Adaptive tracking for two trial types and tracker reconstruction from .tab

This file shows how to interleave multiple Tracker objects using expyfun.stimuli.TrackerDealer as well as how to reconstruct the dealer from the .tab file logged by experiment controller with expyfun.io.reconstruct_dealer(),

In this case, a modeled human subject generates two curves (one for each trial type: 1 & 2).

@author: maddycapp27

import matplotlib.pyplot as plt
import numpy as np

from expyfun import ExperimentController
from expyfun.analyze import sigmoid
from expyfun.io import reconstruct_dealer
from expyfun.stimuli import TrackerDealer, TrackerUD

# define parameters of modeled subject (using sigmoid probability)
true_thresh = [30, 40]  # true thresholds for trial types 1 and 2
slope = 0.1
chance = 0.5

Defining Tracker Parameters

In this example, the tracker parameters are exactly the same for each instance of the up-down adaptive tracker. These are defined such that the step sizes vary for both up v. down (the up step size is larger by a factor of 3) and based on the number of reversals (the first element in each list is the step size until the number of reversals dictacted by the second element in change_criteria have occurred (i.e. the up step size will be 9 until 5 reversals have occurred, then the up step size will be 3.))

up = 1
down = 1
step_size_up = [9, 3]
step_size_down = [3, 1]
stop_reversals = 30
stop_trials = np.inf
start_value = 45
change_indices = [5]
change_rule = "reversals"
x_min = 0
x_max = 90

# parameters for the tracker dealer
max_lag = 2
pace_rule = "reversals"
rng_dealer = np.random.RandomState(3)  # random seed to select trial type

Initializing and Running Trackers

The two trackers in this example use all of the same parameters and then are passed into the dealer. After the dealer is created, the type of each trial (returned as an index of the array of individual trackers) and trial level for that trial can be acquired. expyfun.ExperimentController is used to generate log files with expyfun.stimuli.TrackerUD and expyfun.stimuli.TrackerDealer information.

std_args = ["test"]  # experiment name
std_kwargs = dict(
    full_screen=False,
    window_size=(1, 1),
    participant="foo",
    session="01",
    stim_db=0.0,
    noise_db=0.0,
    verbose=True,
    version="dev",
)

with ExperimentController(*std_args, **std_kwargs) as ec:
    # initialize two tracker objects--one for each trial type
    tr_ud = [
        TrackerUD(
            ec,
            up,
            down,
            step_size_up,
            step_size_down,
            stop_reversals,
            stop_trials,
            start_value,
            change_indices,
            change_rule,
            x_min,
            x_max,
        )
        for _ in range(2)
    ]

    # initialize TrackerDealer object
    td = TrackerDealer(ec, tr_ud, max_lag, pace_rule, rng_dealer)

    # Initialize human state
    rng_human = np.random.RandomState(1)  # random seed for modeled subject

    for ss, level in td:
        # Get information of which trial type is next and what the level is at
        # that time from TrackerDealer
        td.respond(
            rng_human.rand()
            < sigmoid(level - true_thresh[sum(ss)], lower=chance, slope=slope)
        )

exp_name: test
date: 2024-12-23 22_40_19.997645
file: /home/circleci/project/examples/experiments/tracker_dealer.py
participant: foo
session: 01

Reconstructing the TrackerDealer Object

The TrackerDealer object has many built in analysis functions that are can only be access through the object itself (not the log files alone). By using expyfun.io.reconstruct_dealer(), the object can be recreated such that the analysis functions are accessible. Note that the function always returns a list of objects. Similar reconstructions of single trackers can be done with expyfun.io.reconstruct_tracker().

td_tab = reconstruct_dealer(ec.data_fname)[0]

Plotting the Results

axes = plt.subplots(2, 1)[1]
for i in [0, 1]:
    fig, ax, lines = td_tab.trackers.ravel()[i].plot(ax=axes[i], n_skip=4)

    ax.legend(loc="best")
    ax.set_title(
        f"Adaptive track of model human trial type {i + 1} (true threshold "
        f"is {true_thresh[i]})"
    )
    fig.tight_layout()