The Quest algorithm has been ported directly from the MATLAB sources available with the PsychToolbox. The MATLAB source code was written by Denis Pelli. Thanks to Denis for allowing it to be released under the BSD license to the (Python) world.
This Python version does not depend on the Vision Egg, and may be useful in other contexts.
Download the source directly from the Quest package at the SourceForge download page.
The intensity scale is abstract, but usually we think of it as representing log contrast. Specify true threshold of simulated observer: 0.5 Estimate threshold: 0.4 Trial 1 at 0.7 is right Trial 2 at 0.4 is right Trial 3 at -0.2 is wrong Trial 4 at 0.7 is right Trial 5 at 0.4 is right Trial 6 at 0.4 is wrong Trial 7 at 1.1 is right Trial 8 at 0.9 is right Trial 9 at 0.7 is right Trial 10 at 0.7 is right 36 ms/trial Mean threshold estimate is 0.60 +/- 0.38 Quest beta analysis. Beta controls the steepness of the Weibull function. Now re-analyzing with beta as a free parameter. . . . logC sd beta sd gamma 0.62 0.39 3.7 4.5 0.500 Actual parameters of simulated observer: logC beta gamma 0.50 3.5 0.50
The example above is taken directly from the demo() function of Quest.py and is a direct translation of the demo included in the original MATLAB source:
1 print 'The intensity scale is abstract, but usually we think of it as representing log contrast.' 2 3 tActual = None 4 while tActual is None: 5 sys.stdout.write('Specify true threshold of simulated observer: ') 6 input = raw_input() 7 try: 8 tActual = float(input) 9 except: 10 pass 11 12 tGuess = None 13 while tGuess is None: 14 sys.stdout.write('Estimate threshold: ') 15 input = raw_input() 16 try: 17 tGuess = float(input) 18 except: 19 pass 20 21 tGuessSd = 2.0 # sd of Gaussian before clipping to specified range 22 pThreshold = 0.82 23 beta = 3.5 24 delta = 0.01 25 gamma = 0.5 26 q=QuestObject(tGuess,tGuessSd,pThreshold,beta,delta,gamma) 27 28 # Simulate a series of trials. 29 trialsDesired=100 30 wrongRight = 'wrong', 'right' 31 timeZero=time.time() 32 for k in range(trialsDesired): 33 # Get recommended level. Choose your favorite algorithm. 34 tTest=q.quantile() 35 #tTest=q.mean() 36 #tTest=q.mode() 37 38 tTest=tTest+random.choice([-0.1,0,0.1]) 39 40 # Simulate a trial 41 timeSplit=time.time(); # omit simulation and printing from reported time/trial. 42 response=q.simulate(tTest,tActual) 43 print 'Trial %3d at %4.1f is %s'%(k+1,tTest,wrongRight[response]) 44 timeZero=timeZero+time.time()-timeSplit; 45 46 # Update the pdf 47 q.update(tTest,response); 48 49 # Print results of timing. 50 print '%.0f ms/trial'%(1000*(time.time()-timeZero)/trialsDesired) 51 52 # Get final estimate. 53 t=q.mean() 54 sd=q.sd() 55 print 'Mean threshold estimate is %4.2f +/- %.2f'%(t,sd) 56 #t=QuestMode(q); 57 #print 'Mode threshold estimate is %4.2f'%t 58 59 print '\nQuest beta analysis. Beta controls the steepness of the Weibull function.\n' 60 q.beta_analysis() 61 print 'Actual parameters of simulated observer:' 62 print 'logC beta gamma' 63 print '%5.2f %4.1f %5.2f'%(tActual,q.beta,q.gamma)
- Watson, A. B. and Pelli, D. G. (1983) QUEST: a Bayesian adaptive psychometric method. Percept Psychophys, 33 (2), 113-20.
Pelli, D. G. (1987) The ideal psychometric procedure. Investigative Ophthalmology & Visual Science, 28 (Suppl), 366.
- King-Smith, P. E., Grigsby, S. S., Vingrys, A. J., Benes, S. C., and Supowit, A. (1994) Efficient and unbiased modifications of the QUEST threshold method: theory, simulations, experimental evaluation and practical implementation. Vision Res, 34 (7), 885-912.
The Python Quest package is released under a BSD-style license. (The Vision Egg itself has a LGPL license.)