def data(num):
""" Run a single subjects data. """
sdir = subdir(num)
roi_names = [
"wartaskAB_Right Accumbens_str2.nii.gz",
"wartaskAB_Right Accumbens.nii.gz",
"wartaskAB_Right Caudate.nii.gz"]
# --
# Get that Ss data and trial information.
sdata = get_behave_data(num)
sdata.update(get_similarity_data(num))
sdata.update(get_rl_data(num))
trials = get_trials_combined()
durations = get_durations()
# --
# Go!
roi_results = []
for name in roi_names:
print(name)
spath = os.path.join(sdir, name)
# Init this roi's models
roiglm = fmri.catreward.roi.base.Catreward(
1.5, spath, trials, durations, sdata)
# Get, reformat (extract), and store the results.
roi_results.append(roiglm.run(name))
return roi_results
def meandata(num):
""" Run a single subjects data, drawing the BOLD data
from the /bold dir text files. """
sdir = subdir(num)
roi_names = [
"wartaskAB_Right Accumbens_str2_bold.txt",
"wartaskAB_Right Accumbens_bold.txt",
"wartaskAB_Right Caudate_bold.txt"]
# --
# Get that Ss data and trial information.
sdata = get_behave_data(num)
sdata.update(get_similarity_data(num))
sdata.update(get_rl_data(num))
trials = get_trials()
durations = get_durations()
# --
# Go!
roi_results = []
for name in roi_names:
print(name)
spath = os.path.join(sdir, 'bold', name)
# Init this roi's models
roiglm = fmri.catreward.roi.exps.base.CatMean(
1.5, spath, trials, durations, sdata)
# Get, reformat (extract), and store the results.
roi_results.append(roiglm.run(name))
return roi_results
def run(num, roi_class_name, trials_name):
""" For subject <num>, run all models for all ROIs.
Note: Must be run from the parent folder of all the fMRI datasets."""
sdir = subdir(num)
roi_class_path = roi_class_name.split('.')
txt_classes = ['CatMean', 'Nobox', 'CatMeanFir', 'Subtime']
if roi_class_name == 'Catreward':
roi_names = get_roi_names(kind='nii')
elif roi_class_name in txt_classes:
roi_names = get_roi_names(kind='txt')
sdir = os.path.join(sdir, 'bold')
## Swich to the bold dir too
else:
raise ValueError("Could not find suitable ROI names.")
# --
# Get that Ss data and trial information.
sdata = get_behave_data(num)
sdata.update(get_similarity_data(num))
sdata.update(get_rl_data(num))
durations = get_durations()
trials = None
if trials_name == 'get_trials':
trials = get_trials()
elif trials_name == 'get_trials_combined':
trials = get_trials_combined()
else:
raise ValueError("trials_name was unkown.")
# --
# Go!
results = []
for name in roi_names:
print(name)
spath = os.path.join(sdir, name)
# Init this roi's models
Roiclass = getattr(bclasses, roi_class_name)
roiglm = Roiclass(1.5, spath, trials, durations, sdata)
# Get, reformat (extract), and store the results.
# And del it to keep memory reasonable
results.append(roiglm.run(name))
return results
def main(num, roi_names, roi_class_name):
""" The main worker for run3.py.
<num> is a valid subject number code (101-118)
<roi_names> is the names of the rois you wish to analyze
<roi_class_name> is the name of the fmri.catreward.roi.exps.base
class you want to use for the experiment.
This script saves each Ss data to seperate files, returning nothing. """
# --
# Get that Ss data and trial information.
trials = get_trials_combined()
durations = get_durations()
sdata = get_behave_data(num)
sdata.update(get_similarity_data(num))
sdata.update(get_rl_data(num))
sdata.update(get_cumrewards(num))
## Add all data to sdata, a dict
# --
# The fMRI data is located at:
sdir = subdir(num)
# --
results = []
for name in roi_names:
print(name)
spath = os.path.join(sdir, 'bold', name)
# Init this roi's models
Roiclass = getattr(bclasses, roi_class_name)
roiglm = Roiclass(1.5, spath, trials, durations, sdata)
# Get, reformat (extract), and store the results.
# And del it to keep memory reasonable
results.append(roiglm.run(name))
return results
def run(num, similarity_name=None, reward_name="acc", alpha=0.3):
""" Run a RL model based on <num>'s data, <similarity_name> is
the of the similarity metric data you want to use, see
fmri.catreward.roi.data.get_similarity_data() for details.
