Getting started with calmr

In calmr: Canonical Associative Learning Models and their Representations

```{R, include = FALSE} knitr::opts_chunk$set( fig.width = 7, collapse = TRUE, comment = "#>", message = FALSE, warning = FALSE )

To perform your first simulation you will need:

1. A `data.frame` specifying the experiment design, and
2. A list with the parameters for the model you will be using.

## The design data.frame

Let's specify a blocking design.

```r
library(calmr)
my_blocking <- data.frame(
  Group = c("Exp", "Control"),
  Phase1 = c("10A(US)", "10C(US)"),
  Phase2 = c("10AB(US)", "10AB(US)"),
  Test = c("1#A/1#B", "1#A/1#B")
)
# parsing the design and showing the original and what was detected
parsed <- parse_design(my_blocking)
parsed

A few rules about the design data.frame:

1. Each row represents a group.
2. The first column contains the group labels.
3. Every other column represents a phase in the experiment.

The trials in each phase column are specified using a very rigid notation. A handful of observations about it:

1. Trials are preceded by a number. That number represents the number of times that trial will be given in each phase. "10A(US)" means that the "A(US)" trial will be given 10 times.
2. The presence and absence of the unconditioned stimulus are not denoted with the traditional "+" and "-" symbols. Instead, here we use parenthesis to denote "complex" stimuli. These can be thought of as an element with a complex name (i.e., with more than one character). As such, "(US)" specifies a single element to represent the US.
3. In the same vein, multiple characters with no parentheses denote individual elements. For example, "AB" implies the presence of two stimuli, A and B.
4. The "/" character is used as a trial separator (it does not imply randomization by itself). Thus, "1A/1B" specifies that a single "A" trial and a single "B" trial will be given during that phase.
5. The "!" character is used to denote randomization. For example, "!10A/10B" implies that 10 trials of A and 10 trials of B will be given in random order.
6. The ">" character is used to denote trial periods. For example, "1A>B" implies a single trial in which A is followed by B.
7. The "#" character is used to denote probe trials. In contrast to real life, probe trials here entail no update of the model's associations. As such, probe trials can be used to track the development of key associations, with no repercussion to what the model learns on normal training trials.

If you want to check whether your phase string will work with the package, you can use phase_parser().

Warning: The function returns a list with a lot of information used by the models in the package, but the rule of thumb is that if you see a wall of text, your phase string is working.

# not specifying the number of AB trials. Error!
phase_parser("AB/10AC")
# putting the probe symbol out of order. Error!
phase_parser("#10A")
# considering a configural cue for elements AB
trial <- phase_parser("10AB(AB)(US)")
# different USs
trial <- phase_parser("10A(US1)/10B(US2)")
# tons of information! Phase parser is meant for internal use only.
# you are better of using `parse_design()` on a design `data.frame`
str(trial)

The parameters list

Now we need to pick a model and its parameters.

To get the models currently supported in calmr, you can call supported_models().

supported_models()

After choosing a model, you can get some default parameters for your design with get_parameters().

my_pars <- get_parameters(my_blocking, model = "RW1972")
# Increasing the beta parameter for US presentations
my_pars$betas_on["US"] <- .6
my_pars

For a reference on how each model is parametrized, check out the model's reference page. For example, the reference page for the "RW1972" model is here.

Or, if that many equations tire your eyes, you can consult the model parameter reference.

Simulating

With all of the above, we can run our simulation using the run_experiment() function. This function also takes extra arguments to manipulate the number of iterations to run the experiment for (important for designs with randomized trials), whether to organize trials in miniblocks, and extra configuration for more complex models (see the help page for make_experiment() for additional details).

Below, we keep it simple and run the experiment for a single iteration.

my_experiment <- run_experiment(
  my_blocking, # note we do not need to pass the parsed design
  model = "RW1972",
  parameters = my_pars
)
# returns a `CalmrExperiment` object
class(my_experiment)
# CalmrExperiment is an S4 class, so it has slots
slotNames(my_experiment)
# some of the experience given to group Exp on the first (and only) iteration
head(my_experiment@experiences[[1]])
# the number of times we ran the model (groups x iterations)
length(experiences(my_experiment))
# an experiment has results with different levels of aggregation
class(my_experiment@results)
slotNames(my_experiment@results)
# shorthand method to access aggregated_results
results(my_experiment)

If you are an advanced R user you will be able to dig into the data straight away. However, the package also includes some methods to get a quick look at the results.

Plotting

Let's use plot method to create some plots. Each model supports different types of plots according to the results they can produce (e.g., associations, responses, saliences, etc.)

# get all the plots for the experiment
plots <- plot(my_experiment)
names(plots)
# or get a specific type of plot
specific_plot <- plot(my_experiment, type = "associations")
names(specific_plot)
# show which plots are supported by the model we are using
supported_plots("RW1972")

In this case, the RW model supports both associations (associations) and responses (responses).

Stimulus associations

The columns in the plots below are the phases of the design and the rows denote the source of the association.

The colors within each panel determine the target of the association. For example, associations towards the US are shown in yellow.

plot(my_experiment, type = "associations")

Responding

Fairly similar to the above, but responding is a function of the stimuli presented in each trial.

plot(my_experiment, type = "responses")

Graphing

You can also take a look at the state of the model's associations at any point during the experiment, using the graph method. The graphs are created using the ggnetwork package.

# some general options for ggnetwork
my_graph_opts <- get_graph_opts("small")
# passing the argument t to specify the trial we're interested in.
# end of acquisition
patch_graphs(graph(my_experiment, t = 10, options = my_graph_opts))
# end of blocking
patch_graphs(graph(my_experiment, t = 20, options = my_graph_opts))

Final thoughts

The design philosophy behind calmr package revolves around simplicity and ease of access.

The user only needs to specify a design as well as a model to generate model predictions. In fact, there is also an app that lets users access the basic package functionality using a GUI.

That said, the package has plenty of features for more advanced R users. If you're one of them, make sure to check the other vignettes when you are ready.

calmr documentation built on June 8, 2025, 1:03 p.m.