def test_mle_fit_psycho(self):
expected = {
'weibull50': (np.array([0.0034045, 3.9029162, .1119576]), -334.1149693046583),
'weibull': (np.array([0.00316341, 1.72552866, 0.1032307]), -261.235178611311),
'erf_psycho': (np.array([-9.78747259, 10., 0.15967605]), -193.0509031440323),
'erf_psycho_2gammas': (np.array([-11.45463779, 9.9999999, 0.24117732, 0.0270835]),
-147.02380025592902)
}
for model in self.test_data.keys():
pars, L = psy.mle_fit_psycho(self.test_data[model], P_model=model, nfits=10)
expected_pars, expected_L = expected[model]
self.assertTrue(np.allclose(expected_pars, pars, atol=1e-3),
f'unexpected pars for {model}')
self.assertTrue(np.isclose(expected_L, L, atol=1e-3),
f'unexpected likelihood for {model}')
# Test on of the models with function pars
params = {
'parmin': np.array([-5., 10., 0.]),
'parmax': np.array([5., 15., .1]),
'parstart': np.array([0., 11., 0.1]),
'nfits': 5
}
model = 'erf_psycho'
pars, L = psy.mle_fit_psycho(self.test_data[model], P_model=model, **params)
expected = [-5, 15, 0.1]
self.assertTrue(np.allclose(expected, pars, rtol=.01), f'unexpected pars for {model}')
self.assertTrue(np.isclose(-195.55603, L, atol=1e-5), f'unexpected likelihood for {model}')
def compute_psychometric(trials,
signed_contrast=None,
block=None,
plotting=False):
"""
Compute psychometric fit parameters for trials object
:param trials: trials object that must contain contrastLeft, contrastRight and probabilityLeft
:type trials: dict
:param signed_contrast: array of signed contrasts in percent, where -ve values are on the left
:type signed_contrast: np.array
:param block: biased block can be either 0.2 or 0.8
:type block: float
:return: array of psychometric fit parameters - bias, threshold, lapse high, lapse low
"""
if signed_contrast is None:
signed_contrast = get_signed_contrast(trials)
if block is None:
block_idx = np.full(trials.probabilityLeft.shape, True, dtype=bool)
else:
block_idx = trials.probabilityLeft == block
if not np.any(block_idx):
return np.nan * np.zeros(4)
prob_choose_right, contrasts, n_contrasts = compute_performance(
trials, signed_contrast=signed_contrast, block=block, prob_right=True)
if plotting:
psych, _ = psy.mle_fit_psycho(np.vstack(
[contrasts, n_contrasts, prob_choose_right]),
P_model='erf_psycho_2gammas',
parstart=np.array([0., 40., 0.1, 0.1]),
parmin=np.array([-50., 10., 0., 0.]),
parmax=np.array([50., 50., 0.2, 0.2]),
nfits=10)
else:
psych, _ = psy.mle_fit_psycho(
np.vstack([contrasts, n_contrasts, prob_choose_right]),
P_model='erf_psycho_2gammas',
parstart=np.array([np.mean(contrasts), 20., 0.05, 0.05]),
parmin=np.array([np.min(contrasts), 0., 0., 0.]),
parmax=np.array([np.max(contrasts), 100., 1, 1]))
return psych
def fit_psychfunc(df):
choicedat = df.groupby('signed_contrast').agg({
'choice': 'count',
'choice2': 'mean'
}).reset_index()
if len(choicedat) >= 4: # need some minimum number of unique x-values
pars, L = psy.mle_fit_psycho(
choicedat.values.transpose(),
P_model='erf_psycho_2gammas',
parstart=np.array([0, 20., 0.05, 0.05]),
parmin=np.array([choicedat['signed_contrast'].min(), 5, 0., 0.]),
parmax=np.array([choicedat['signed_contrast'].max(), 40., 1, 1]))
else:
pars = [np.nan, np.nan, np.nan, np.nan]
df2 = {
'bias': pars[0],
'threshold': pars[1],
'lapselow': pars[2],
'lapsehigh': pars[3]
}
df2 = pd.DataFrame(df2, index=[0])
df2['ntrials'] = df['choice'].count()
return df2
def compute_psychometric(trials, signed_contrast=None, block=None):
"""
Compute psychometric fit parameters for trials object
:param trials: trials object that must contain contrastLeft, contrastRight and probabilityLeft
:type trials: dict
:param signed_contrast: array of signed contrasts in percent, where -ve values are on the left
