This repository has been archived by the owner on Dec 4, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathREADME.Rmd
146 lines (125 loc) · 7.16 KB
/
README.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
---
#title: "Bayesian mixed models, shrinkage, sparse and interaction kernel regression"
#author: "Luc Janss, Maria Izabel Cavassim Alves and Marni Tausen"
#date: "`r Sys.Date()`"
output:
md_document:
variant: markdown_github
---
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```
# Bayesian mixed models, shrinkage, sparse and interaction kernel regression
[](https://travis-ci.org/MarniTausen/BayzR)[](https://coveralls.io/github/MarniTausen/BayzR?branch=master)
- [How to install](#how-to-install)
- [Short manual for using the bayz() main function](#Short-Manual)
- [Example](#examples)
- [Summarizing and using output](usingOutput.html)
The Rbayz packages has a single main function 'bayz' that
fits mixed-linear, Bayesian shrinkage, (sparse) kernel-regression, (kernel)
interaction and multi-trait models with complex covariance structures using an extended R-formula syntax.
The R help ?bayz contains basic help information; full details are available on <ljanss.github.io/Rbayz>.
# How to install
Most users would like to install our binary packages, that we have available
for Windows 10 and MacOS in the bayz.biz website <http://www.bayz.biz/>.
If you need or want to use the source code for installation, we have
two repositories on github: in ljanss/BayzR, and in MarniTausen/BayzR.
Installation from the ljanss/BayzR is recommended as first option,
this repository contains regularly 'frozen' version that are tested on different
data sets and compile-tested on Windows and Mac-OS. The version in
MarniTausen/BayzR is our development version and may only be of interest
to those that want to co-develop and submit changes to our source code.
You may run into bugs or compilation errors when using the MarniTausen/BayzR
version because we may just be in the middle of developing / testing something.
To install from the github repositories, the following is needed:
1). download and install Rtools. This is not an R-package but a separate part of the
R system and can be found on r-project.org.
2). install the devtools package.
```{r, eval=FALSE}
install.packages("devtools")
```
3). Install dependencies lme4, coda and Rcpp.
```{r, eval=FALSE}
install.packages("lme4")
install.packages("coda")
install.packages("Rcpp")
```
4). Download and install/compile the BayzR package using:
```{r, eval=FALSE}
library(devtools)
devtools::install_github("ljanss/BayzR")
```
You may be asked to install several other packages.
If you are sure you want the development-version, replace in step 3 to install from
MarniTausen/BayzR. If the compilation succeeds, you are ready to use the main function bayz() after
loading the package with library(BayzR).
Short Manual
============
The bayz function fits various mixed-linear and Bayesian shrinkage models with complex
covariance structures using an extended R-formula syntax.
The model formula in bayz has the basic syntax of an R formula but with all
explanatory (right-hand-side) terms wrapped by a function to specify how to fit
the explanatory variables in the model. This may look like Yield ~ fx(Year) + rn(Variety)
to fit Yield with Year as a fixed factor and Variety as a random factor.
The equivalent lme4 model would be Yield ~ factor(Year) + (1|Variety).
The list of model functions currently available is:
fx() : fixed factors (with interactions)
rn() : random factors (with interactions)
rg() : fixed regressions (with interactions or nested in a factor)
rr() : random regressions
The model functions allow to specify interactions of variables, hierarchies, use of
matrices as input data, complex covariance structures, and prior distributions can
be changed to modify standard mixed model in Bayesian shrinkage models.
Interactions between fators are specified using the colon, for instance by writing for a
fixed Year-Location interaction fx(Year:Location). Bayz does
NOT support automatic expansion with main effects by using the 'star' (Year\*Location) or
'forward slash' (Year/Location) syntaxes, hence bayz requires to manually add the desired
main effects in the model (but
note that bayz uses the 'forward slash' to specify hierarchical models).
Interactions between factors can be specified to any degree.
The 'star' syntax can be used to indicate interaction
between covariates, like rg(TempSum\*Precip), where it is simply interpreted as
multiplication.
Interaction between a covariate and a factor is specified using the 'pipe'
character as in lme4 models. It can be used to specify fixed nested regressions as
rg(TempSum|Year:Location) to specify regressions on TempSum within each Year-Location, or
to speficy random slope models in rr(), for instance rr(TempSum|Variety).
The bayz call has a data argument to specify an input data-frame (the "main data")
to contain model variables,
but bayz will also search the R environment if variables are not found in the main data.
Typically, input in the form of matrices such as large sets of covariates, proportional
variance-covariance / correlation / kernel / similarity matrices used in variance models,
as well as data for hierarchical models, are not in the main data.
Matrices / kernels should be prepared with row-names to match a variable
in the data. For matrices / kernels used in variance models, the link is straightforward,
for instance in rn(Variety:Location, V=KG\*KE), KG and KE can be a genetic and environmental
kinship / kernel matrix, respectively, and KG should have row-names matching Variety levels,
while KE should have row-names matching Location levels.
To fit a large set of covariates, the match is specified using a hierachical specification
rr(Variety/Metabolites), where Metabolites is then a matrix of covariates which must have
row-names matching Variety levels. In both cases, such kernel or covariate matrices
must have unique levels, but the main data may have repeated levels and in different order.
If one has repeated metabolite data on each Variety, for instance at multiple time-points,
then consider that Variety is not the appropriate link to the data, but Variety:Time is.
For random effects a variance-covariance structure can be specified using a
V= option within the model-term function, for example rn(Variety, V=KG).
When the fit is for an interaction of factors, the variance specification should
expand to include one term for each variable, separated by stars (which should
be read as Kronecker products). The variance structure
is then built up from a combination of given (proportional) variance-covariance / correlation /
kernel matrices and predefined acronyms IDEN, DIAG and VCOV that indicate parameterized
matrices (with parameters to be estimated from the data).
Alternatively, the variance structure can be specified as a linear model using V=~,
which is interpreted as a use of a log-linear model for the variances.
# Examples
```{r, eval=FALSE}
library(BayzR)
# A dummy test for bayz
y = rnorm(1000)
A = as.factor(rep(1:20,50))
B = as.factor(rep(1:10,each=100))
dat1 = data.frame(y,A,B)
library(BayzR)
fit1 = bayz(y~fx(A)+fx(B)+rn(A:B),data=dat1,chain=c(2000,100,20))
summary(fit1)
```