highD: Simulated example data on relative high dimension

In TriangularCell/PgaMsgl: Proximal Gradient Algorithm for Multi-variate Sparse Group Lasso

Description

This is a list includes simulated example data on relative high dimension for PgaMsgl testing.

Format

A list with 18 components:

X

Simulated covariate matrix X in the model Y=XB with dimension 25x50, i.e. X is composed with values of 50 variates in 25 samples. Elements in X are independently standard normal distributed.

Beta0

Simulated coefficient matrix B in the model Y=XB with dimension 50x2500. B is with 10x100 groups tiling it, and each group is a 5×25 module of B. 100 out of 1000 groups were randomly selected, each with 15 elements randomly selected to be with a random value in [-1, 1].

Y

Simulated response matrix Y in the model Y=XB with dimension 25x2500, i.e. Y is composed of values of 2500 responses in 25 samples. It is generated by multiply X with Beta0.

Gm

Matrix of the group structure of coefficient matrix Beta0. It is a 1000x4 matrix with each row indicating a group, four columns indicate the row-start, row-end, column-start and column-end of the group. The row/column index is 1-based.

mi

Maximum number of iteration, which is 2000.

mg

Maximum number of groups in matrix B to be reserved, which is 100.

mc

Maximum number of single coefficients in matrix B to be reserved, which is 1500.

B0

Simulated initial matrix of B, which is composed of all zeros.

B1

Simulated initial matrix of B, which is composed of all ones.

B2

Simulated initial matrix of B, which is generated adding random noise (rnorm(1,mean=1,sd=1)) to the real solution, Beta0.

B3

Simulated initial matrix of B, which is composed of all random noises (rnorm(1,mean=1,sd=1)).

B4

Simulated initial matrix of B, which is to have all non-zero elements in Beta0 as 1.

B5

Simulated initial matrix of B, which is to have non-zero values in Beta0 has 80% probability to be 1, and zero values has 20% probability to be 1.

B6

Simulated initial matrix of B, which is to have non-zero values in Beta0 has 80% probability to be 1, and zero values to be 0.

B7_001

Simulated initial matrix of B. An element was set as 1 if the Benjamini-Hochberg adjusted P-value of the simple linear regression between the corresponding dependent variable and independent variable was less than 0.01, otherwise it was set as 0.

B7_001_coeff

Simulated initial matrix of B. An element was set as the coefficient of the simple linear regression between the corresponding dependent variable and independent variable, if its Benjamini-Hochberg adjusted P-value was less than 0.01, otherwise it was set as 0.

B7_01

Simulated initial matrix of B. Similar to B7_001, but the Benjamini-Hochberg adjusted P-value cutoff was 0.1.

B7_01_coeff

Simulated initial matrix of B. Similar to B7_001_coeff, but the Benjamini-Hochberg adjusted P-value cutoff was 0.1.

Author(s)

Yiming Qin

Examples

data(highD)
system.time(highD_result <- PgaMsgl(highD$X, highD$Y, highD$B0, model="L121", highD$Gm, highD$mi, highD$mg, highD$mc))

TriangularCell/PgaMsgl documentation built on May 28, 2019, 9:33 a.m.