psychNET: Psychometric networks estimated by multivariate time series...
In psychNET: Psychometric Networks for Intensive Longitudinal Data
Description
Usage
Arguments
Details
Value
Author(s)
References
Examples
Description
Wrapper function of multivariate time-series models, which can be used to obtain symptom networks (i.e., networks of symptom-symptom interactions).
Usage
1
2
Arguments
data
: Usually a "matrix", "data.frame", or "longitudinal" object of dimension (n*T) x p where n is the number of individuals. For time-series data from multiple persons, an additional numeric column explicitly named "ID" indicating from which person the measurements are coming from should be included in the data. Additionally, with experience sampling data that are nested within days, additional columns named "DAY" and "BEEP" can be included in data from which we calculate the consecutiveness of the measurements. Alternatively, a column named "TIME" denoting the measuremnt consecutiveness can be provided in data. See details.
model
: This argument controls the model to be fitted to the data. The available options are: "VAR", "SVAR", "SMVAR", "SVECM", "GVAR", "SVARHL", "SVARHLX", "SVARHLMA", "DFM", "MLVAR", "GGVAR".
lag
: Specifies the lag order of the process. When model is "DFM" lag corresponds to the order of the factor process. When model is "SVARHL", "SVARHLX", or "SVARHLMA" lag can be NULL and the model chooses the optimal lag via regularized hierarchical lag structures.
criterion
: The information criterion to be used in order to tune the penalty parameters of regularized (sparse) VAR models. Available options are: "CV", "AIC", "BIC", "EBIC", "GIC", and "MBIC". This argument depends on the model argument (see details).
nFact
: This argument is used only when model="DFM" and controls the number of the static factors in the dynamic factor model.
penalty
: This argument controls the type of regularization to be used and depends on the model argument (see details). Available options are "LASSO", "ENET", "SCAD", "MCP", and "HLag".
lambda1
: A numeric vector with length greater than one that specifies the values of the penalty parameter for regularized VARs. Typically, this corresponds to the penalty parameter for the lagged effects. When model is "GVAR" or "GGVAR" then lambda1 is used as the penalty value for the precision matrix.
lambda2
: A numeric vector with length greater than one. This argument is used only when model is "GVAR" or "GGVAR" and controls the value of the penalty parameter for the lagged effects.
covariates
: This argument is used only when model is "VAR" or "SVARHLX" in order to specify covariates for VARX or SVARX models.
impute
: String that is used to specify the imputation method when missing values exist in the data or when missing values are inserted to transform non-equidistant measurements into equidistant. It can be a single string or a vector of strings with length equal to the number of symptoms in the data. In the latter case, each symptom can be imputed using an explicit method. Available options are: "Kalman.arima", "Kalman.struct", "Interpol.linear", "Interpol.spline", "Interpol.stine", "LOCF", "NOCB", "MA.simple", "MA.linear", "MA.exponent", "mean", "mode", "median", "random", "Season.int.linear", "Season.int.spline", "Season.int.stine", "Season.LOCF", "Season.NOCB","Season.mean", "Season.median", "Season.mode", "Season.MA", "Season.kalman", "Season.random".
transform
: String that is used to specify the transormation function. It can be a single string or a vector of strings with length equal to the number of symptoms in the data. In the latter case, each symptom can be transformed using an explicit function. Available options are: "log", "log10", "Copula_discr", "Copula_skew", "Zero.mean", "Standardize", "Power", "Logit", "Square.root", "Power2", "Power3", "Cube.root".
...
Argument that depends on the value of the model argument. It is used to pass additional arguments to the model fitting function.
Details
The data
The data to be used in the psychNET function must strictly contain symptom expression data where the measurements are ordered by time. The only additional variables that are allowed in the data except symptom expression are a variable explicitly coded as "ID" when multiple persons need to be analyzed together, a variable coded as "DAY" when multiple measurements per day are taken and a variable called "BEEP" to indicate the measurement intensity within a given day. The variables "DAY" and "BEEP" are used internally to construct a time variable that denotes the consecutiveness of the measurements. The time variable can also be provided in the data explicitly named as "TIME". The only models in the package that can handle non-equidistant observations naturally are the "GVAR", "SMVAR", and "MLVAR". In any other model, non-equidistant measurements first are transformed to equidistant by inserting missing values, which are then imputed using the imputation method specified in the impute argument.
The models
Here we provide some more details with respect to the models implemented in the package.
VAR model (VAR)
The traditional VAR model is fitted with the psychNET function by setting model="VAR". This model is originally implememnted in the function VAR that comes with the R package vars. In this implementation the unknown model parameters are estimated by ordinary least squares (OLS) per equation. The arguments lag and covariates of the psychNET function correspond to the parameter p and exogen respectively of the original VAR function. Other parameters that are used in the original function can be passed through the ... structure (technically known as ellipsis). For details on arguments that can be passed to the three dots when fitting a VAR the user can type ?vars::VAR into the console in order to access its help file.
Sparse VAR model (SVAR)
A sparse VAR model is fitted with the psychNET function by setting model="SVAR". This model is implememnted in the function fitVAR from the R package sparsevar and the unknown model parameters are estimated via penalized multivariate least squares. Three different types of penalization are available that can be passed to the function through the penalty argument. If you set penalty="ENET" an elasticnet penalty is applied to the parameters with other option being "SCAD" for SCAD type of penalization and "MCP" for MC+ penalty. The only option available for tuning the penalty parameter is k-fold cross validation. The parameter lag of the psychNET function corresponds to the parameter p of the fitVAR function, the parameter penalty is used in the same way as in the fitVAR function while criterion substitutes the argument method of the fitVAR function. Additional arguments can be passed to the fitVAR function by using the ... structure of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?sparsevar::fitVAR.
