def calc_trial_like(self, trial, save_post=False):
# Compute likelihood from reaction time
# see what response was made and map it to the choice
if trial.Choice == 'Gamble':
choice = np.array([1])
elif trial.Choice == 'Certain':
choice = np.array([2])
else:
# they made no choice
# we could consider skipping these
choice = np.array([0])
# calc the like
if self.ignore_non_resp and choice == np.array([0]):
log_like = 0.0
else:
# calc the log like
log_like = np.log(
wfpt_like(choice,
np.array([trial.RT]),
v_mean=trial['Ediff'],
a=self.params['a'],
w_mode=self.params['w'],
t0=self.params['t0'],
nsamp=self.wfpt_nsamp,
max_time=self.max_time,
trange_nsamp=self.trange_nsamp))[0]
# * * * * * * * * * * * * * * * * * * * * * * * * * * * *
# if the trial is also a mood trial we could also add in a like calc
# for a model, too
# * * * * * * * * * * * * * * * * * * * * * * * * * * * *
if np.isnan(trial['Mood']):
mood_log_like = 0.0
if not np.isnan(trial['Mood']):
curr_mood = dists.logit(trial['Mood'] / 1000)
pred_mood = self.params['b'] + \
self.params['w_LTA'] * trial['LTA_sum'] + \
self.params['w_RPE'] * trial['RPE_sum']
log_like = log_like + np.log(
norm.pdf(curr_mood, pred_mood, np.sqrt(self.params['s_v'])))
##Mood_log_likelihood
mood_log_like = np.log(
norm.pdf(curr_mood, pred_mood, np.sqrt(self.params['s_v'])))
# see if running conditional sim
if save_post:
# run wfpt_gen
choices, rts = wfpt_gen(v_mean=trial['Ediff'],
a=self.params['a'],
w_mode=self.params['w'],
wfpt_nsamp=self.wfpt_nsamp,
nsamp=self.gen_nsamp,
trange=np.linspace(
0, self.max_time - self.params['t0'],
self.trange_nsamp))
# calc prob of making the observed choice
ind = choices == choice
p_choice = ind.mean()
# calc mean log rt
choice_mean_log_rt = np.log(rts[ind] + self.params['t0']).mean()
return log_like, p_choice, choice_mean_log_rt
return log_like, mood_log_like
def calc_trial_like(self, trial, save_post=False):
# Compute likelihood from reaction time
# see what response was made and map it to the choice
if trial.Choice == 'Gamble':
choice = np.array([1])
elif trial.Choice == 'Certain':
choice = np.array([0])
else:
# they made no choice
# we could consider skipping these
choice = np.array([2])
# calc the like
if self.ignore_non_resp and choice == np.array([2]):
log_like = 0.0
else:
#### Exclusion of RT (likelihood depends only on choices)
p_c = dists.invlogit(trial['v_mean']) # prob. of making a choice
likelihood = (p_c**choice) * ((1 - p_c)**(1 - choice))
log_like = np.log(likelihood)
## Debug
#import pdb; pdb.set_trace()
# calc the log like
# log_like = np.log(wfpt_like(choice, np.array([trial.RT]),
# v_mean=trial['Ediff'], a=self.params['a'],
# w_mode=self.params['w'], t0=self.params['t0'],
# nsamp=self.wfpt_nsamp,
# max_time=self.max_time,
# trange_nsamp=self.trange_nsamp))[0]
# * * * * * * * * * * * * * * * * * * * * * * * * * * * *
# if the trial is also a mood trial we could also add in a like calc
# for a model, too
# * * * * * * * * * * * * * * * * * * * * * * * * * * * *
if np.isnan(trial['Mood']):
mood_log_like = 0.0
if not np.isnan(trial['Mood']):
curr_mood = dists.logit(trial['Mood'] / 1000)
### Modification for P_win
pred_mood = self.params['b'] + self.params['w_p'] * trial['P_win']
log_like = log_like + np.log(
