Skip to contents

Function: Learning Rate

Source: R/func_alpha.R
func_alpha.Rd

$$Q_{new} = Q_{old} + \alpha \cdot (R - Q_{old})$$

Usage

func_alpha(
  shown,
  is.fp,
  qvalue,
  reward,
  utility,
  system,
  rownum,
  params,
  hidden,
  ...
)

Arguments

shown

Which options shown in this trial.

is.fp

Is it the first time picking this option?

qvalue

The expected Q values of different behaviors produced by different systems when updated to this trial.

reward

The feedback received by the agent from the environment at trial(t) following the execution of action(a)

utility

The subjective value (internal representation) assigned by the agent to the objective reward.

system

When the agent makes a decision, is a single system at work, or are multiple systems involved? see system

rownum

The trial number

params

Parameters used by the model's internal functions, see params

hidden

All hidden variables within the MDP process belong here.

...

It currently contains the following information; additional information may be added in future package versions.

  • idinfo:

    • subid

    • block

    • trial

  • exinfo: contains information whose column names are specified by the user.

    • Frame

    • RT

    • NetWorth

    • ...

  • behave: includes the following:

    • action: the behavior performed by the human in the given trial.

    • latent: the object updated by the agent in the given trial.

    • simulation: the actual behavior performed by the agent.

    • position: the position of the stimulus on the screen.

  • cue and rsp: Cues and responses within latent learning rules, see behrule

  • state: The state stores the stimuli shown in the current trial—split into components by underscores—and the rewards associated with them.

Value

A List

  • output [NumericVector]

    A numeric value representing the updated Q-value after learning.

    This function specifies how prediction error (PE) is incorporated into value updating, using a learning rate that determines whether updates are more conservative or more aggressive in response to PE.

  • hidden [CharacterVector]

    User-defined internal variables generated by this function. These represent intermediate (latent) states produced during computation, which can be read or modified by other functions in the MDP process.

Body

func_alpha <- function(
    shown,
    is.fp,
    qvalue,
    reward,
    utility,
    params,
    rownum,
    system,
    hidden,
    ...
){

  list2env(list(...), envir = environment())

  # If you need extra information(...)
  # Column names may be lost(C++), indexes are recommended
  # e.g.
  # Trial  <- idinfo[3]
  # Frame  <- exinfo[1]
  # Action <- behave[1]

  Q0        <-  params[["Q0"]]
  alpha     <-  params[["alpha"]]
  alphaN    <-  params[["alphaN"]]
  alphaP    <-  params[["alphaP"]]

  if (is.nan(Q0) && first) {
    update <- utility
    hidden[1] <- "first"
    return(list(output = update, hidden = hidden))
  }

  # Determine the model currently in use based on which parameters are free.
  if (
    system == "RL" && !(is.null(alpha)) && is.null(alphaN) && is.null(alphaP)
  ) {
    model <- "TD"
  } else if (
    system == "RL" && is.null(alpha) && !(is.null(alphaN)) && !(is.null(alphaP))
  ) {
    model <- "RSTD"
  } else if (
    system == "WM"
  ) {
    model <- "WM"
  } else {
    stop("Unknown Model! Plase modify your learning rate function")
  }

  # TD
  if (model == "TD") {
    update <- qvalue + alpha * (utility - qvalue)
  # RSTD
  } else if (model == "RSTD" && utility < qvalue) {
    update <- qvalue + alphaN * (utility - qvalue)
  } else if (model == "RSTD" && utility >= qvalue) {
    update <- qvalue + alphaP * (utility - qvalue)
  # WM
  } else if (model == "WM") {
    update <- reward
  }

  return(list(output = update, hidden = hidden))
}