def setUp(self, n_subjects=25, TR=1.5):
self.TR = TR
self.log = logging.getLogger('hb_test')
self.data, self.onsets, self.parameters = (simulate_fmri_experiment(
[{
'name': 'Condition A',
'mu_group': 5,
'std_group': 3
}, {
'name': 'Condition B',
'mu_group': 10,
'std_group': 3
}],
n_subjects=n_subjects,
run_duration=60,
n_trials=10,
TR=TR))
self.hfit = HierarchicalBayesianModel()
df = []
for subject, d in self.data.reset_index().groupby(['subject']):
fitter = nideconv.ResponseFitter(d.signal.values, 1. / self.TR)
fitter.add_event('Condition A',
self.onsets.loc[subject, 'Condition A'].onset,
interval=[0, 20])
fitter.add_event('Condition B',
self.onsets.loc[subject, 'Condition B'].onset,
interval=[0, 20])
self.hfit.add_run(fitter, subject)
cue_pars = {'name':'cue',
'mu_group':-.5, # Slight negative response for cue
'std_group':0,
'onsets':cue_onsets}
stim_pars = {'name':'stim',
'mu_group':1, # Positive response for stimulus presentation
'std_group':0,
'onsets':stim_onsets}
conditions = [cue_pars,
stim_pars]
data, onsets, parameters = simulate.simulate_fmri_experiment(conditions,
run_duration=60,
noise_level=0.05)
##############################################################################
# Underlying data-generating model
# --------------------------------
# Because we simulated the data, we know that the event-related responses should
# exactly follow the *canonical Hemodynamic Response Function* [1]_are
from nideconv.utils import double_gamma_with_d
import numpy as np
plt.figure(figsize=(8, 2.5))
t = np.linspace(0, 20, 100)
ax1 = plt.subplot(121)
plt.title('Ground truth cue-related response')
# fit a model, and see how well our model fits the data
from nideconv import simulate
from nideconv import ResponseFitter
conditions = [{
'name': 'Condition A',
'mu_group': 5,
'std_group': 1,
'onsets': [0, 20]
}]
# Simulate data with very short run time and TR for illustrative purposes
data, onsets, pars = simulate.simulate_fmri_experiment(conditions,
TR=0.2,
run_duration=40,
noise_level=.2,
n_rois=1)
# Make ResponseFitter-object to fit these data
rf = ResponseFitter(input_signal=data,
sample_rate=5) # Sample rate is inverse of TR (1/TR)
rf.add_event('Condition A',
onsets.loc['Condition A'].onset,
interval=[0, 20],
basis_set='canonical_hrf')
rf.fit()
rf.plot_model_fit()
"""
from nideconv import simulate
conditions = [{
'name': 'Correct',
'mu_group': .25,
'std_group': .1,
'n_trials': 16
}, {
'name': 'Error',
'mu_group': .5,
'std_group': .1,
'n_trials': (1, 6)
}]
data, onsets, pars = simulate.simulate_fmri_experiment(conditions,
n_subjects=9,
n_runs=1,
TR=1.5)
##############################################################################
# First, we fit this data using the traditional frequentist GLM with Fourier
# basis functions:
from nideconv import GroupResponseFitter
gmodel = GroupResponseFitter(data,
onsets,
input_sample_rate=1 / 1.5,
concatenate_runs=False)
gmodel.add_event('Correct',
basis_set='fourier',
n_regressors=9,
interval=[0, 21])
gmodel.add_event('Error',
"""
Deconvolution on group of subjects
==================================
The `GroupResponseFitter`-object of `Nideconv` offers an easy way to fit the data of many subjects together
well-specified model
`nideconv.simulate` can simulate data from multiple subjects.
"""
from nideconv import simulate
data, onsets, pars = simulate.simulate_fmri_experiment(n_subjects=8,
n_rois=2,
n_runs=2)
##############################################################################
# Now we have an indexed Dataframe, `data`, that contains time series for every
# subject, run, and ROI:
print(data.head())
##############################################################################
#
print(data.tail())
##############################################################################
# We also have onsets for every subject and run:
print(onsets.head())
##############################################################################
#
##############################################################################
print(onsets.tail())