Sparse VAR model with hierarchical lag structures and additional covariates (SVARHL and SVARHLX)
A sparse VAR that offers the possibility of simultaneous lag estimation is fitted by setting model="SVARHL". Additional covariates can be used by setting model="SVARHLX" and providing the covariates data to the argument covariates of the psychNET function. These two models are available with the R package bigtime in the functions sparseVAR and sparseVARX respectively. These models are estimated by penalized least squares optimized by a proximal gradient algorithm. These models offer two types of penalization and the tuning of the penalty parameters is done by k-fold cross validation. One option is to penalize the VAR coefficients by hierarchical group LASSO regularization and the other option is the standard LASSO also known as L1 regularization. For hierarchical group LASSO the user must set penalty="HLag" and for standard LASSO penalty="LASSO". In the psychNET function, penalty argument corresponds to the VARpen and VARXpen arguments of the functions sparseVAR and sparseVARX respectively, while lag is associated with the parameter p of these two functions. Typically, the function internally calculates a grid for the regularization parameter corresponding to sparse penalty. The user can provide her/his own grid of regularization parameters with the argument lambda1. The latter acts like the arguments VARlseq and VARXlPhiseq of the original functions sparseVAR and sparseVARX respectively. Other arguments of these functions can be passed via the ... structure of the psychNET function. If you want the function to estimate the lag of the process simultaneously you must set lag=NULL. For details on the original functions and additional parameters that can be used you can look at ?bigtime::sparseVAR and ?bigtime::sparseVARX.
Sparse VAR model with simultaneous covariance estimation (GVAR)
A sparse VAR with simultaneous covariance estimation (also known as graphical VAR) is fitted by setting model="GVAR". This model is available with the R package graphicalVAR in the function graphicalVAR. The model parameters are estimated by otimizing a penalized maximum likelihood. The only penalization option when model="GVAR" is LASSO, which means that penalty argument must be equal to "LASSO". The tuning of the penalty parameters is done by the extended Bayesian information criterion (EBIC) by setting criterion="EBIC". Model selection via standard BIC is possible by using an additional argument gamma=0 in the psychNET function. This model uses two distinct penalties, one for the autoregressive parameters and one for the covariance parameters. Typically, the function calculates its own grid of penalty parameters for the regularization, however, the user can provide her/his own grid by using the parameters lambda1 (for covariance penalization) and lambda2 (for autoregressive penalization) respectively in the function psychNET. Other arguments of the original graphicalVAR function can be passed via the ... structure of the psychNET function.
!!WARNING!! Do not use the ... to provide the following arguments of the graphicalVAR function: vars, scale, idvar, beepvar, dayvar, centerWithin, deleteMissings since they are already specified internally. If you want to use idvar, beepvar, dayvar arguments of the original function graphicalVAR, these variables must be provided as columns in the data object explicitly named as "ID", "BEEP", and "DAY" respectively. For scale and centerWithin arguments you can use the argument transform of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?graphicalVAR::graphicalVAR.
Sparse VAR model for mixed type of time series (SMVAR)
A sparse VAR with simultaneous covariance estimation (also known as graphical VAR) is fitted by setting model="SMVAR". This model is available with the R package mgm in the function mvar. The model parameters are estimated by otimizing a penalized least squares function per equation. The only penalization option when model="SMVAR" is elasticnet, which means that penalty argument must be equal to "ENET". The tuning of the penalty parameters is done by the extended Bayesian information criterion (EBIC) by setting criterion="EBIC" or k-fold cross validation by setting criterion="CV". Typically, the function internally calculates a grid for the regularization parameter. The user can provide her/his own grid of regularization parameters with the argument lambda1. The latter acts analogous to the argument lambdaSeq of the original mvar function. Model selection via standard BIC is possible by using an additional argument gamma=0 in the psychNET function. Other arguments of the original mvar function can be passed via the ... structure of the psychNET function.
!!WARNING 1!! When model="SMVAR" the data must be given in a "data.frame" format with the following properties:
Gaussian variables: must be of class "numeric"
Poisson variables: must be of class "integer"
Categorical variables: must be of class "factor"
Ordinal variables: must be of class "ordered factor" (SMVAR for ordinal type of variables to be expected in a following version of the package psychNET).
!!WARNING 2!! Do not use the ... to provide the following arguments of the mvar function: type, level, scale, beepvar, dayvar since they are already specified internally. If you want to use beepvar and dayvar arguments of the original function mvar, these variables must be provided as columns in the data object explicitly named as "BEEP", and "DAY" respectively. For scale you can use the argument transform of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?mgm::mvar.
Sparse VEC model (SVECM)
A sparse VEC (vector error correction) model is fitted with the psychNET function by setting model="SVECM". This model is implememnted in the function fitVECM from the R package sparsevar and the unknown model parameters are estimated via penalized multivariate least squares. Three different types of penalization are available that can be passed to the function through the penalty argument. If you set penalty="ENET" an elasticnet penalty is applied to the parameters with other option being "SCAD" for SCAD type of penalization and "MCP" for MC+ penalty. The only option available for tuning the penalty parameter is k-fold cross validation. The parameter lag of the psychNET function corresponds to the parameter p of the fitVECM function, the parameter penalty is used in the same way as in the fitVECM function while criterion substitutes the argument method of the fitVAR function. Additional arguments can be passed to the fitVAR function by using the ... structure of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?sparsevar::fitVECM. The argument logScale of the orgiginal function is set to FALSE, but you can use the argument transform of the psychNET function instead.
Sparse VARMA model with hierarchical lag structures (SVARHLMA)
A sparse VARMA (vector autoregressive moving average) model that offers the possibility of simultaneous lag estimation is fitted by setting model="SVARHLMA". This model is available with the R package bigtime in the function sparseVARMA. The model parameters are estimated by penalized least squares optimized by a proximal gradient algorithm. The model offers two types of penalization and the tuning of the penalty parameters is done by k-fold cross validation. One option is to penalize the VAR coefficients by hierarchical group LASSO regularization and the other option is the standard LASSO. For hierarchical group LASSO the user must set penalty="HLag" and for standard LASSO penalty="LASSO". In the psychNET function, the penalty argument corresponds to the VARpen argument of the function sparseVARMA, while lag is associated with the parameter p of this function. Typically, the function internally calculates a grid for the regularization parameter corresponding to sparse penalty. The user can provide her/his own grid of regularization parameters with the argument lambda1. The latter acts like the arguments VARlseq of the original function sparseVARMA. Other arguments of this function can be passed via the ... structure of the psychNET function. If you want the function to estimate the lag of the process simultaneously you must set lag=NULL. For details on the original function and additional parameters that can be used you can look at ?bigtime::sparseVARMA.
Dynamic factor model (DFM)
A DFM is fitted by setting model="DFM" in the psychNET function. This model is available with the R package dynfactoR (available at github) in the function dfm. The model parameters are estimated by an EM algorithm. The argument nFact of the psychNET function is used to specify the number of static factors and it is analogous to the argument r of the original dfm function. The argument lag of the psychNET function is used to specify the lag order of the factor process analogously to the argument p of the dfm function. The DFM that we implement here is restricted to have an identity system state covariance matrix and zeros at the upper diagonal elements of the factor loading matrix for identifiability purposes. The user of the psychNET package can specify her/his own threshold for algorithm convergence by using an additional argument threshold.
