def plot_seasonality(
self,
order="default",
center=True,
model="additive",
median=False,
title="Seasonality",
x_lab="Index",
y_lab="Y",
time_index=True,
x_rotation=45,
**kwargs
):
"""
Plots series seasonality.
"""
y = self.values["X"]
x = self.values["index"].astype("O") if time_index else np.arange(len(y))
order = self.frequency if order is "default" else order
avg_seasonality = seasonality(y, order, center, model, median)
plt.plot(x, avg_seasonality, **kwargs)
plt.title(title)
plt.ylabel(y_lab)
plt.xlabel(x_lab)
plt.xticks(rotation=x_rotation)
plt.tight_layout()
def impute_random(self, trailing=True, mode=None, **kwargs):
"""
Returns tseries object with missing values of the series filled
with a random number from the series up to that point.
"""
data = self.values["X"]
order = kwargs.pop("order", self.frequency)
if mode is "decompose":
seasonal = seasonality(data, order, **kwargs).reshape(len(data))
de_seasonal = data - seasonal
self.values["X"] = seasonal + nan_random(de_seasonal, trailing)
elif mode is "split":
k = 0
r = []
while k < order:
r.append(nan_random(data[k::order], trailing))
k += 1
rz = np.array(list(zip_longest(*r, fillvalue=np.nan))).flatten()
self.values["X"] = np.array([i for i in rz if not np.isnan(i).any()])
else:
self.values["X"] = nan_random(data, trailing)
return self
def decompose(self, order="default", center=True, model="additive", median=False):
"""
Returns array of decomposition results in the order of:
original series, trend, seasonality, random.
"""
data = self.values["X"]
order = self.frequency if order is "default" else order
trends = trend(data, order, center)
avg_seasonality = seasonality(data, order, center, model, median)
random = remainder(data, order, center, model, median)
decomp_series = np.column_stack((data, trends, avg_seasonality, random))
dtypes = np.dtype(
[
("series", data.dtype),
("trend", trends.dtype),
("seasonality", avg_seasonality.dtype),
("random", random.dtype),
]
)
decomposition = np.empty(len(decomp_series), dtype=dtypes)
decomposition["series"] = decomp_series[:, 0]
decomposition["trend"] = decomp_series[:, 1]
decomposition["seasonality"] = decomp_series[:, 2]
decomposition["random"] = decomp_series[:, 3]
return decomposition
def impute_interpolation(self, mode=None, **kwargs):
"""
Returns tseries object with missing values of the series filled
via linear interpolation.
"""
data = self.values["X"]
order = kwargs.pop("order", self.frequency)
if mode is "decompose":
seasonal = seasonality(data, order, **kwargs).reshape(len(data))
de_seasonal = data - seasonal
self.values["X"] = seasonal + nan_linear_interpolation(de_seasonal)
elif mode is "split":
k = 0
r = []
while k < order:
r.append(nan_linear_interpolation(data[k::order]))
k += 1
rz = np.array(list(zip_longest(*r, fillvalue=np.nan))).flatten()
self.values["X"] = np.array([i for i in rz if not np.isnan(i).any()])
else:
self.values["X"] = nan_linear_interpolation(data)
return self
def decompose_seasonality(
self, order="default", center=True, model="additive", median=False
):
"""
Returns array of series seasonality.
"""
data = self.values["X"]
order = self.frequency if order is "default" else order
avg_seasonality = seasonality(data, order, center, model, median)
return avg_seasonality