<reward_name> is the name of the data to be used as rewards in the
model. Options are 'acc' ({0,1}, i.e. behavioral accuracy or
'gl' ({-1,1}, short for gain/lose).
Returns:
-------
values, rpes, and sim_rewards (as a list) tied to the trial
structure returned by fmri.catreward.roi.data.get_trials() """
# --
# Get subject's data
sdata = get_behave_data(num)
sdata.update(get_similarity_data(num))
responses = np.array(sdata["resp"])
rewards = None
if reward_name == "acc":
rewards = np.array(sdata["acc"], dtype=np.float32)
elif reward_name == "gl":
rewards = np.array(sdata["gl"], dtype=np.float32)
trials = np.array(fmri.catreward.roi.data.get_trials())
conds = list(set(trials))
## conds are the unqie entries in trials
values = np.zeros_like(trials, dtype=np.float32)
rpes = np.zeros_like(trials, dtype=np.float32)
sim_rewards = np.zeros_like(trials, dtype=np.float32)
## Returned....
# Each cond has n states,
# matching the number of
# responses (approx 2: {1,6}).
#
# Wrong button presses
# are included, however these
# are never rewarded so stay at 0.
for cond in conds:
if cond == 0:
continue
## Drop jitter.
# Create states and their rewards.
mask = trials == cond
states_c = responses[mask]
rewards_c = rewards[mask] ## _c for cond...
# Get the RL alg we want to run.
# based on similarity_name
if similarity_name == None:
# No similarity:
values_c, rpes_c = rl.reinforce.b_delta(rewards_c, states_c, alpha)
sim_rewards_c = rewards_c
## To give a consistent return
## just map rewards to sim_rewards
else:
# Get the similarity data, filter it by mask, and run RL.
similarity_c = np.array(sdata[similarity_name])[mask]
values_c, rpes_c, sim_rewards_c = rl.reinforce.b_delta_similarity(rewards_c, states_c, similarity_c, alpha)
# sim_rewards_c does not need to be
# unpacked when similarity_name is None
sim_rewards_c = rl.misc.unpack(sim_rewards_c, states_c)
# Unpack values and rpes
# based on states_c
values_c = rl.misc.unpack(values_c, states_c)
rpes_c = rl.misc.unpack(rpes_c, states_c)
# Now use the mask to map values_c, etc,
# into trials space
values[mask] = values_c
rpes[mask] = rpes_c
sim_rewards[mask] = sim_rewards_c
return values.tolist(), rpes.tolist(), sim_rewards.tolist()
def fit(num, similarity_name=None, reward_name="acc", res=0.05):
""" Fit <model_name> on <num>'s data, returning first the
model fit score (liklihood) and fit parameters - alpha and beta.
<res> is the parameter resolution (0-1).
For available <model_name>s see rl.fit. """
# --
# Get subject's data
sdata = get_behave_data(num)
sdata.update(get_similarity_data(num))
responses = np.array(sdata["resp"])
rewards = None
if reward_name == "acc":
rewards = np.array(sdata["acc"], dtype=np.float32)
elif reward_name == "gl":
rewards = np.array(sdata["gl"], dtype=np.float32)
trials = np.array(fmri.catreward.roi.data.get_trials())
conds = list(set(trials))
## conds are the uniqu entries in trials
# Each cond has n states,
# matching the number of
# responses (approx 2: {1,6}).
#
# Wrong button presses
# are included, however these
# are never rewarded so stay at 0.
params = None
log_L = 0
for cond in conds:
print(cond)
if cond == 0:
continue
## Drop jitter.
# Create states and their rewards.
mask = trials == cond
states_c = responses[mask]
rewards_c = rewards[mask] ## _c for cond...
# Get the RL alg we want to run.
# based on similarity_name
params_c = None
log_L_c = None
if similarity_name == None:
# No similarity, so just fit:
params_c, log_L_c = rl.fit.ml_delta(rewards_c, states_c, res)
else:
# Get the similarity data, filter it by mask, and fit.
similarity_c = np.array(sdata[similarity_name])[mask]
params_c, log_L_c = rl.fit.ml_delta_similarity(rewards_c, states_c, similarity_c, res)
# Add cond log_L_c to the overall log_L score
log_L += log_L_c
# params is the average for all conds
if params == None:
params = deepcopy(params_c)
params = (np.array(params_c) + np.array(params)) / 2.0
return tuple(params), log_L