Multi-level VAR model for group level dynamics (MLVAR)
An MLVAR can be fitted by setting model="MLVAR". This model is available with the R package mlVAR in the function mlVAR.
!!WARNING!! Do not use the ... to provide the following arguments of the mlVAR function: vars, scale, idvar, beepvar, dayvar, estimator, and scaleWithin since they are already specified internally. If you want to use idvar, beepvar, dayvar arguments of the original function mlVAR, these variables must be provided as columns in the data object explicitly named as "ID", "BEEP", and "DAY" respectively. For the scaleWithin argument you can use the argument transform of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?mlVAR::mlVAR. The only estimation possible is "lmer".
Sparse VAR model with simultaneous covariance estimation for group level dynamics (GGVAR)
A sparse VAR with simultaneous covariance estimation for multiple individuals (also known as graphical VAR) is fitted by setting model="GGVAR". This model is available with the R package SparseTSCGM in the function sparse.tscgm. The model parameters are estimated by optimizing a penalized maximum likelihood. Penalization options when model="GGVAR" is LASSO and SCAD, which means that penalty argument must be equal to "LASSO" for LASSO penalization and equal to "SCAD" for SCAD penalization. The tuning of the penalty parameters is done by AIC, BIC, EBIC, GIC (generalized information criterion), and MBIC (modified Bayesian information criterion) by using the criterion argument which is the same as the optimality argument in the sparse.tscgm function. This model uses two distinct penalties, one for the autoregressive parameters and one for the covariance parameters. Typically, the function calculates its own grid of penalty parameters for the regularization, however, the user can provide her/his own grid by using the parameters lambda1 (for covariance penalization) and lambda2 (for autoregressive penalization) respectively in the function psychNET. Other arguments of the original graphicalVAR function can be passed via the ... structure of the psychNET function. Be aware that only VAR models with up to 2 lags are possible.
Additional details
The psychNET function will estimate (based on the lag of the process) one or more temporal conditional independence graphs unless the model equals to "DFM". For model="DFM" and lag=1 the function will estimate the equivalent VAR(1), whereas for lag greater than one, the temporal network of the factor interactions will be estimated. Contemporaneous conditional independence graphs are available only when the model argument equals to: "GVAR", "GGVAR", "MLVAR", or "DFM" with one lag.
Value
The function psychNET returns a list object of class pnt.
Author(s)
Spyros E. Balafas (author, creator), Sanne Booij, Marco A. Grzegorczyk, Hanneke Wardenaar-Wigman, Ernst C. Wit
Maintainer: Spyros E. Balafas (s.balafas@rug.nl)
References
Lutkepohl, H. (2005). New Introduction to Multiple Time Series Analysis. Springer, New York.
Basu, S., Michailidis, G. (2015). Regularized estimation in sparse high-dimensional time series models. Ann. Statist. 43, no. 4, 1535-1567.
Abegaz, F., Wit, E. (2013). Sparse time series chain graphical models for reconstructing genetic networks. Biostatistics. 14, 3: 586-599.
Haslbeck, J., Waldorp, L. J. (2015). mgm: Structure Estimation for time-varying Mixed Graphical Models in high-dimensional Data.
Nicholson, W. B., Bien, J., Matteson, D. S. (2017). High Dimensional Forecasting via Interpretable Vector Autoregression..
Wilms, I., Basu, S., Bien, J., Matteson D. S. (2017). Sparse Identification and Estimation of High-Dimensional Vector AutoRegressive Moving Averages.
Epskamp, S., Waldorp, L. J., Mottus, R., Borsboom, D. (2016). The Gaussian Graphical Model in Cross-sectional and Time-series Data.
Examples
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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
## load the psychNET library
library(psychNET)
## load the Canada dataset from the 'vars' package
data("Canada", package = "vars")
Canada_data_frame <- data.frame(Canada)
## fit a VAR model
VAR_model <- psychNET(Canada_data_frame, model = "VAR", lag = 1, type = "const")
# print the result
VAR_model
# summarize the resulting network
summary(VAR_model)
# summarize the VAR model using the original summary method
vars:::summary.varest(VAR_model$fit)
# plot the VAR model results using the original plot method
vars:::plot.varest(VAR_model$fit)
# plot the resulting network
plot(VAR_model)
## fit a sparse VAR model
sparse_VAR_model <- psychNET(Canada_data_frame, model = "SVAR", lag = 1)
# print the result
sparse_VAR_model
# summarize the resulting network
summary(sparse_VAR_model)
# plot the resulting network
plot(sparse_VAR_model)
## fit a sparse VAR model as the one in the 'bigtime' package
sparse_lassVAR_model <- psychNET(Canada_data_frame,
model = "SVARHL",penalty = "LASSO", lag = 1, VARgran=c(500,1000))
# print the result
sparse_lassVAR_model
# summarize the resulting network
summary(sparse_lassVAR_model)
# plot the resulting network
plot(sparse_lassVAR_model)
## Load the psychNET package
library(psychNET)
################################## N=1 models ####################################
##################################################################################
####################################################################
######## REPRODUCE EXAMPLE OF A VAR FROM THE 'vars' PACKAGE ########
####################################################################
## load the 'vars' package
library(vars)
## Load the Canada dataset from the vars package
data(Canada)
## check the structure of the data
str(Canada)
## The data is in time series format. It needs to be transformed into
## a matrix, data.frame or longitudinal object for the psychNET package
Canada_data_frame <- data.frame(Canada)
## fitting a VAR using the vars package
varmod <- vars::VAR(Canada, p = 2, type = "none")
## fitting the same VAR using the psychNET package
psychvar <- psychNET(Canada_data_frame, model = "VAR", lag = 2, type = "none")
## Check if the results are the same
all.equal(Acoef(varmod), psychvar$results$A, check.attributes = FALSE)
################################################################################
####### Fit A DYNAMIC FACTOR TO THE CANADA DATA FROM THE 'vars' PACKAGE #######
################################################################################
## install and load the 'devtools' package and the
## by using install.packages("devtools") and then
## library(devtools)
## install the 'dynfactoR' package available on github
## by using devtools::install_github("rbagd/dynfactoR")
## library(dynfactoR)
## Load the Canada dataset from the vars package
data(Canada, package = "vars")
## check the structure of the data
str(Canada)
## The data is in time series format. It needs to be transformed into
## a matrix, data.frame or longitudinal object for the psychNET package
Canada_data_frame <- data.frame(Canada)
