IshitaTakeshi / SoftConfidenceWeighted.jl

Exact Soft Confidence-Weighted Learning

SoftConfidenceWeighted.jl

This is an online supervised learning algorithm which utilizes the four salient properties:

• Large margin training
• Confidence weighting
• Capability to handle non-separable data
• Adaptive margin

The paper is here.

Usage

SCW has 2 formulations of its algorithm which are SCW-I and SCW-II.
You can choose which to use by the parameter of init.

Note

1. This package performs only binary classification, not multiclass classification.
2. Training labels must be 1 or -1. No other labels allowed.

Training from matrix

Feature vectors are given as the columns of the matrix X.

using SoftConfidenceWeighted

# C and ETA are hyperparameters.
# X is a data matrix which each column represents a data vector.
# y is corresponding labels.

model = init(C = 1, ETA = 1, type_ = SCW1)
model = fit!(model, X_train, y_train)
y_pred = predict(model, X_test)