norm.pdf(curr_mood, pred_mood, np.sqrt(self.params['s_v'])))
##Mood_log_likelihood
mood_log_like = np.log(
norm.pdf(curr_mood, pred_mood, np.sqrt(self.params['s_v'])))
# see if running conditional sim
if save_post:
# run wfpt_gen
choices, rts = wfpt_gen(v_mean=trial['Ediff'],
a=self.params['a'],
w_mode=self.params['w'],
wfpt_nsamp=self.wfpt_nsamp,
nsamp=self.gen_nsamp,
trange=np.linspace(
0, self.max_time - self.params['t0'],
self.trange_nsamp))
# calc prob of making the observed choice
ind = choices == choice
p_choice = ind.mean()
# calc mean log rt
choice_mean_log_rt = np.log(rts[ind] + self.params['t0']).mean()
return log_like, p_choice, choice_mean_log_rt
return log_like, mood_log_like
def calc_trial_like(self, trial, save_post=False):
# Compute likelihood from reaction time
# see what response was made and map it to the choice
if trial.Choice == 'Gamble':
choice = np.array([1])
elif trial.Choice == 'Certain':
choice = np.array([2])
else:
# they made no choice
# we could consider skipping these
choice = np.array([0])
# calc the like
if self.ignore_non_resp and choice == np.array([0]):
log_like = 0.0
else:
###### Debug ########
# Reac_time= np.array([trial.RT])
# v_mean=trial['Ediff']
# GAMMA=self.params['gamma']
# BETA = self.params['beta']
# A=self.params['a']
# W=self.params['w']
# T0=self.params['t0']
# P0=self.params['p0']
# Sv=self.params['s_v']
# LAMBDA=self.params['lambda']
# WLTA=self.params['w_LTA']
# WRPE=self.params['w_RPE']
# B=self.params['b']
# Trange_nsamp=self.trange_nsamp
# # calc the log like
# W1=self.params['w']
# print('w parameter line 138(calc_trial_like):',W1,flush=True)
log_like = np.log(
wfpt_like(choice,
np.array([trial.RT]),
v_mean=trial['Ediff'],
a=self.params['a'],
w_mode=self.params['w'],
t0=self.params['t0'],
nsamp=self.wfpt_nsamp,
max_time=self.max_time,
trange_nsamp=self.trange_nsamp))[0]
#### Debug
# if np.isnan(log_like):
# print((GAMMA, BETA, W, A, T0, P0, LAMBDA, WLTA, WRPE, B, Sv), flush=True)
# print((Reac_time,v_mean), flush=True)
# print((choice,Trange_nsamp), flush=True)
# * * * * * * * * * * * * * * * * * * * * * * * * * * * *
# if the trial is also a mood trial we could also add in a like calc
# for a model, too
# * * * * * * * * * * * * * * * * * * * * * * * * * * * *
if np.isnan(trial['Mood']):
mood_log_like = 0.0
if not np.isnan(trial['Mood']):
curr_mood = dists.logit(trial['Mood'] / 1000)
pred_mood = self.params['b'] + \
self.params['w_LTA'] * trial['LTA_sum'] + \
self.params['w_RPE'] * trial['RPE_sum']
log_like = log_like + np.log(
norm.pdf(curr_mood, pred_mood, np.sqrt(self.params['s_v'])))
##Mood_log_likelihood
mood_log_like = np.log(
norm.pdf(curr_mood, pred_mood, np.sqrt(self.params['s_v'])))
# see if running conditional sim
if save_post:
# run wfpt_gen
choices, rts = wfpt_gen(v_mean=trial['Ediff'],
a=self.params['a'],
w_mode=self.params['w'],
wfpt_nsamp=self.wfpt_nsamp,
nsamp=self.gen_nsamp,
trange=np.linspace(
0, self.max_time - self.params['t0'],
self.trange_nsamp))
# calc prob of making the observed choice
ind = choices == choice
p_choice = ind.mean()
# calc mean log rt
choice_mean_log_rt = np.log(rts[ind] + self.params['t0']).mean()
return log_like, p_choice, choice_mean_log_rt
return log_like, mood_log_like