## fitting a DFM using the dynfactoR package
dfmmod <- psychNET:::dfm(Canada_data_frame, r=2, p = 1,
rQ= "identity", rC= "upper",max_iter = 100000)
## fitting the same DFM using the psychNET package
psychdfm <- psychNET(Canada_data_frame, model = "DFM", nFact = 2, lag = 1)
## Check if the results are the same
all.equal(dfmmod$A, psychdfm$results$A_fact[[1]], check.attributes = FALSE)
all.equal(dfmmod$C, psychdfm$results$B_fac_symptoms, check.attributes = FALSE)
all.equal(dfmmod$Q, psychdfm$results$System_Covariance, check.attributes = FALSE)
all.equal(dfmmod$R, psychdfm$results$Obs_Covariance, check.attributes = FALSE)
#####################################################################################
########### FIT A SPARSE VAR TO THE CANADA DATA FROM THE 'vars' PACKAGE ############
#####################################################################################
## Load the Canada dataset from the vars package
data(Canada, package = "vars")
## The data is in time series format. It needs to be transformed into
## a matrix, data.frame or longitudinal object for the psychNET package
Canada_data_frame <- data.frame(Canada)
## fitting a SVAR using the sparsevar package
set.seed(1)
svarmod <- psychNET:::fitVAR(Canada, p = 1, penalty="SCAD", method="cv", logScale=FALSE)
## fitting the same SVAR using the psychNET package
set.seed(1)
psychsvar <- psychNET(Canada_data_frame, model = "SVAR",penalty = "SCAD", lag = 1, criterion="CV")
## Check if the results are the same
all.equal(svarmod$A, psychsvar$results$A, check.attributes = FALSE)
########################################################################################
####### EXAMPLE OF A SPARSE VAR WITH HIERARCHICAL LAGS FROM THE 'bigtime' PACKAGE ######
########################################################################################
## load the 'bigtime' package
library(bigtime)
## Load the Y dataset from the bigtime package
data(Y, package = "bigtime")
## fitting a SVAR with hierarchical lags
## using the bigtime package
svarhlmod <- sparseVAR(Y, VARpen ="HLag")
## fitting the same model using the psychNET package
psychsvarhl <- psychNET(Y, model = "SVARHL", penalty = "HLag", criterion="CV")
## Check if the results are the same
all.equal(svarhlmod$Phihat, psychsvarhl$fit$Phihat, check.attributes = FALSE)
######################################################################################
########### REPRODUCE EXAMPLE OF A SPARSE MIXED VAR FROM THE 'mgm' PACKAGE ###########
######################################################################################
## load the 'mgm' package
library(mgm)
# 1) Define mVAR model as in the mgm manual
p <- 6 # Six variables
type <- c("c", "c", "c", "c", "g", "g") # 4 categorical, 2 gaussians
level <- c(2, 2, 4, 4, 1, 1) # 2 categoricals with m=2, 2 categoricals with m=4, two continuous
max_level <- max(level)
lags <- 1:3 # include lagged effects of order 1-3
n_lags <- length(lags)
# Specify thresholds
thresholds <- list()
thresholds[[1]] <- rep(0, level[1])
thresholds[[2]] <- rep(0, level[2])
thresholds[[3]] <- rep(0, level[3])
thresholds[[4]] <- rep(0, level[4])
thresholds[[5]] <- rep(0, level[5])
thresholds[[6]] <- rep(0, level[6])
# Specify standard deviations for the Gaussians
sds <- rep(NULL, p)
sds[5:6] <- 1
# Create coefficient array
coefarray <- array(0, dim=c(p, p, max_level, max_level, n_lags))
# a.1) interaction between continuous 5<-6, lag=3
coefarray[5, 6, 1, 1, 2] <- .4
# a.2) interaction between 1<-3, lag=1
m1 <- matrix(0, nrow=level[2], ncol=level[4])
m1[1,1:2] <- 1
m1[2,3:4] <- 1
coefarray[1, 3, 1:level[2], 1:level[4], 1] <- m1
# a.3) interaction between 1<-5, lag=9
coefarray[1, 5, 1:level[1], 1:level[5], 3] <- c(0, 1)
# 2) Sample
set.seed(1)
dlist <- mvarsampler(coefarray = coefarray,
lags = lags,
thresholds = thresholds,
sds = sds,
type = type,
level = level,
N = 200,
pbar = TRUE)
# 3) Transform data into a data.frame for psychoNET suitability
# each categorical variable is coded as factor, each poisson as integer, gaussian as numeric
d1 <- as.data.frame(dlist$data)
d1$V1 <- as.factor(d1$V1)
d1$V2 <- as.factor(d1$V2)
d1$V3 <- as.factor(d1$V3)
d1$V4 <- as.factor(d1$V4)
dat <- d1
## fitting the SMVAR model using the mgm package
smvarmod <- mvar(data = dlist$data,
type = type,
level = level,
lambdaSel = "EBIC",
lags = 1:3,
scale = FALSE,
signInfo = FALSE,
overparameterize = FALSE)
## fitting the same model using the psychNET package
psychsmvar <- psychNET(dat, model = "SMVAR",
lag = 3,
criterion = "EBIC",
signInfo = FALSE,
overparameterize = FALSE)
all.equal(smvarmod$wadj,psychsmvar$fit$wadj)
################################### N>1 models ########################################
#######################################################################################
######################################################################################
########### REPRODUCE EXAMPLE OF A MULTILEVEL VAR FROM THE 'mlVAR' PACKAGE ###########
######################################################################################
## load the 'mlVAR' package
library(mlVAR)
## Simulate data:
Model <- mlVARsim(nPerson = 50, nNode = 3, nTime = 50, lag=1)
# Estimate an MLVAR with correlated random effects using the mlVAR package
mlvarmod <- mlVAR(Model$Data, vars = Model$vars,
idvar = Model$idvar, lags = 1, temporal = "correlated")
# Estimate the same MLVAR using the psychNET package
psychmlvar <- psychNET(Model$Data, model="MLVAR", lag = 1, temporal = "correlated")
# Check if the results are equal
all.equal(mlvarmod$results, psychmlvar$fit$results)
#######################################################################################
########### REPRODUCE EXAMPLE OF A GGVAR VAR FROM THE 'SparseTSCGM' PACKAGE ###########
#######################################################################################
## load the 'SparseTSCGM' package
library(SparseTSCGM)
## Simulate data:
seed = 321
datas <- sim.data(model="ar1", time=10,n.obs=10, n.var=5,seed=seed,
prob0=0.35, network="random")
data.fit <- datas$data1
# Estimate a group graphical VAR (also known as time series chain graphical model)
# using the SparseTSCGM package
ggvarmod <- sparse.tscgm(data=data.fit,
lam1=NULL, lam2=NULL, nlambda=NULL,
model="ar1", penalty="scad",optimality="bic",
control=list(maxit.out = 10, maxit.in = 100))
# Estimate the same model using the psychNET package
psychggvar <- psychNET(data.fit, model="GGVAR", lag=1, penalty="SCAD", criterion="BIC",
control=list(maxit.out = 10, maxit.in = 100))
# Check if the results are equal
all.equal(ggvarmod$theta, psychggvar$fit$theta, check.attributes = FALSE)
all.equal(ggvarmod$gamma, psychggvar$fit$gamma, check.attributes = FALSE)
psychNET documentation built on April 14, 2020, 6:39 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Description Usage Arguments Details Value Author(s) References Examples
Description
Wrapper function of multivariate time-series models, which can be used to obtain symptom networks (i.e., networks of symptom-symptom interactions).
