How to setup DDModeling in R-studio

First things first

In order to use the package you have to install and load it

# install.packages("devtools")
# devtools::install_github("Pelzer402/DDModeling")
library(DDModeling)

Defining models

Model definition in DDModeling is straightforward using the DDModel function.

DSTP_M1 <-  DDModel(model="DSTP",task="flanker",
                    CDF_perc = c(0.1,0.3,0.5,0.7,0.9),CAF_perc = c(0.0,0.2,0.4,0.6,0.8,1.0))

Note that the function can be used with specific task presets (as of DDModeling 0.0.1.0 only flanker is available), which will lessen the needed input of the function tremendously. Since DDModel returns a S4-class object, you can access (and modify by override) it’s slots by using the “@” operator. For explaination of the slots please see the DDModel-class help.

DSTP_M1@DM
#>              Ter    a    c mu_t mu_f mu_RS2 mu_SS
#> Upper_Limit 0.45 0.38 0.38 0.15 0.25   0.55   1.2
#> Lower_Limit 0.15 0.14 0.14 0.05 0.05   0.25   0.4

Simulating a model

The function Sim_DDModel will generate model simulations (as of DDModeling 0.0.1.0 this will only return a simulation for a randomly drawn parameterset of the domain specified in the model).

Fitting data

Suppose you did an experiment and want to fit the data to your model. Here is what needs to be done.

Read and format data

For demonstraion purposes the DDModeling package includes a dataset called FLANKER_DATA which constitudes 64 subjects who were part of a flanker task experiment. We will now use one of these samples:

Subj1 <- FLANKER_DATA[[2]]

Then we can generate a representation (i.e. CDF/CAF) from this by using the DDRep function.

Subj1_Rep <- DDRep(model = DSTP_M1,raw = Subj1)
Subj1_Rep
#> CDF: 
#> $Cong
#>   cond perc time   N
#> 1 Cong  0.1  294  22
#> 2 Cong  0.3  335  68
#> 3 Cong  0.5  365 114
#> 4 Cong  0.7  416 159
#> 5 Cong  0.9  516 205
#> 
#> $Incong
#>     cond perc time   N
#> 1 Incong  0.1  286  26
#> 2 Incong  0.3  330  78
#> 3 Incong  0.5  364 131
#> 4 Incong  0.7  403 184
#> 5 Incong  0.9  469 236
#> 
#> 
#> CAF: 
#> $Cong
#>   cond perc time      acc N_A N_B
#> 1 Cong  0.1  277 0.468750  30  34
#> 2 Cong  0.3  321 0.625000  40  24
#> 3 Cong  0.5  354 0.859375  55   9
#> 4 Cong  0.7  403 0.781250  50  14
#> 5 Cong  0.9  523 0.828125  53  11
#> 
#> $Incong
#>     cond perc time      acc N_A N_B
#> 1 Incong  0.1  272 0.750000  48  16
#> 2 Incong  0.3  320 0.718750  46  18
#> 3 Incong  0.5  355 0.828125  53  11
#> 4 Incong  0.7  397 0.906250  58   6
#> 5 Incong  0.9  502 0.906250  58   6
#> 
#> 
#> Parameter: 
#>   Ter a c mu_t mu_f mu_RS2 mu_SS
#> 1   0 0 0    0    0      0     0

Note that in order for this to work, your data (i.e. Subj1) has to be formated in a very specific way! Make sure that your data

• is represented in a data.frame
• contains only three coulmns with the specific names cond, resp and time (for explaination what each coulmn refers to and how it should be formated see the DDRep help page)
• the factor levels of the cond coulmn corresond to the conditions specified in your model

Generate a grid

DDModeling uses a combination of grid-search and downhill simplex in order to fit a given model. Therefore there is a need to calculate a grid before fitting. This can be done with the Generate_GRID function:

Grid1 <- Generate_GRID(model = DSTP_M1, path = getwd(), name = "Grid1")

Several notes for this functions are needed:

• You will be asked to specify the dimensions of the grid when calling the function. These refer to the number of equally spaced evaluation points on the domains for each parameter specified in the model. We suggest to consider the computational cost of choosing high grid dimension in combination with a highly parameterized model, as there will be the need of model simulations equal to the product of all specified dimensions!
• As of DDModeling 0.0.1.0 all grid simulations will be conducted using 10000 trials
• The calculation will be multi-threaded to the number of threads minus one of a given system
• The progress bar (as of DDModeling 0.0.1.0) is only a placeholder
• Since the grid can be, depending on the dimensions, quiete large, it will be saved to disc
• The return value of the functions is not a grid per definition, but rather the path to the grid saved on disc

Fit

Finally we can fit the data using the Fit_DDModel function

Subj1_Fit <- Fit_DDModel(model = DSTP_M1,rep = Subj1_Rep,grid_path = Grid1,s_sampling=FALSE,trials=10000)

There are several things to note here, which will be discussed with further development.

Note of interest

Since this package is still in early stages of development (and not yet submitted to CRAN), many of the functions and theire handling are subject to change. However the functionality described on this site will allways be part of the final product.