kkt.check: KKT conditions plots for MSTweedie objects

In fontaine618/MSTweedie: Multi-source Sparse Tweedie Modelling

Description

This functions plots the values of the KKT conditions in the solution path of a MSTweedie object.

Usage

kkt.check(fit, eps = fit$eps, cond = c(1, 2),
      from = 1, to = length(fit$lambda), ...)

Arguments

fit	MSTweedie object.
eps	Value of the threshold to add to the plots. Default is the convergence threshold of the fit object, i.e. fit$eps.
cond	A vector of integers between 1 and 4 indicating which conditions to plot. See "Details".
from	The index of the first regularization parameter to include. Default is 1.
to	The index of the last regularization parameter to include. Default is length(fit$lambda).
...	Additionnal graphical parameters to pass to plot.

Details

The reference contains detailed explanation of these KKT conditions.

If code contains 1, then the sequence of values for the non-zero KKT condition is plotted:

||U_j||=λ v_j,β_j\neq 0.

If code contains 2, then the sequence of values for the zero KKT condition is plotted:

||U_j||≤qλ v_j,β_j= 0.

If code contains 3, then the sequence of the detailed values for the non-zero KKT condition is plotted:

-U_j/λ v_j=0,β_j\neq 0, k\not\in M(β_j).

If code contains 4, then the sequence of the aggregate KKT condition is plotted:

\frac{-∑_{j=1}^pU_j^\topβ_j}{∑_{j=1}^pv_j||β_j||} = λ.

Value

Multiple plots are produced.

Author(s)

Simon Fontaine, Yi Yang, Bo Fan, Wei Qian and Yuwen Gu.

Maintainer: Simon Fontaine [e-mail]

References

Fontaine, S., Yang, Y., Fan, B., Qian, W. and Gu, Y. (2018). "A Unified Approach to Sparse Tweedie Model with Big Data Applications to Multi-Source Insurance Claim Data Analysis," to be submitted.

See Also

MSTweedie

Examples

# import package
library(MSTweedie)

# load data
data(AutoClaim)

# performs 10-folds CV with L1/Linf regularization
fit <- MSTweedie(x = AutoClaim, y=1, source=4, reg='Linf')

# plot the first two kkt conditions
par(mfrow=c(2,1))
kkt.check(fit)

fontaine618/MSTweedie documentation built on May 25, 2019, 5:22 p.m.