def discretize(self, n_bins=3, inplace=False):
"""Retruns a discretized pandas.Series
Args:
n_bins (int): Number of bins or steps to discretize the function
inplace (bool): if True, perform operation in-place
"""
try:
from scipy.optimize import minimize
from itertools import chain
except ImportError:
raise ImportError("The sklearn package must be installed to "
"use this optional feature.")
if self.archetypes:
# if multiindex, group and apply operation on each group.
# combine at the end
results = {}
edges = {}
ampls = {}
for name, sub in self.groupby(level=0):
hour_of_min = sub.time_at_min[1]
sf = [1 / (i * 1.01) for i in range(1, n_bins + 1)]
sf.extend([sub.min()])
sf_bounds = [(0, sub.max()) for i in range(0, n_bins + 1)]
hours = [
hour_of_min - hour_of_min * 1 / (i * 1.01)
for i in range(1, n_bins + 1)
]
# Todo hours need to work fow datetime index
hours.extend([len(sub)])
hours_bounds = [(0, len(sub)) for i in range(0, n_bins + 1)]
start_time = time.time()
log("discretizing EnergySeries {}".format(name), lg.DEBUG)
res = minimize(
rmse,
np.array(hours + sf),
args=(sub.values),
method="L-BFGS-B",
bounds=hours_bounds + sf_bounds,
options=dict(disp=True),
)
log(
"Completed discretization in {:,.2f} seconds".format(
time.time() - start_time),
lg.DEBUG,
)
edges[name] = res.x[0:n_bins + 1]
ampls[name] = res.x[n_bins + 1:]
results[name] = Series(piecewise(res.x))
self.bin_edges_ = Series(edges).apply(Series)
self.bin_scaling_factors_ = DataFrame(ampls)
result = concat(results)
else:
hour_of_min = self.time_at_min
sf = [1 / (i * 1.01) for i in range(1, n_bins + 1)]
sf.extend([self.min()])
sf_bounds = [(0, self.max()) for i in range(0, n_bins + 1)]
hours = [
hour_of_min - hour_of_min * 1 / (i * 1.01)
for i in range(1, n_bins + 1)
]
hours.extend([len(self)])
hours_bounds = [(0, len(self)) for i in range(0, n_bins + 1)]
start_time = time.time()
# log('discretizing EnergySeries {}'.format(name), lg.DEBUG)
res = minimize(
rmse,
np.array(hours + sf),
args=(self.values),
method="L-BFGS-B",
bounds=hours_bounds + sf_bounds,
options=dict(disp=True),
)
log(
"Completed discretization in {:,.2f} seconds".format(
time.time() - start_time),
lg.DEBUG,
)
edges = res.x[0:n_bins + 1]
ampls = res.x[n_bins + 1:]
result = Series(piecewise(res.x))
bin_edges = Series(edges).apply(Series)
self.bin_edges_ = bin_edges
bin_edges.loc[-1, 0] = 0
bin_edges.sort_index(inplace=True)
bin_edges = bin_edges.diff().dropna()
bin_edges = bin_edges.round()
self.bin_scaling_factors_ = DataFrame({
"duration": bin_edges[0],
"scaling factor": ampls
})
self.bin_scaling_factors_.index = np.round(edges).astype(int)
if inplace:
self.update(result)
self.__class__ = EnergySeries
self.__finalize__(result)
else:
result.__class__ = EnergySeries
return result.__finalize__(self)
