mixturs

Unofficial implementation of Dirichlet Process Mixture Model Sub-Clusters algorithm.

Features • Usage • Examples

To use as a library, add the following to your Cargo.toml. Executable builds can be found at https://github.com/EgorDm/mixturs/releases.

[dependencies]
mixturs = "0.1"

Features

• Cluster points without knowing the number of clusters in advance
• Fastest CPU implementation of the algorithm
• Python bindings to cluster numpy data
• Command line tool for generating segmented images from JPG/PNG input files

Examples

Rust Examples:

// Load data into a col major matrix
let x: DMatrix<f64> = ...;

// Set model options
let mut model_options = ModelOptions::<NIW>::default(dim);
model_options.alpha = 100.0;
model_options.outlier = None;

// Set fit options
let mut fit_options = FitOptions::default();
fit_options.init_clusters = 1;

// Configure callbacks
let mut callback = MonitoringCallback::from_data(
    EvalData::from_sample(&x, Some(&y), 1000)
);
callback.add_metric(AIC);
callback.add_callback(PlotCallback::new(
    3,
    "examples/data/plot/synthetic_2d".into(),
    EvalData::from_sample(&x, None, 1000)
));
callback.set_verbose(true);

// Fit the model
let mut model = Model::from_options(model_options);
model.fit(
    x.clone_owned(),
    &fit_options,
    Some(callback),
);

Python Examples:

import numpy as np
from mixtupy import *

# Load data
x = ...

# Configure model
mo = ModelOptions(2)
model = Model(mo)

# Fit model
fo = FitOptions()
fo.iters = 200
fo.aic = True
model.fit(x, fo, y=y)

# Predict data point labels
probs, labels = model.predict(x)

# Extract cluster parameters
print(model.cluster_weights())
print(model.cluster_params())

Reference

[1] J. Chang and J. W. Fisher III, “Parallel Sampling of DP Mixture Models using Sub-Cluster Splits,” in Advances in Neural Information Processing Systems, 2013.

[2] O. Dinari, A. Yu, O. Freifeld, and J. Fisher, “Distributed MCMC Inference in Dirichlet Process Mixture Models Using Julia,” in 2019 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID).