can_INST_compensate.m

Are there parameter regimes in which the instance-based model "INST" (implemented in work/models/protoanchor.m) exhibits less assimilation without feedback that it does with feedback under skewed stimulus distributions?

The tendency off all responses to drift upwards without feedback is eliminated in order to see the context-induced effects more clearly.

File: work/models/protoanchor/can_INST_compensate.m
Date: 2007-06-05
Alexander Petrov, http://alexpetrov.com

Contents

Make a test bank of stimulus sequences

Parameters for the INST model

iP = 
        scale: 'LINEAR'
        N_cat: 7
      SM_conv: 1.0000e-03
       cat_sz: 0.0400
       perc_k: 0.0400
        mem_k: 0.0700
        avail: [7x1 double]
    instances: [7x3 double]
      cutoffs: [-2.4000 -0.8000 0.8000 2.4000]
       temper: 0.0400
      history: 0.0400
        alpha: 0.3000
        decay: 1
          ITI: 4

Run the model and calculate ARL profiles

Plot the group ARLs for the INST model

See CMUExper/fdbk1/data/model_ARL_scplot.m

Define Assimilation := ARL(N)-ARL(P)

See CMUExper/fdbk1/data/assimil_profiles.m

Collapse across blocks and print report

Params: H=0.040, T=0.040, d=1.00, c=[-2.4 -0.8 0.8 2.4], 100 runs
-----------------------------------------------------------------
Red line:  fdbk1 = +0.67, nofb1 = +0.88, f-n = -0.21   % fdbk 1st
Blue line: nofb2 = +0.51, fdbk2 = +0.65, f-n = +0.15   % nofb 1st
Interactn: first = +0.16, secnd = -0.22, TTL = -0.06

Scatterplot for the Negative groups 3 & 6. INST model

For Negatively skewed stimuli, higher ARLs indicate assimilation and lower ARLs indicate compensation.

Scatterplot for the Positive groups 2 & 5. INST model

For Negatively skewed stimuli, higher ARLs indicate assimilation and lower ARLs indicate compensation.