def printSimResults(results): stopCurve, goRTs, ssrts = utils.stop_curve(results) print "Stop Curve:" print stopCurve print "GO RT: %0.3f " % (goRTs)
def simulate(theta, mean, std, ntrials=20000, timebound=.650, learn=False, acc=None, ssv_decay=False, sim=False):
#########################################################
# TESTING LEARN FX #
#########################################################
"""
Notes for learning:
*set learn=True
*if you want the accumulated evidence vectors set return_all=True
*IMPORTANT: theta needs to contain parameters
'cor_lr': learning rate following correct trials
'err_lr': learning rate following error trials
*IMPORTANT: acc is a binary vector containing the accuracy for each trial
* when you're fitting (using fitre.fit_reactive_data())
then it will use the subject's actual accuracy data
* if you're just simulating with this function then you need
to provide this vector
I'd reccomend pulling an example from the subject dataframe
when you're in preliminary testing stages (df = your subject dataframe)
To do this get your accuracy vector from some subject (e.g. subject 12)
example_acc=df[df['idx']==12]['acc'].values
ntrials=len(example_acc)
theta={'pGo':.5, 'a':0.4, 'z':.19, 'v':1.1, 't':0.235, 'ssv':-1, 'cor_lr':.09, err_lr:-.11}
Then set fitre.simulate(theta, ntrials=ntrials, acc=example_acc, learn=True, return_all=True)
The output will be a pandas dataframe. The df will have a column 'v' containing the drift-rate
for that trial, you can use this column to bin drift-rates and look at the accuracy, rt info for each
*once you have good estimates for cor_lr and err_lr,
try optimizing to set of behavioral data using fitre.fit_reactive_data()
"""
acc = []
columns = ['rt', 'choice', 'response', 'acc', 'trial_type', 'ssd', 'v', 'probe'] #pGo
mfx= fitfx_learn.radd_fitfx
if learn:
learnfrom={'previous':0, 'time_dif' : 0, 'prev_response': 0}
else:
learnfrom=None
df = pd.DataFrame(columns=columns, index=np.arange(0,ntrials))
for i in xrange(ntrials):
(ttype,ssd,is_probe) = utils.generate_trial(mean,std)
theta['ssd'] = ssd
if learn and i == 0:
sim_out = mfx(theta, timebound=timebound, ttype=ttype, learnfrom=None)
else:
sim_out = mfx(theta, timebound=timebound, ttype=ttype, learnfrom=learnfrom)
sim_out['probe'] = is_probe
df.loc[i]=pd.Series({c:sim_out[c] for c in df.columns})
acc.append(sim_out['acc'])
if learn:
learnfrom['prev_outcome'] = int(sim_out['acc'])
learnfrom['time_dif'] = 0.500 - sim_out['rt']
learnfrom['prev_response'] = sim_out['response']
df[['rt', 'acc', 'response','v']] = df[['rt', 'acc', 'response', 'v']].astype(float)
if sim:
return df
else:
#get mean vectors
sacc,rt,sem =utils.stop_curve(df)
go_rt=utils.get_rtcor(df, tb=timebound)*10
return sacc, go_rt
