def innermost(self, *args, **kwargs):
from datamatrix import convert as cnv
df_in = cnv.to_pandas(self)
fnc = getattr(df_in, function_name)
df_out = fnc(*args, **kwargs)
return df_out
def test_convert():
refdm = DataMatrix(length=3)
refdm[u'tést'] = 1, 2, u''
refdm.B = u'mathôt', u'b', u'x'
refdm.C = u'a,\\b"\'c', 8, u''
testdm = io.readtxt('testcases/data/data.csv')
check_dm(refdm, testdm)
testdm = cnv.from_json(cnv.to_json(testdm))
check_dm(refdm, testdm)
testdf = cnv.to_pandas(testdm)
testdm = cnv.from_pandas(testdf)
check_dm(refdm, testdm)
print(np.array(testdf))
print(np.array(testdm))
assert (np.array(testdf, dtype=str) == np.array(testdm, dtype=str)).all()
def corr_img(img, trace, mcparams, xyz, deconv_bold):
"""Get a per-voxel correlation image between a nifti image and a trace"""
pcor = np.empty(img.shape[:-1])
pcor[:] = np.nan
imgdat = img.get_data()
dm = mcparams[SKIP_FIRST:len(trace)]
dm.bold = FloatColumn
dm.trace = FloatColumn
dm.trace = trace[SKIP_FIRST:]
df = cnv.to_pandas(dm)
for i, (x, y, z) in enumerate(zip(*xyz)):
if DOWNSAMPLE and (x % 2 or y % 2 or z % 2):
continue
df.bold = flt(imgdat[x, y, z])[SKIP_FIRST:len(trace)]
if deconv_bold:
df.bold = deconv_prf(df.bold)
results = smf.ols(FORMULA_VOXEL_TRACE, df).fit()
pcor[x, y, z] = results.tvalues[1]
if DOWNSAMPLE:
pcor[x + 1, y + 1, z + 1] = results.tvalues[1]
return nib.Nifti2Image(pcor, img.affine)
def __dataframe__(self):
from datamatrix import convert as cnv
return cnv.to_pandas(self)
def do_run(sub, run, row, xyz):
print(
'Starting sub: {}, run: {}, voxels: {}'.format(
sub,
run,
len(xyz[0])
)
)
img = get_avmovie_data(sub, run)
traces = {
'pupil': get_pupil_data(sub, run),
'arousal': get_arousal_data(run), # doesn't depend on subject
'luminance': get_luminance_data(sub, run),
'change': get_change_data(sub, run)
}
mcparams = get_mcparams(sub, run)
# Per-voxel correlationsx, y, z
for tracename, trace in traces.items():
row['t_bold_{}'.format(tracename)] = corr_img(
img,
trace,
mcparams,
xyz,
deconv_bold=tracename == 'pupil' # PRF deconvolution
)
# Per-ROI correlations
voxels = img.get_data()[xyz[0], xyz[1], xyz[2], :]
avg = np.nanmean(voxels, axis=0)
# Create a DataFrame with equally long traces, also including the
# average bold value
trace_len = min(len(trace) for trace in traces.values())
print('trace length: {}'.format(trace_len))
dm = mcparams[SKIP_FIRST:trace_len]
dm.bold = FloatColumn
dm.bold = avg[SKIP_FIRST:trace_len]
for tracename, trace in traces.items():
dm[tracename] = FloatColumn
dm[tracename] = trace[SKIP_FIRST:trace_len]
df = cnv.to_pandas(dm)
# Correlate each trace with the average bold signal
for tracename in traces:
results = smf.ols(
FORMULA_BOLD_TRACE.format(tracename),
data=df
).fit()
t = results.tvalues[1]
print('t(boldavg ~ {}) = {:.4f}'.format(tracename, t))
row['t_boldavg_{}'.format(tracename)] = t
# Correlate each trace with each other trace
for tracename1, tracename2 in itertools.product(TRACES, TRACES):
if tracename1 <= tracename2:
continue
results = smf.ols(
FORMULA_TRACE_TRACE.format(tracename1, tracename2),
data=df
).fit()
t = results.tvalues[1]
print('t({} ~ {}) = {:.4f}'.format(tracename1, tracename2, t))
row['t_{}_{}'.format(tracename1, tracename2)] = t
row.run = run