To start running distance metrics, please use main.py. Its help should contain all you need to undertand how to perform the experiments on the data.
python main.py --help
To run the various methods use:
python main.py method [--rank RANK] [-cmc] [--title TITLE] [method_options]
Note that there may be other mandatory arguments needed for the various methods used. The options are shown below alongside the method names. The syntax is: [--optional_argument [arg_values]], --mandatory_argument [mandatory_arg_values], mandatory_positional_argument.
Methods attempted:
- Baseline Euclidean Distance:
euclid - Mahalanobis (various methods):
mahala [--model [mmc, lmnn, nca, lfda, mlkr]] - k-means clustering:
kmeans --clusters [num_clusters] --iterations [max_iterations] - kernel methods:
kernel --kernel [cosine, poly, laplacian, chi2] - neural network:
neuralnet --model [mlp, mlp-conv] - transformer methods:
transformer --transformer [pca, quantile, normalizer, standard_scaler, gauss_rand] [--components n_components] - distance methods:
distance --kernel [cosine, poly, laplacian, chi2]
You can use --rank to reduce the required rank measurements, which will reduce computational complexity where possible.
You can optionally plot a graph of the results using the option -cmc (cumulative match curve). Its title can be specified with --title.