Package Guide

To work with the Rocketeer.jl package, you should know:

Overview

This package implements a module that generates a set of randomly distributed kernels for extracting features on a time series dataset for classification with some other machine learning method. It is based upon the work in this paper and this Python implementation, but it is not affiliated with these works or these original authors.

What this package is

A package for generating, applying, saving, and loading these random feature kernels.

What this package is not

This package does not implement any machine learning (i.e., parameter updating) methods themselves for learning upon these feature on the back end, nor does it implement any data manipulation tools on the front end. These you are responsible depending upon your application ("batteries not included").

Installation

The Rocketeer.jl package can be installed using the Julia package manager. From the Julia REPL, type ] to enter the Pkg REPL mode and run

julia> ]
(@v1.9) pkg> add Rocketeer

Alternatively, it can be added to your environment in a script with

using Pkg
Pkg.add("Rocketeer")

If you wish to have the latest changes between releases, you can directly add the GitHub repo at an arbitrary branch (such as develop) as a dependency with

julia> ]
(@v1.9) pkg> add https://github.com/AP6YC/Rocketeer.jl#develop

Instantiation and Usage

After installation, load the package in a script in the usual Julia pattern (with either using or import):

# `using` brings all exported names into this scope
using Rocketeer

# `import` just brings the module name itself
import Rocketeer

Create a RocketModule with the following hyperparameters:

  1. input_length: the expected window length of the input data.

This is used to determine the distribution of the depth of the kernels.

  1. n_kernels: the number of

like so:

## Specify these hyperparameters
my_rocket = RocketModule(input_length, n_kernels)
## Use the default values (input_length=5, n_kernels=100)
my_default_rocket = RocketModule()

Input data is assumed to be a vector of Real numbers, such as those created when calling rand:

X = rand(input_length)

To apply the kernels and extract their features, simply call apply_kernels upon the RocketModule and your data:

features = apply_kernels(my_rocket, X)

This results in a feature set of dimension (n_kernels, 2) due to the way that the Rocket kernel features are defined. For more details, please see the original paper.

Saving and Loading

This package implements two functions for convenient saving and loading of the RocketModule. To save a set of kernels for later reuse, simple use save_rocket:

filename = "my_rocket_file"
save_rocket(my_rocket, filename)

and load with load_rocket:

my_loaded_rocket = load_rocket(filename)

These two use JLD2.jl under the hood. You can also choose to save the RocketModule with any other package, as the stateful information of the module is implemented with native Julia types (e.g., Vectors of Real numbers, etc.) that have well-defined serializations for saving and loading.