Usage
1 2 |
Arguments
data |
: Usually a |
model |
: This argument controls the model to be fitted to the data. The available options are: |
lag |
: Specifies the lag order of the process. When |
criterion |
: The information criterion to be used in order to tune the penalty parameters of regularized (sparse) VAR models. Available options are: |
nFact |
: This argument is used only when |
penalty |
: This argument controls the type of regularization to be used and depends on the |
lambda1 |
: A numeric vector with length greater than one that specifies the values of the penalty parameter for regularized VARs. Typically, this corresponds to the penalty parameter for the lagged effects. When |
lambda2 |
: A numeric vector with length greater than one. This argument is used only when |
covariates |
: This argument is used only when |
impute |
: String that is used to specify the imputation method when missing values exist in the data or when missing values are inserted to transform non-equidistant measurements into equidistant. It can be a single string or a vector of strings with length equal to the number of symptoms in the data. In the latter case, each symptom can be imputed using an explicit method. Available options are: |
transform |
: String that is used to specify the transormation function. It can be a single string or a vector of strings with length equal to the number of symptoms in the data. In the latter case, each symptom can be transformed using an explicit function. Available options are: |
... |
Argument that depends on the value of the |
Details
The data
The data to be used in the psychNET function must strictly contain symptom expression data where the measurements are ordered by time. The only additional variables that are allowed in the data except symptom expression are a variable explicitly coded as "ID" when multiple persons need to be analyzed together, a variable coded as "DAY" when multiple measurements per day are taken and a variable called "BEEP" to indicate the measurement intensity within a given day. The variables "DAY" and "BEEP" are used internally to construct a time variable that denotes the consecutiveness of the measurements. The time variable can also be provided in the data explicitly named as "TIME". The only models in the package that can handle non-equidistant observations naturally are the "GVAR", "SMVAR", and "MLVAR". In any other model, non-equidistant measurements first are transformed to equidistant by inserting missing values, which are then imputed using the imputation method specified in the impute argument.
The models
Here we provide some more details with respect to the models implemented in the package.
VAR model (VAR)
The traditional VAR model is fitted with the psychNET function by setting model="VAR". This model is originally implememnted in the function VAR that comes with the R package vars. In this implementation the unknown model parameters are estimated by ordinary least squares (OLS) per equation. The arguments lag and covariates of the psychNET function correspond to the parameter p and exogen respectively of the original VAR function. Other parameters that are used in the original function can be passed through the ... structure (technically known as ellipsis). For details on arguments that can be passed to the three dots when fitting a VAR the user can type ?vars::VAR into the console in order to access its help file.
Sparse VAR model (SVAR)
A sparse VAR model is fitted with the psychNET function by setting model="SVAR". This model is implememnted in the function fitVAR from the R package sparsevar and the unknown model parameters are estimated via penalized multivariate least squares. Three different types of penalization are available that can be passed to the function through the penalty argument. If you set penalty="ENET" an elasticnet penalty is applied to the parameters with other option being "SCAD" for SCAD type of penalization and "MCP" for MC+ penalty. The only option available for tuning the penalty parameter is k-fold cross validation. The parameter lag of the psychNET function corresponds to the parameter p of the fitVAR function, the parameter penalty is used in the same way as in the fitVAR function while criterion substitutes the argument method of the fitVAR function. Additional arguments can be passed to the fitVAR function by using the ... structure of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?sparsevar::fitVAR.
Sparse VAR model with hierarchical lag structures and additional covariates (SVARHL and SVARHLX)
A sparse VAR that offers the possibility of simultaneous lag estimation is fitted by setting model="SVARHL". Additional covariates can be used by setting model="SVARHLX" and providing the covariates data to the argument covariates of the psychNET function. These two models are available with the R package bigtime in the functions sparseVAR and sparseVARX respectively. These models are estimated by penalized least squares optimized by a proximal gradient algorithm. These models offer two types of penalization and the tuning of the penalty parameters is done by k-fold cross validation. One option is to penalize the VAR coefficients by hierarchical group LASSO regularization and the other option is the standard LASSO also known as L1 regularization. For hierarchical group LASSO the user must set penalty="HLag" and for standard LASSO penalty="LASSO". In the psychNET function, penalty argument corresponds to the VARpen and VARXpen arguments of the functions sparseVAR and sparseVARX respectively, while lag is associated with the parameter p of these two functions. Typically, the function internally calculates a grid for the regularization parameter corresponding to sparse penalty. The user can provide her/his own grid of regularization parameters with the argument lambda1. The latter acts like the arguments VARlseq and VARXlPhiseq of the original functions sparseVAR and sparseVARX respectively. Other arguments of these functions can be passed via the ... structure of the psychNET function. If you want the function to estimate the lag of the process simultaneously you must set lag=NULL. For details on the original functions and additional parameters that can be used you can look at ?bigtime::sparseVAR and ?bigtime::sparseVARX.
