ACT-R Assignment Unit6

Task : Coin Guess

develop a model for a “probability matching” experiment
Experiment Flow：
- presented with a screen saying “Choose”
- either presses the ‘h’ or ‘t’
- h = heads
- t = tails
- key pressed –> screen cleared
- feedback of correct answer (h / t ) holds for 1s
Condition :
- 48 trails
- 90% of the 48 trials, the correct answer is head
fit data: (each block contain 12 trail) proportion of “head” choice
- 1st block : 0.664
- 2nd block : 0.778
- 3rd block : 0.804
- 4th block : 0.818
should run multiple times of the experiment and average results
reference (goal)
- R = 0.994
- MD = 0.016
parameter tuning
- noise in the utilities (set by the :egs parameter)
- rewards associated with successful and unsuccessful responses.

Example experiment functions

choice-model.lisp
choice.lisp
useful commands:
- reset
- run
- run_full_time(float amout_of_time)
- install_device()
- correlation()
- mean_deviation()

Solution

Task

import actr
import numpy as np
# actr.load_act_r_model("ACT-R:tutorial;unit6;choice-model.lisp")
actr.load_act_r_model("ACT-R:tutorial;unit6;choice-model-solution.lisp")
choice_data = [0.664, 0.778, 0.804, 0.818]
response = False
virtual_mode = True

def dominant_ratio(response):
    if response.count("h") > response.count("t"):
        dominant = "h"
    else:
        dominant = "t"
    return np.mean(np.array(response)==dominant)


def respond_to_key_press (model,key):
    global response

    response = key

def trial(window,probability,person=True):
    global response

    actr.add_command("choice-response",respond_to_key_press,"Choice task key response")
    actr.monitor_command("output-key","choice-response")
    actr.add_text_to_exp_window (window, 'choose', x=50, y=100)

    response = ''

    if person:
        while response == '':
            actr.process_events()
    else:
        actr.run(1,not virtual_mode)

    actr.clear_exp_window(window)

    if actr.random(1.0) < probability:
        answer = 'heads'
    else:
        answer = 'tails'

    actr.add_text_to_exp_window (window, answer, x=50, y=100)

    start = actr.get_time(False)

    if person:
        while (actr.get_time(False) - start) < 1000:
            actr.process_events()
    else:
        actr.run(1,not virtual_mode)
    actr.clear_exp_window(window)
    actr.remove_command_monitor("output-key","choice-response")
    actr.remove_command("choice-response")

    return response

def block(block_id,probability,person=True):
    block_res = []
    window = actr.open_exp_window("Choice Experiment",visible=False)
    actr.install_device(window)
    for i in range(12):
        response = trial(window,probability,person)
        block_res.append(response)
    res = dominant_ratio(block_res)

    #print("Block-",block_id+1,"\tAccuracy=",hratio)
    return res


def Experiment(num_block=4,probability=0.9,person=False):
    result_list = []
    for i in range(num_block):
        result_list.append(block(i,probability,person))
    return result_list


def compare(n=100):
    result_list = []
    for i in range(n):
        result_list.append(Experiment(probability=0.9))
        result_list.append(Experiment(probability=0.1))

    result_list = np.array(result_list)
    avg_result = list(np.mean(result_list,axis=0))
    r=actr.correlation(avg_result,choice_data)
    md=actr.mean_deviation(avg_result,choice_data)
    print()
    print("Model Result:",avg_result)
    print("Human Result:",choice_data)
    print()
    print("correlation=",r)
    print("mean deviation=",md)

if __name__ == "__main__":
    compare(100)

Model

(clear-all)

(define-model choice-solution

;(sgp :seed (200 4))

(sgp :v nil :ul t  :egs 3 :esc t)
(chunk-type goal state)
(chunk-type response val)


(add-dm
 (start isa chunk)
 (attending-choose isa chunk) (respond isa chunk)
 (attending-feedback isa chunk) (read-feedback isa chunk) (done isa chunk)
 (goal isa goal state start)
)

(install-device '("motor" "keyboard"))

(P start
    =goal>
       isa     goal
       state   start
==>
    =goal>
        isa goal
        state attending-choose
    +visual-location>
)

(P attend-letter
   =goal>
      ISA         goal
      state       attending-choose
   =visual-location>
   ?visual>
      state       free
==>
   +visual>
      cmd         move-attention
      screen-pos  =visual-location
   =goal>
      ISA         goal
      state       respond
)


(p respond-t
    =goal>
      isa      goal
      state    respond
    =visual>
      isa      visual-object
      value    "choose"
      ;(!print! =val)
    ?manual>
      state    free   
    ?imaginal>
      state    free
   ==>
    +manual>
        cmd press-key
        key "t"
    +imaginal>
        isa     response
        val     "tails"
    =goal>
      isa      goal
      state    attending-feedback
    +visual-location>
)



(p respond-h
    =goal>
      isa      goal
      state    respond
    =visual>
      isa      visual-object
      value    "choose"
      ;(!print! =val)
    ?manual>
      state    free   
    ?imaginal>
      state    free
   ==>
    +manual>
        cmd press-key
        key "h"
    +imaginal>
        isa     response
        val     "heads"
    =goal>
      isa      goal
      state    attending-feedback
    +visual-location>
)


(p attending-feedback
    =visual-location>
    =goal>
      isa      goal
      state    attending-feedback
 ==>
    =goal>
      isa      goal
      state    read-feedback
    +visual-location>
      :attended nil
)



(p read-feedback
   =goal>
       ISA goal
       state read-feedback
   =visual-location>
   ?visual>
      state       free
==>
   +visual>
      cmd         move-attention
      screen-pos  =visual-location
   =goal>
       ISA        goal
       state      done

)

(p done-correct
    =goal>
      isa      goal
      state    done
    =visual>
      isa      visual-object
      value    =val
    =imaginal>
      val  =val
   ==>
    =goal>
      state    start
)

(p done-wrong
    =goal>
      isa      goal
      state    done
    =visual>
      isa      visual-object
      value    =val
    =imaginal>
      -  val  =val
   ==>
    =goal>
      state    start
)

(goal-focus goal)

;(spp respond-h :u 10)
;(spp respond-t :u 10)

(spp attend-letter :reward 0)
(spp done-correct :reward 10)

)

Parameter tuning:
- :ens = 3
- reward of success = 10

Results

R = 0.991
MD = 0.023

Task : Coin Guess

Example experiment functions

Solution

Task

Model

Parameter tuning:

Results