training and test images. This MNIST dataset contains 60,000 training samples and 10,000 testing samples. By default, images are located in a directory called
datain your current working directory. Use the
--dataoption to specify a different data directory.
Train and test an HTM SP-only network:
python -m nupic.vision.mnist.run_mnist_experiment
This example achieves 95.56% accuracy on the 10,000 image training set as of 2015-04-13.
Numenta wishes to express its gratitude to Dr. Yann LeCun of the Courant Institute of Mathematical Sciences, New York, NY, for publishing the MNIST data set. Details regarding MNIST and related publications are available at the following URL: http://yann.lecun.com/exdb/mnist