SCARGC.jl
A Julia Implementation of Stream Classification Algorithm Guidded by Clustering
SCARGC.jl is a Julia implementation of Stream Classification Algorithm Guided by Clustering - SCARGC -, a data stream classifier in non-stationary environments with extreme verification latency. The implementation was made using the paper as base and the official implementation of SCARGC, in MatLab.
Overview
Most of research data on data stream classification with concept drift, assumes that there's a availability of the labels, instantly or with some delay after the classification occurs. Then, these methods verify if the model is outdated based on the classification information obtained on the recent data.
SCARGC.jl is an algorithm that make predictions with a EVL scenary, where no labeled data is received after the initialization of the model. So, for example, we receive a small amount of labeled data, build a classifier model with this data and, after that, there's no labeled data "arriving". We just predict and update the model to get a better result in the next iteration.
Installation
In the Julia's package manager, type:
pkg> add SCARGCTutorials
You can find some notebooks explaining how to use the package in experiments.
Quick example
We can see below a example of using the SCARGC.jl package to predict labels in datasets.
It's important to say that the dataset used in the example is a synthetic dataset, so we already have the instances's label. Using it, we can say the accuracy of the algorithm. If you're using it with some real application, with no data, the accuracy calculus isn't going to happen.
using SCARGC
# loading the dataset
dataset = "../../src/datasets/synthetic/1CHT.txt"
data = SCARGC.extractValuesFromFile(dataset, 16000, 3)
# predicting data
labels, accuracy = SCARGC.scargc_1NN(data, 0.3125, 300, 2)
# then, we can print the accuracy
println("Accuracy: ", accuracy, "%")Accuracy: 99.24764890282131%Results
Today, the results of SCARGC.jl are equivalent to the first implemented, in MatLab.
In the picture below, the blue bars represent the original SCARGC and the black ones represent SCARGC.jl results on those datasets (represented in X asis).