Sparse VAR model with simultaneous covariance estimation (GVAR)
A sparse VAR with simultaneous covariance estimation (also known as graphical VAR) is fitted by setting model="GVAR". This model is available with the R package graphicalVAR in the function graphicalVAR. The model parameters are estimated by otimizing a penalized maximum likelihood. The only penalization option when model="GVAR" is LASSO, which means that penalty argument must be equal to "LASSO". The tuning of the penalty parameters is done by the extended Bayesian information criterion (EBIC) by setting criterion="EBIC". Model selection via standard BIC is possible by using an additional argument gamma=0 in the psychNET function. This model uses two distinct penalties, one for the autoregressive parameters and one for the covariance parameters. Typically, the function calculates its own grid of penalty parameters for the regularization, however, the user can provide her/his own grid by using the parameters lambda1 (for covariance penalization) and lambda2 (for autoregressive penalization) respectively in the function psychNET. Other arguments of the original graphicalVAR function can be passed via the ... structure of the psychNET function.
!!WARNING!! Do not use the ... to provide the following arguments of the graphicalVAR function: vars, scale, idvar, beepvar, dayvar, centerWithin, deleteMissings since they are already specified internally. If you want to use idvar, beepvar, dayvar arguments of the original function graphicalVAR, these variables must be provided as columns in the data object explicitly named as "ID", "BEEP", and "DAY" respectively. For scale and centerWithin arguments you can use the argument transform of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?graphicalVAR::graphicalVAR.
Sparse VAR model for mixed type of time series (SMVAR)
A sparse VAR with simultaneous covariance estimation (also known as graphical VAR) is fitted by setting model="SMVAR". This model is available with the R package mgm in the function mvar. The model parameters are estimated by otimizing a penalized least squares function per equation. The only penalization option when model="SMVAR" is elasticnet, which means that penalty argument must be equal to "ENET". The tuning of the penalty parameters is done by the extended Bayesian information criterion (EBIC) by setting criterion="EBIC" or k-fold cross validation by setting criterion="CV". Typically, the function internally calculates a grid for the regularization parameter. The user can provide her/his own grid of regularization parameters with the argument lambda1. The latter acts analogous to the argument lambdaSeq of the original mvar function. Model selection via standard BIC is possible by using an additional argument gamma=0 in the psychNET function. Other arguments of the original mvar function can be passed via the ... structure of the psychNET function.
!!WARNING 1!! When model="SMVAR" the data must be given in a "data.frame" format with the following properties:
Gaussian variables: must be of class "numeric"
Poisson variables: must be of class "integer"
Categorical variables: must be of class "factor"
Ordinal variables: must be of class "ordered factor" (SMVAR for ordinal type of variables to be expected in a following version of the package psychNET).
!!WARNING 2!! Do not use the ... to provide the following arguments of the mvar function: type, level, scale, beepvar, dayvar since they are already specified internally. If you want to use beepvar and dayvar arguments of the original function mvar, these variables must be provided as columns in the data object explicitly named as "BEEP", and "DAY" respectively. For scale you can use the argument transform of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?mgm::mvar.
Sparse VEC model (SVECM)
A sparse VEC (vector error correction) model is fitted with the psychNET function by setting model="SVECM". This model is implememnted in the function fitVECM from the R package sparsevar and the unknown model parameters are estimated via penalized multivariate least squares. Three different types of penalization are available that can be passed to the function through the penalty argument. If you set penalty="ENET" an elasticnet penalty is applied to the parameters with other option being "SCAD" for SCAD type of penalization and "MCP" for MC+ penalty. The only option available for tuning the penalty parameter is k-fold cross validation. The parameter lag of the psychNET function corresponds to the parameter p of the fitVECM function, the parameter penalty is used in the same way as in the fitVECM function while criterion substitutes the argument method of the fitVAR function. Additional arguments can be passed to the fitVAR function by using the ... structure of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?sparsevar::fitVECM. The argument logScale of the orgiginal function is set to FALSE, but you can use the argument transform of the psychNET function instead.
Sparse VARMA model with hierarchical lag structures (SVARHLMA)
A sparse VARMA (vector autoregressive moving average) model that offers the possibility of simultaneous lag estimation is fitted by setting model="SVARHLMA". This model is available with the R package bigtime in the function sparseVARMA. The model parameters are estimated by penalized least squares optimized by a proximal gradient algorithm. The model offers two types of penalization and the tuning of the penalty parameters is done by k-fold cross validation. One option is to penalize the VAR coefficients by hierarchical group LASSO regularization and the other option is the standard LASSO. For hierarchical group LASSO the user must set penalty="HLag" and for standard LASSO penalty="LASSO". In the psychNET function, the penalty argument corresponds to the VARpen argument of the function sparseVARMA, while lag is associated with the parameter p of this function. Typically, the function internally calculates a grid for the regularization parameter corresponding to sparse penalty. The user can provide her/his own grid of regularization parameters with the argument lambda1. The latter acts like the arguments VARlseq of the original function sparseVARMA. Other arguments of this function can be passed via the ... structure of the psychNET function. If you want the function to estimate the lag of the process simultaneously you must set lag=NULL. For details on the original function and additional parameters that can be used you can look at ?bigtime::sparseVARMA.
Dynamic factor model (DFM)
A DFM is fitted by setting model="DFM" in the psychNET function. This model is available with the R package dynfactoR (available at github) in the function dfm. The model parameters are estimated by an EM algorithm. The argument nFact of the psychNET function is used to specify the number of static factors and it is analogous to the argument r of the original dfm function. The argument lag of the psychNET function is used to specify the lag order of the factor process analogously to the argument p of the dfm function. The DFM that we implement here is restricted to have an identity system state covariance matrix and zeros at the upper diagonal elements of the factor loading matrix for identifiability purposes. The user of the psychNET package can specify her/his own threshold for algorithm convergence by using an additional argument threshold.
Multi-level VAR model for group level dynamics (MLVAR)
An MLVAR can be fitted by setting model="MLVAR". This model is available with the R package mlVAR in the function mlVAR.
!!WARNING!! Do not use the ... to provide the following arguments of the mlVAR function: vars, scale, idvar, beepvar, dayvar, estimator, and scaleWithin since they are already specified internally. If you want to use idvar, beepvar, dayvar arguments of the original function mlVAR, these variables must be provided as columns in the data object explicitly named as "ID", "BEEP", and "DAY" respectively. For the scaleWithin argument you can use the argument transform of the psychNET function. For details on the original function and additional parameters that can be used you can look at ?mlVAR::mlVAR. The only estimation possible is "lmer".
