def plot_all_series(data, eps=0.1):
dates = reduce(pd.DatetimeIndex.union, [util.days(d.data.index) for d in data])
ax = prepare_axis(dates)
n = len(data)
intervals = np.column_stack([np.linspace(eps - 0.5, 0.5 - eps, n, False), [(1.0 - 2 * eps) / n] * n])
for interval, d in zip(intervals, data):
d.plot(dates[0], ax, interval)
plt.show()
def plot_all_series(data, eps=0.1):
dates = reduce(pd.DatetimeIndex.union,
[util.days(d.data.index) for d in data])
ax = prepare_axis(dates)
n = len(data)
intervals = np.column_stack([
np.linspace(eps - 0.5, 0.5 - eps, n, False), [(1.0 - 2 * eps) / n] * n
])
for interval, d in zip(intervals, data):
d.plot(dates[0], ax, interval)
plt.show()
def from_sojourns(cls, sojourns, subj):
# FIXME this whole method is horrifying
durs = range(10, 61, 10)
day_idx = util.days(sojourns.raw_data.index)
# FIXME does this still work if we pass in the index?
out = cls.prepare_output().reindex(index=day_idx[:-1])
out_cols = set(out.columns)
out['subject'] = subj
out['day'] = util.WEEKDAYS[out.index.weekday]
out['sleep_ranges'] = map(sleep_ranges,
util.slice_data(sojourns.data, day_idx))
out['total_counts'] = [total(day, 'counts') for day in
util.slice_data(sojourns.raw_data, day_idx)]
if 'steps' in sojourns.raw_data.columns:
out['AG_steps'] = [total(day, 'steps') for day in
util.slice_data(sojourns.raw_data, day_idx)]
if 'AP.steps' in sojourns.raw_data.columns:
out['AP_steps'] = [2*total(day, 'AP.steps') for day in
util.slice_data(sojourns.raw_data, day_idx)]
for classifier in util.classifiers:
grouped = util.group_by_classifier(sojourns.data, classifier)
sliced = util.slice_data(grouped, day_idx)
hours = util.clock_hours(grouped.index)
key = 'min_%s' % classifier.name
if key in out_cols:
out[key] = [delta(day)[day[classifier.cname].notnull()].sum() /
pd.Timedelta(1, 'm')
for day in sliced]
key = '%s_periods' % classifier.name
if key in out_cols:
out[key] = [day[classifier.cname].nunique() for day in sliced]
key = 'mean_%s_len' % classifier.name
# Weight by minute, not by bout.
if key in out_cols:
out[key] = [(delta(day) / delta(day).where(
day[classifier.cname].notnull()).sum() *
day['bout_Dur']).sum() / pd.Timedelta(1, 'm')
for day in sliced]
# Include bouts that extend into neighboring days
key = 'median_%s_len' % classifier.name
if key in out_cols:
out[key] = [day.groupby(classifier.cname).first()['bout_Dur']
.median() / pd.Timedelta(1, 'm')
for day in sliced]
# Incorrectly allow intensity to bleed across midnight because the
# current architecture doesn't allow us to do this right.
key = 'mean_%s_intensity' % classifier.name
if key in out_cols:
out[key] = [delta(day) / delta(day).where(
day[classifier.cname].notnull()).sum() *
day['counts'].sum()
for day in sliced]
# Unused
# This is wrong; it can't be computed from processed data
# key = 'median_%s_intensity' % classifier.name
# if key in out_cols:
# out[key] = [day.ix[day[classifier.cname].notnull(),'counts']
# .median()
# for day in sliced]
for dur in durs:
key = '%s_length_%d' % (classifier.name, dur)
if key in out_cols:
out[key] = [delta(day)[day[classifier.cname].notnull() &
(day['bout_Dur'] >=
pd.Timedelta(dur, 'm'))] \
.sum() / pd.Timedelta(1, 'm')
for day in sliced]
for hh, hour in zip(hours, util.slice_data(grouped, hours)):
# unlike sliced, hour can be empty
key = '%s_circadian_%d' % (classifier.name, hh.hour)
if key in out_cols:
out.ix[pd.Timestamp(hh.date(), tz=util.tz),key] = \
delta(hour)[hour[classifier.cname].notnull()].sum() / \
pd.Timedelta(1, 'm')
for dur in durs:
key = '%s_circadian_%d_length_%d' % (classifier.name,
hh.hour, dur)
if key in out_cols:
out.ix[pd.Timestamp(hh.date(), tz=util.tz),key] = \
delta(hour)[hour[classifier.cname].notnull() &
(hour['bout_Dur'] >=
pd.Timedelta(dur, 'm'))
].sum() / pd.Timedelta(1, 'm')
out['counts_per_min'] = out['total_counts'] / out['min_awake']
out['break_rate'] = (60*out['sedentary_periods'] /
out['min_sedentary'])
return cls(out)