:type signed_contrast: np.array
:param block: biased block can be either 0.2 or 0.8
:type block: float
:return: array of psychometric fit parameters - bias, threshold, lapse high, lapse low
"""
if signed_contrast is None:
signed_contrast = get_signed_contrast(trials)
if block is None:
block_idx = np.full(trials.probabilityLeft.shape, True, dtype=bool)
else:
block_idx = trials.probabilityLeft == block
contrasts, n_contrasts = np.unique(signed_contrast[block_idx], return_counts=True)
rightward = trials.choice == -1
# Calculate the proportion rightward for each contrast type
prob_choose_right = np.vectorize(lambda x: np.mean(rightward[(x == signed_contrast) &
block_idx]))(contrasts)
psych, _ = psy.mle_fit_psycho(
np.vstack([contrasts, n_contrasts, prob_choose_right]),
P_model='erf_psycho_2gammas',
parstart=np.array([np.mean(contrasts), 20., 0.05, 0.05]),
parmin=np.array([np.min(contrasts), 0., 0., 0.]),
parmax=np.array([np.max(contrasts), 100., 1, 1]))
return psych
def plot_psychometric(x, y, subj, **kwargs):
# summary stats - average psychfunc over observers
df = pd.DataFrame({
'signed_contrast': x,
'choice': y,
'choice2': y,
'subject_nickname': subj
})
df2 = df.groupby(['signed_contrast', 'subject_nickname']).agg({
'choice2':
'count',
'choice':
'mean'
}).reset_index()
df2.rename(columns={
"choice2": "ntrials",
"choice": "fraction"
},
inplace=True)
df2 = df2.groupby(['signed_contrast']).mean().reset_index()
df2 = df2[['signed_contrast', 'ntrials', 'fraction']]
# only 'break' the x-axis and remove 50% contrast when 0% is present
# print(df2.signed_contrast.unique())
if 0. in df2.signed_contrast.values:
brokenXaxis = True
else:
brokenXaxis = False
# fit psychfunc
pars, L = psy.mle_fit_psycho(
df2.transpose().values, # extract the data from the df
P_model='erf_psycho_2gammas',
parstart=np.array([0, 20., 0.05, 0.05]),
parmin=np.array([df2['signed_contrast'].min(), 5, 0., 0.]),
parmax=np.array([df2['signed_contrast'].max(), 40., 1, 1]))
if brokenXaxis:
# plot psychfunc
g = sns.lineplot(np.arange(-27, 27),
psy.erf_psycho_2gammas(pars, np.arange(-27, 27)),
**kwargs)
# plot psychfunc: -100, +100
sns.lineplot(np.arange(-36, -31),
psy.erf_psycho_2gammas(pars, np.arange(-103, -98)),
**kwargs)
sns.lineplot(np.arange(31, 36),
psy.erf_psycho_2gammas(pars, np.arange(98, 103)),
**kwargs)
# if there are any points at -50, 50 left, remove those
if 50 in df.signed_contrast.values or -50 in df.signed_contrast.values:
df.drop(df[(df['signed_contrast'] == -50.) |
(df['signed_contrast'] == 50)].index,
inplace=True)
# now break the x-axis
df['signed_contrast'] = df['signed_contrast'].replace(-100, -35)
df['signed_contrast'] = df['signed_contrast'].replace(100, 35)
else:
# plot psychfunc
g = sns.lineplot(np.arange(-103, 103),
psy.erf_psycho_2gammas(pars, np.arange(-103, 103)),
**kwargs)
df3 = df.groupby(['signed_contrast', 'subject_nickname']).agg({
'choice2':
'count',
'choice':
'mean'
}).reset_index()
# plot datapoints with errorbars on top
if df['subject_nickname'].nunique() > 1:
# put the kwargs into a merged dict, so that overriding does not cause an error
sns.lineplot(
df3['signed_contrast'], df3['choice'], **{
**{
'err_style': "bars",
'linewidth': 0,
'linestyle': 'None',
'mew': 0.5,
'marker': 'o',
'ci': 68
},
**kwargs
})
if brokenXaxis:
g.set_xticks([-35, -25, -12.5, 0, 12.5, 25, 35])
g.set_xticklabels(['-100', '-25', '-12.5', '0', '12.5', '25', '100'],
size='small',
rotation=60)
g.set_xlim([-40, 40])
break_xaxis(y=-0.004)
else:
g.set_xticks([-100, -50, 0, 50, 100])
g.set_xticklabels(['-100', '-50', '0', '50', '100'],
size='small',
rotation=60)
g.set_xlim([-110, 110])
g.set_ylim([0, 1.02])
g.set_yticks([0, 0.25, 0.5, 0.75, 1])
g.set_yticklabels(['0', '25', '50', '75', '100'])