Sparse VAR model with simultaneous covariance estimation for group level dynamics (GGVAR)
A sparse VAR with simultaneous covariance estimation for multiple individuals (also known as graphical VAR) is fitted by setting model="GGVAR". This model is available with the R package SparseTSCGM in the function sparse.tscgm. The model parameters are estimated by optimizing a penalized maximum likelihood. Penalization options when model="GGVAR" is LASSO and SCAD, which means that penalty argument must be equal to "LASSO" for LASSO penalization and equal to "SCAD" for SCAD penalization. The tuning of the penalty parameters is done by AIC, BIC, EBIC, GIC (generalized information criterion), and MBIC (modified Bayesian information criterion) by using the criterion argument which is the same as the optimality argument in the sparse.tscgm function. This model uses two distinct penalties, one for the autoregressive parameters and one for the covariance parameters. Typically, the function calculates its own grid of penalty parameters for the regularization, however, the user can provide her/his own grid by using the parameters lambda1 (for covariance penalization) and lambda2 (for autoregressive penalization) respectively in the function psychNET. Other arguments of the original graphicalVAR function can be passed via the ... structure of the psychNET function. Be aware that only VAR models with up to 2 lags are possible.
Additional details
The psychNET function will estimate (based on the lag of the process) one or more temporal conditional independence graphs unless the model equals to "DFM". For model="DFM" and lag=1 the function will estimate the equivalent VAR(1), whereas for lag greater than one, the temporal network of the factor interactions will be estimated. Contemporaneous conditional independence graphs are available only when the model argument equals to: "GVAR", "GGVAR", "MLVAR", or "DFM" with one lag.
Value
The function psychNET returns a list object of class pnt.
Author(s)
Spyros E. Balafas (author, creator), Sanne Booij, Marco A. Grzegorczyk, Hanneke Wardenaar-Wigman, Ernst C. Wit
Maintainer: Spyros E. Balafas (s.balafas@rug.nl)
References
Lutkepohl, H. (2005). New Introduction to Multiple Time Series Analysis. Springer, New York.
Basu, S., Michailidis, G. (2015). Regularized estimation in sparse high-dimensional time series models. Ann. Statist. 43, no. 4, 1535-1567.
Abegaz, F., Wit, E. (2013). Sparse time series chain graphical models for reconstructing genetic networks. Biostatistics. 14, 3: 586-599.
Haslbeck, J., Waldorp, L. J. (2015). mgm: Structure Estimation for time-varying Mixed Graphical Models in high-dimensional Data.
Nicholson, W. B., Bien, J., Matteson, D. S. (2017). High Dimensional Forecasting via Interpretable Vector Autoregression..
Wilms, I., Basu, S., Bien, J., Matteson D. S. (2017). Sparse Identification and Estimation of High-Dimensional Vector AutoRegressive Moving Averages.
Epskamp, S., Waldorp, L. J., Mottus, R., Borsboom, D. (2016). The Gaussian Graphical Model in Cross-sectional and Time-series Data.
Examples
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 | ## load the psychNET library
library(psychNET)
## load the Canada dataset from the 'vars' package
data("Canada", package = "vars")
Canada_data_frame <- data.frame(Canada)
## fit a VAR model
VAR_model <- psychNET(Canada_data_frame, model = "VAR", lag = 1, type = "const")
# print the result
VAR_model
# summarize the resulting network
summary(VAR_model)
# summarize the VAR model using the original summary method
vars:::summary.varest(VAR_model$fit)
# plot the VAR model results using the original plot method
vars:::plot.varest(VAR_model$fit)
# plot the resulting network
plot(VAR_model)
## fit a sparse VAR model
sparse_VAR_model <- psychNET(Canada_data_frame, model = "SVAR", lag = 1)
# print the result
sparse_VAR_model
# summarize the resulting network
summary(sparse_VAR_model)
# plot the resulting network
plot(sparse_VAR_model)
## fit a sparse VAR model as the one in the 'bigtime' package
sparse_lassVAR_model <- psychNET(Canada_data_frame,
model = "SVARHL",penalty = "LASSO", lag = 1, VARgran=c(500,1000))
# print the result
sparse_lassVAR_model
# summarize the resulting network
summary(sparse_lassVAR_model)
# plot the resulting network
plot(sparse_lassVAR_model)
## Load the psychNET package
library(psychNET)
################################## N=1 models ####################################
##################################################################################
####################################################################
######## REPRODUCE EXAMPLE OF A VAR FROM THE 'vars' PACKAGE ########
####################################################################
## load the 'vars' package
library(vars)
## Load the Canada dataset from the vars package
data(Canada)
## check the structure of the data
str(Canada)
## The data is in time series format. It needs to be transformed into
## a matrix, data.frame or longitudinal object for the psychNET package
Canada_data_frame <- data.frame(Canada)
## fitting a VAR using the vars package
varmod <- vars::VAR(Canada, p = 2, type = "none")
## fitting the same VAR using the psychNET package
psychvar <- psychNET(Canada_data_frame, model = "VAR", lag = 2, type = "none")
## Check if the results are the same
all.equal(Acoef(varmod), psychvar$results$A, check.attributes = FALSE)
################################################################################
####### Fit A DYNAMIC FACTOR TO THE CANADA DATA FROM THE 'vars' PACKAGE #######
################################################################################
## install and load the 'devtools' package and the
## by using install.packages("devtools") and then
## library(devtools)
## install the 'dynfactoR' package available on github
## by using devtools::install_github("rbagd/dynfactoR")
## library(dynfactoR)
## Load the Canada dataset from the vars package
data(Canada, package = "vars")
## check the structure of the data
str(Canada)
## The data is in time series format. It needs to be transformed into
## a matrix, data.frame or longitudinal object for the psychNET package
Canada_data_frame <- data.frame(Canada)
## fitting a DFM using the dynfactoR package
dfmmod <- psychNET:::dfm(Canada_data_frame, r=2, p = 1,
rQ= "identity", rC= "upper",max_iter = 100000)
## fitting the same DFM using the psychNET package
psychdfm <- psychNET(Canada_data_frame, model = "DFM", nFact = 2, lag = 1)
