Code for the AAAI 2014 paper: Manifold Spanning Graphs
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

Manifold Spanning Graphs

Code for the paper:

Carey and Mahadevan, Manifold Spanning Graphs, AAAI-2014: Twenty-Eighth Conference on Artificial Intelligence. July, 2014.

Setup Instructions

Ensure that stable versions of the following Python 2 libraries are installed:

Optionally, install bottleneck for faster k-nearest neighbor search.

Next, download and build BMatchingSolver. This is required for the b-matching parts of the experiments. Once built, edit the BMatchingSolver_PATH in to reflect the path to the binary you built.

Finally, download the MNIST data. The paper used the t10k set, which is available on Yann LeCun's site. To convert the data into .npy files for use in, run the following code:

cd data/
gunzip t10k-*-ubyte.gz
python -c 'import numpy as np; x=np.fromfile("t10k-images-idx3-ubyte", dtype=np.uint8);"test_data", x[16:].reshape((10000,28,28),order="C"))'
python -c 'import numpy as np; x=np.fromfile("t10k-labels-idx1-ubyte", dtype=np.uint8);"test_labels", x[8:])'

Running Experiments

Run the experiments to generate the figures:


Both files have some options, which you can see by passing them the --help flag.