# Plugins-antibias

From Bcontrol

From the help file (i.e., in Matlab, type "help @antibias/AntibiasSection"):

% [x, y] = AntibiasSection(obj, action, [arg1], [arg2], [arg3]) % % Section that calculates biases and calculates probability of choosing a stimulus % given the previous history. % % Antibias assumes that trials are of two classes, Left desired answer % and Right desired answer, and that their outcome is either Correct or % Incorrect. Given the history of previous trial classes, and the history % of previous corrects/incorrects, Antibias makes a local estimate of % fraction correct for each class, combines that with a prior probability % of making the next trial Left, and produces a recommended probability % for choosing the next trial as Left. Antibias will tend to make the % class with the smaller frac correct the one with the higher probability. % The strength of that tendency is quantified by a parameter, beta. % (See probabilistic_trial_selector.m for details on the math of how the % tendency is generated.) % % Local estimates of fraction correct are computed using an exponential % kernel, most recent trial the most strongly weighted. The tau of this % kernel is a GUI parameter. Two different estimates are computed: one for % use in computing Left probability and Right probability; and a second % simply for GUI display purposes. The two estimates can have different % taus for their kernels. % % GUI DISPLAY: When initialized, this plugin will put up two panels and a % title. In each panel, there is a slider that controls the tau of the % recent-trials exponential kernel. One panel will display the percent % corrects for Right and Left, as computed with its kernel. The second panel % will display the a posteriori probabilities of making the next trial a % "Left" trial or making the next trial a "Right" trial. This second panel % has its own tau slider, and it also has a GUI parameter, beta, that % controls how strongly the history matters. If beta=0, history doesn't % matter, and the a priori LeftProbability dominates. If beta=Inf, then % history matters above all: the next trial will be of the type with lowest % fraction correct, for sure. % % See the bottom of this help file for examples of usage. % % arg1, arg2, arg3 are optional and their meaning depends on action (see % below). % % PARAMETERS: % ----------- % % obj Default object argument. % % action One of: % % 'init' To initialise the plugin and set up the GUI for it. This % action requires two more arguments: The bottom left % x y position (in units of pixels) of where to start placing % the GUI elements of this plugin. % % 'update' This call will recompute both local estimates of % fraction correct, and will recompute the recommended % LeftProb, p(Left). This action requires three more arguments: % HitHist, LProb, a scalar b/w 0 and 1; HitHist, a vector of 1s % SidesHist and 0s and of length n_done_trials where 1 represents % correct and 0 represents incorrect, first element % corresponds to first trial; and SidesHist, a vector of % 'l's and 'r's and of length n_done_trials where 'l' % represents 'left', 'r' represents 'right' first element % corresponds to first trial. % % 'get_posterior_probs' Returns a vector with two components, % [p(Left) p(Right)]. % % % 'update_biashitfrac' This call will recompute the local estimate of fraction % Left correct and fraction Right correct used for % LeftProb, antibiasing, and will also recompute the recommended % HitHist, Left probability. This action % SidesHist, requires three more arguments: LProb, a scalar b/w 0 % and 1; HitHist, a vector of 1s and 0s and of length % n_done_trials where 1 represents correct and 0 % represents incorrect, first element corresponds to % first trial; and SidesHist, a vector of 'l's and 'r's % and of length n_done_trials where 'l' represents % 'left', 'r' represents 'right' first element % corresponds to first trial. % % 'update_hitfrac' This call is not related to computing the posterior % Left probability, but will recompute only the local estimate % HitHist, of fraction correct that is not used for antibiasing. % SidesHist This action requires two more arguments: HitHist, a % vector of 1s and 0s and of length n_done_trials where 1 % represents correct and 0 represents incorrect, first % element corresponds to first trial; and SidesHist, a vector of % 'l's and 'r's and of length n_done_trials where 'l' % represents 'left', 'r' represents 'right' first element % corresponds to first trial. % % 'reinit' Delete all of this section's GUIs and data, % and reinit, at the same position on the same % figure as the original section GUI was placed. % % % x, y Relevant to action = 'init'; they indicate the initial % position to place the GUI at, in the current figure window % % RETURNS: % -------- % % if action == 'init' : % % [x1, y1, w, h] When action == 'init', Antibias will put up GUIs and take % up a certain amount of space of the figure that was current when % AntiBiasSection(obj, 'init', x, y) was called. On return, [x1 y1] % will be the top left corner of the space used; [x y] (as passed % to Antibias in the init call) will be the bottom left corner; % [x+w y1] will be the top right; and [x+w y] will be the bottom % right. h = y1-y. All these are in units of pixels. % % % if action == 'get_posterior_probs' : % % [L R] When action == 'get_posterior_probs', a two-component vector is % returned, with p(Left) and p(Right). If beta=0, then p(Left) % will be the same as the last LeftProb that was passed in. % % % USAGE: % ------ % % To use this plugin, the typical calls would be: % % 'init' : On initializing your protocol, call % AntibiasSection(obj, 'init', x, y); % % 'update' : After a trial is completed, call % AntibiasSection(obj, 'update', LeftProb, HitHist, SidesHist) % % 'get_posterior_probs' : After a trial is completed, and when you are % deciding what kind of trial to make the next trial, get the plugins % opinion on whether the next trial should be Left or Right by calling % AntibiasSection(obj, 'get_posterior_probs') % % See PARAMETERS section above for the documentation of each of these calls. %