ACTR-ME: ACT-R, Made Easy

A Python package to use ACT-R to analyze behavioral data.

What is this?

This is a prototype for how ACT-R could be reimplemented to allow for faster creation of reasonable and sound models within a typical cognitive psych / cognitive neuroscience workflow. It allows for simulating behavior as well as fitting trial-by-trial data of participant data.

(It is not a full reimplementation of ACT-R, and it does not conform precisely the the ACT-R theory).

Examples

Some simple, interactive examples of its use are included in the examples Jupyter Notebook.

Why?

ACT-R is a much better theory to explain behavior that most other models, but researchers still prefer other packages such DDM, RL. In fact, even memory researchers seem to prefer DDM or RL to ACT-R. Why? Most psych/neuro folks want to use models to measure and compare things (patients, conditions, items) on their way to understand the mind. ACT-R can do that, but not in an easy way that fits with the typical, data analysis workflow of behavioral researchers.

Design Choices

A few design choices were made:

• All code is written in native Python to allow faster integration with common data science libraries (Numpy, Scipy, Matplotlib)
• Production rules are not used unless strictly necessary
• All modules are optional: It is possible to create a model that uses, for example, only declarative memory, or only declarative memory combined with a visual module.
• Modules can be connected with each other for simple, sequential workflow
• Models are simple collections of interconnected modules.
• It should be trivial to create a model to analyze a datafile.

Classes and Definitions

• A model is a collection of interconnected modules. A model functions as the interface between the interconnected
  modules and an outside environment, which could be either another other software or a datafile.
  • A model has inputs and outputs. Inputs and outputs need to be specified from the list of modules within the model. A model automatically includes four outputs: the response, its associated probability, the response time, and the associated response time probability. The probabilities might not be computed.
  • A model must define a run function; the run function examines the model's current inputs and updates and controls how these inputs flow through the modules.
  • A model must define its own timing. The model keeps track of its internal activity, updates its modules about its the current time.
• A module is a component that provides some basic cognitive ability, such as declarative or procedural memory, modeled at the desired level of interest. A models possesses inputs and outputs. A model's output can be connected to another model's input, thus describing the flow of control.
  • A module must contain at least one input and one output.
  • A module must define a run function; the run function examines the module's current inputs and updates and its outputs through its own internal logic and timing.
  • A module must keep track of its own internal time. Before each run, the model might update the module's current time for consistency.
• Inputs and Outputs are the abstract ways in which modules pass information through each other. They can be of two types:
  • Symbolic inputs and outputs are dictionary-like key-value structures ("chunks" and "slot-value" pairs in ACT-R lingo)
  • Numeric inputs and outputs are Python float.
• Inputs and outputs are connected by Connection objects, which define the logic of the connection.
  • All connections are between an input and an output. A dataframe can serves as either an input or an output
  • Connections have an internal logic. For example:
    • Connections between a numeric input and a numeric output simply overwrite the output's value with the input's.
    • Connections between two symbols could either overwrite the entire symbol, or add its contents to the output.
    • Connections between a numeric input and a symbolic output add the numeric value to the output's slot-value pairs.
      • This connection requires a specification of which key will be overwritten.
      • If the key does not exist, it will be added.
    • Connections between a symbolic input and a numeric output extract the numeric value from a slot and use it as the value of the output.
      • This connection requires a specification of which slot will be looked at.
      • The slot must exist.
      • The slot's value will be coerced to float.