def test_split_by_group():
df = pd.DataFrame({
'group': ['a', 'a', 'b', 'b'],
'val': [1.0, 1.0, 2.0, 2.0]
})
data = utils.split_by_group(df, 'group')
assert np.allclose(data['a']['val'].values, 1.0)
assert np.allclose(data['b']['val'].values, 2.0)
def calc_peaks(date_df, model_df, potential_peaked_groups, poly_fit):
df = model_df.merge(date_df)
data = utils.split_by_group(df, 'location')
peak_date = compute_peak_date(potential_peaked_groups,
data,
poly_fit,
time_resolution=0.1)
peak_day = compute_peak_day(potential_peaked_groups,
data,
poly_fit,
time_resolution=0.1)
return data, peak_date, peak_day
def get_peak_detector(self):
self.df_by_group = split_by_group(self.df, self.col_group)
log_derf_obs = []
times = []
peaked = []
self.groups = []
for grp, df in self.df_by_group.items():
if grp in self.peaked_groups or grp in self.not_peaked_groups:
log_derf_obs.append(df[self.col_log_derf_obs].to_numpy())
times.append(df[self.col_t].to_numpy())
if grp in self.peaked_groups:
peaked.append(1)
else:
peaked.append(0)
self.groups.append(grp)
self.peak_detector = PieceWiseLinearPeakDetector(
log_derf_obs, self.groups, times, peaked)
self.peak_detector.train_peak_classifier()
def __init__(self, df, col_obs, col_group, col_est=None, models_dict=None):
if col_est is None and models_dict is None:
raise RuntimeError('must have either a column of estimates or CurveModels to generate estimates')
self.df = df
self.col_obs = col_obs
self.col_group = col_group
self.col_est = col_est
self.models_dict = models_dict
self.df_by_group = split_by_group(self.df, self.col_group)
self.obs_by_group = {}
self.est_by_group = {}
self.groups = []
for grp, df in self.df_by_group.items():
self.obs_by_group[grp] = df[self.col_obs].to_numpy()
if self.col_est is not None:
self.est_by_group[grp] = df[self.col_est].to_numpy()
else:
model = self.models_dict[grp]
self.est_by_group[grp] = model.fun(model.t, model.params)
self.groups.append(grp)