[][src]Crate hotsax

This crate contains an implementation of the HOT SAX algorithm, and the brute force algorithm, as proposed by Keogh et al.. It also includes the HS-Squeezer algorithm, since it offers useful optimizations, while still being heavily based on the HOT SAX algorithm. However, from testing, the performance of this implementation is similar or worse than HOT SAX. If you find any possible optimizations, please open an issue.

During the implementation some other functions had to be made, such as paa, znorm, and gaussian. These functions are exposed, due to their utility apart from being used in HOT SAX.

The code is well commented in order to explain the implementation, in the case that people want to learn how the HOT SAX algorithm works by looking at an implementation. If a part is vaguely commented, feel free to leave an issue.

Note that only Float vectors are supported. If your data is made up of integers, you need to convert it to float first.

Example of use

This example is not tested
use std::error::Error;
use plotly::{Plot, Scatter};

// Parses the CSV file from the dataset.
let mut rdr = csv::ReaderBuilder::new()

// Deserialize CSV data into a vector of floats.
let mut data : Vec<f64> = Vec::new();
for result in rdr.deserialize() {

// Prepare a plot
let mut plot = Plot::new();

// Retrieve the largest discord. This should approx. match the one found in the paper.
// It uses the same settings: a discord size of 256 and a=3.
// word_size was assumed to be 3.
let discord_size = 256;
let discord = hotsax::Anomaly::with(&data, discord_size)
    .use_slice(1000..)      // Skips the beginning due to an abnormality.
    .find_largest_discord() // Finds the largest discord in the subslice.
    .unwrap().1;            // Only gets the location.

// Plot the entire dataset as a blue color.
let trace1 = Scatter::new((1..=data.len()).collect(), data.clone())

// Plot the discord itself as a red color.
let trace2 = Scatter::new((discord+1..discord+discord_size+1).collect(), data[discord..discord+128].to_vec())

// Add traces to the plot.

// Shows the plot to verify.


pub use anomaly::Anomaly;
pub use dim_reduction::paa;
pub use dim_reduction::sax;
pub use util::gaussian;
pub use util::znorm;
pub use util::mean;
pub use util::std_dev;
pub use squeezer::squeezer;
pub use anomaly::Algorithm;



Implements anomaly detection algorithms, including the brute force and HOT SAX algorithms as specified by Keogh's paper, found here.


Dimensionality reduction techniques.


Clustering functions and squeezer impl


Miscellaneous utility functions.