## Check if the results are the same
all.equal(dfmmod$A, psychdfm$results$A_fact[[1]], check.attributes = FALSE)
all.equal(dfmmod$C, psychdfm$results$B_fac_symptoms, check.attributes = FALSE)
all.equal(dfmmod$Q, psychdfm$results$System_Covariance, check.attributes = FALSE)
all.equal(dfmmod$R, psychdfm$results$Obs_Covariance, check.attributes = FALSE)
#####################################################################################
########### FIT A SPARSE VAR TO THE CANADA DATA FROM THE 'vars' PACKAGE ############
#####################################################################################
## Load the Canada dataset from the vars package
data(Canada, package = "vars")
## The data is in time series format. It needs to be transformed into
## a matrix, data.frame or longitudinal object for the psychNET package
Canada_data_frame <- data.frame(Canada)
## fitting a SVAR using the sparsevar package
set.seed(1)
svarmod <- psychNET:::fitVAR(Canada, p = 1, penalty="SCAD", method="cv", logScale=FALSE)
## fitting the same SVAR using the psychNET package
set.seed(1)
psychsvar <- psychNET(Canada_data_frame, model = "SVAR",penalty = "SCAD", lag = 1, criterion="CV")
## Check if the results are the same
all.equal(svarmod$A, psychsvar$results$A, check.attributes = FALSE)
########################################################################################
####### EXAMPLE OF A SPARSE VAR WITH HIERARCHICAL LAGS FROM THE 'bigtime' PACKAGE ######
########################################################################################
## load the 'bigtime' package
library(bigtime)
## Load the Y dataset from the bigtime package
data(Y, package = "bigtime")
## fitting a SVAR with hierarchical lags
## using the bigtime package
svarhlmod <- sparseVAR(Y, VARpen ="HLag")
## fitting the same model using the psychNET package
psychsvarhl <- psychNET(Y, model = "SVARHL", penalty = "HLag", criterion="CV")
## Check if the results are the same
all.equal(svarhlmod$Phihat, psychsvarhl$fit$Phihat, check.attributes = FALSE)
######################################################################################
########### REPRODUCE EXAMPLE OF A SPARSE MIXED VAR FROM THE 'mgm' PACKAGE ###########
######################################################################################
## load the 'mgm' package
library(mgm)
# 1) Define mVAR model as in the mgm manual
p <- 6 # Six variables
type <- c("c", "c", "c", "c", "g", "g") # 4 categorical, 2 gaussians
level <- c(2, 2, 4, 4, 1, 1) # 2 categoricals with m=2, 2 categoricals with m=4, two continuous
max_level <- max(level)
lags <- 1:3 # include lagged effects of order 1-3
n_lags <- length(lags)
# Specify thresholds
thresholds <- list()
thresholds[[1]] <- rep(0, level[1])
thresholds[[2]] <- rep(0, level[2])
thresholds[[3]] <- rep(0, level[3])
thresholds[[4]] <- rep(0, level[4])
thresholds[[5]] <- rep(0, level[5])
thresholds[[6]] <- rep(0, level[6])
# Specify standard deviations for the Gaussians
sds <- rep(NULL, p)
sds[5:6] <- 1
# Create coefficient array
coefarray <- array(0, dim=c(p, p, max_level, max_level, n_lags))
# a.1) interaction between continuous 5<-6, lag=3
coefarray[5, 6, 1, 1, 2] <- .4
# a.2) interaction between 1<-3, lag=1
m1 <- matrix(0, nrow=level[2], ncol=level[4])
m1[1,1:2] <- 1
m1[2,3:4] <- 1
coefarray[1, 3, 1:level[2], 1:level[4], 1] <- m1
# a.3) interaction between 1<-5, lag=9
coefarray[1, 5, 1:level[1], 1:level[5], 3] <- c(0, 1)
# 2) Sample
set.seed(1)
dlist <- mvarsampler(coefarray = coefarray,
lags = lags,
thresholds = thresholds,
sds = sds,
type = type,
level = level,
N = 200,
pbar = TRUE)
# 3) Transform data into a data.frame for psychoNET suitability
# each categorical variable is coded as factor, each poisson as integer, gaussian as numeric
d1 <- as.data.frame(dlist$data)
d1$V1 <- as.factor(d1$V1)
d1$V2 <- as.factor(d1$V2)
d1$V3 <- as.factor(d1$V3)
d1$V4 <- as.factor(d1$V4)
dat <- d1
## fitting the SMVAR model using the mgm package
smvarmod <- mvar(data = dlist$data,
type = type,
level = level,
lambdaSel = "EBIC",
lags = 1:3,
scale = FALSE,
signInfo = FALSE,
overparameterize = FALSE)
## fitting the same model using the psychNET package
psychsmvar <- psychNET(dat, model = "SMVAR",
lag = 3,
criterion = "EBIC",
signInfo = FALSE,
overparameterize = FALSE)
all.equal(smvarmod$wadj,psychsmvar$fit$wadj)
################################### N>1 models ########################################
#######################################################################################
######################################################################################
########### REPRODUCE EXAMPLE OF A MULTILEVEL VAR FROM THE 'mlVAR' PACKAGE ###########
######################################################################################
## load the 'mlVAR' package
library(mlVAR)
## Simulate data:
Model <- mlVARsim(nPerson = 50, nNode = 3, nTime = 50, lag=1)
# Estimate an MLVAR with correlated random effects using the mlVAR package
mlvarmod <- mlVAR(Model$Data, vars = Model$vars,
idvar = Model$idvar, lags = 1, temporal = "correlated")
# Estimate the same MLVAR using the psychNET package
psychmlvar <- psychNET(Model$Data, model="MLVAR", lag = 1, temporal = "correlated")
# Check if the results are equal
all.equal(mlvarmod$results, psychmlvar$fit$results)
#######################################################################################
########### REPRODUCE EXAMPLE OF A GGVAR VAR FROM THE 'SparseTSCGM' PACKAGE ###########
#######################################################################################
## load the 'SparseTSCGM' package
library(SparseTSCGM)
## Simulate data:
seed = 321
datas <- sim.data(model="ar1", time=10,n.obs=10, n.var=5,seed=seed,
prob0=0.35, network="random")
data.fit <- datas$data1
# Estimate a group graphical VAR (also known as time series chain graphical model)
# using the SparseTSCGM package
ggvarmod <- sparse.tscgm(data=data.fit,
lam1=NULL, lam2=NULL, nlambda=NULL,
model="ar1", penalty="scad",optimality="bic",
control=list(maxit.out = 10, maxit.in = 100))
# Estimate the same model using the psychNET package
psychggvar <- psychNET(data.fit, model="GGVAR", lag=1, penalty="SCAD", criterion="BIC",
control=list(maxit.out = 10, maxit.in = 100))
# Check if the results are equal
all.equal(ggvarmod$theta, psychggvar$fit$theta, check.attributes = FALSE)
all.equal(ggvarmod$gamma, psychggvar$fit$gamma, check.attributes = FALSE)
|
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.