reg_2d {UniIsoRegression}

Isotonic Regression on 2D input.

Description

Isotonic regression on weighted or unweighted 2D input with L1, L2 metric and other options.

Usage

    reg_2d(y_vec, w_vec, metric)

Arguments

y_vec

The 2D NumericMatrix of input data that we use to regression. It must be the same size as the w_vec argument.

w_vec

The 2D NumericMatrix of the weight of the input data. The default value is 1 for every entry. It must be the same size as y_vec.

metric

This is an integer input, metric = 1 stands for using L1 metric, metric = 2 stands for using L2 metric

Details

See the paper about 2D regression in the reference.

Value

A 2D NumericMatrix of the regression result which has the same size of y_vec.

Error Messages

• The size of y_vec is 0: Empty data.

• The rows of w_vec doesn't match the rows of y_vec: Data and weight have different number of rows

• The columns of w_vec doesn't match the rows of y_vec: Data and weight have different number of columns

• The entry of w_vec has negative value: Negative weight detected

• Metric input is not in 1,2,3: Metric does not exist

Author(s)

Zhipeng Xu, Chenkai Sun, Aman Karunakaran, Quentin Stout [e-mail] https://github.com/xzp1995/UniIsoRegression

References

Q.F. Stout, Isotonic median regression via partitioning, Algorithmica 66 (2013), pp. 93-112 doi.org/10.1007/s00453-012-9628-4

Examples

    library(UniIsoRegression)
    #===2d monotonic===
    y=matrix(c(2, 4, 3, 1, 5, 7,9,0), nrow=2, ncol=4, byrow = TRUE)
    weight=matrix(c(1, 10, 3, 9, 5, 7,9,10), nrow=2, ncol=4, byrow = TRUE)

    #l_1 metric
    temp=UniIsoRegression::reg_2d(y, weight, metric = 1)
    print(temp)

    #l_2 metric
    temp=UniIsoRegression::reg_2d(y, weight, metric = 2)
    print(temp)