def _curve_fitting(self, phase, start_date, end_date):
"""
Perform curve fitting for the phase.
Args:
phase (str): phase name
start_date (str): start date of the phase
end_date (str): end date of the phase
Returns:
tuple
(pandas.DataFrame): Result of curve fitting
Index: reset index
Columns:
- (phase name)_predicted: predicted value of Susceptible
- (phase_name)_actual: actual value of Susceptible
- (phase_name)_Recovered: Recovered
(int): minimum value of R, which is the change point of the curve
"""
sr_df = self.sr_df.copy()
sta = self.date_obj(start_date)
end = self.date_obj(end_date)
sr_df = sr_df.loc[(sr_df.index >= sta) & (sr_df.index <= end), :]
trend = Trend(sr_df)
df = trend.run()
if trend.rmsle() > self.max_rmsle:
df[f"{self.S}{self.P}"] = None
# Get min value for vline
r_value = int(df[self.R].min())
# Rename the columns
phase = self.INITIAL if phase == "0th" else phase
df = df.rename({f"{self.S}{self.P}": f"{phase}{self.P}"}, axis=1)
df = df.rename({f"{self.S}{self.A}": f"{phase}{self.A}"}, axis=1)
df = df.rename({f"{self.R}": f"{phase}_{self.R}"}, axis=1)
return (df, r_value)
def show(self, area, change_dates=None, **kwargs):
"""
show the S-R trend in a figure.
Args:
area (str): area name
change_dates (list[str] or None): list of change points
kwargs: keyword arguments of covsirphy.trend_plot()
Note:
@change_dates must be specified if ChangeFinder.run() was not done.
"""
# Curve fitting
start_dates, end_dates = self.date_range(change_dates)
nested = [
self._curve_fitting(self.num2str(num), start_date, end_date)
for (num, (start_date, end_date))
in enumerate(zip(start_dates, end_dates))
]
df_list, vlines = zip(*nested)
comp_df = pd.concat([self.sr_df, *df_list], axis=1)
comp_df = comp_df.rename({self.S: f"{self.S}{self.A}"}, axis=1)
comp_df = comp_df.apply(
lambda x: pd.to_numeric(x, errors="coerce", downcast="integer"), axis=0)
# Show figure
pred_cols = [col for col in comp_df.columns if col.endswith(self.P)]
if len(pred_cols) == 1:
title = f"{area}: S-R trend without change points"
else:
_list = self._change_dates[:]
strings = [", ".join(_list[i: i + 6]) for i in range(0, len(_list), 6)]
change_str = ",\n".join(strings)
title = f"{area}: S-R trend changed on\n{change_str}"
Trend.show_with_many(
result_df=comp_df, predicted_cols=pred_cols, title=title, v=vlines[1:], **kwargs)
def show(self, show_figure=True, filename=None):
"""
show the result as a figure and return a dictionary of phases.
Args:
@show_figure (bool): if True, show the result as a figure.
@filename (str): filename of the figure, or None (display figure)
Returns:
(covsirphy.PhaseSeries)
"""
# Create phase dictionary
phase_series = self._create_phases()
phase_dict = phase_series.to_dict()
# Curve fitting
nested = [
self._curve_fitting(phase, info)
for (phase, info) in phase_dict.items()
]
df_list, vlines = zip(*nested)
comp_df = pd.concat(df_list[1:], axis=1)
comp_df[self.R] = comp_df.fillna(0).loc[
:, comp_df.columns.str.endswith(self.R)
].sum(axis=1)
comp_df[f"{self.S}{self.A}"] = comp_df.fillna(0).loc[
:, comp_df.columns.str.endswith(self.A)
].sum(axis=1)
comp_df = comp_df.apply(
lambda x: pd.to_numeric(x, errors="coerce", downcast="integer"),
axis=0
)
# Show figure
if not show_figure:
return phase_series
pred_cols = comp_df.loc[
:, comp_df.columns.str.endswith(self.P)
].columns.tolist()
if len(pred_cols) == 1:
title = f"{self.area}: S-R trend without change points"
else:
_list = self.change_dates[:]
strings = [
", ".join(_list[i: i + 6]) for i in range(0, len(_list), 6)
]
change_str = ",\n".join(strings)
title = f"{self.area}: S-R trend changed on\n{change_str}"
Trend.show_with_many(
result_df=comp_df,
predicted_cols=pred_cols,
title=title,
vlines=vlines[2:],
filename=filename
)
return phase_series
def error_f(self, start_dates, end_dates):
"""
Definition of error score to minimize in the study.
This is weighted average of RMSLE scores.
@start_dates <list[str]>: list of start date of phases (candidates)
@end_dates <list[str]>: list of end date of phases (candidates)
@return <float> : score of the error function to minimize
"""
scores = list()
for (start_date, end_date) in zip(start_dates, end_dates):
population = self.pop_dict[start_date]
trend = Trend(self.clean_df,
population,
self.country,
province=self.province,
start_date=start_date,
end_date=end_date)
trend.analyse()
scores.append(trend.rmsle())
return np.average(scores, weights=range(1, len(scores) + 1))
def _curve_fitting(self, phase, info):
"""
Perform curve fitting for the phase.
Args:
phase (str): phase name
info (dict[str]): start date, end date and population
Returns:
(tuple)
(pandas.DataFrame): Result of curve fitting
Index: reset index
Columns:
- (phase name)_predicted: predicted value of Susceptible
- (phase_name)_actual: actual value of Susceptible
- (phase_name)_Recovered: Recovered
(int): minimum value of R, which is the change point of the curve
"""
start_date = info[self.START]
end_date = info[self.END]
population = info[self.N]
trend = Trend(
self.jhu_data, population, self.country, province=self.province,
start_date=start_date, end_date=end_date
)
trend.analyse()
df = trend.result()
if trend.rmsle() > self.max_rmsle:
df[f"{self.S}{self.P}"] = None
# Get min value for vline
r_value = int(df[self.R].min())
# Rename the columns
phase = self.INITIAL if phase == "0th" else phase
df = df.rename({f"{self.S}{self.P}": f"{phase}{self.P}"}, axis=1)
df = df.rename({f"{self.S}{self.A}": f"{phase}{self.A}"}, axis=1)
df = df.rename({f"{self.R}": f"{phase}_{self.R}"}, axis=1)
return (df, r_value)
def show(self, show_figure=True, filename=None):
"""
show the result as a figure and return a dictionary of phases.
Args:
@show_figure <bool>:
- if True, show the result as a figure.
@filename <str>: filename of the figure, or None (show figure)
Returns:
<covsirphy.PhaseSeries>
"""
# Create phase dictionary
phase_series = self._create_phases()
phase_dict = phase_series.to_dict()
# Curve fitting
df_list = list()
vlines = list()
for (phase, info) in phase_dict.items():
start_date = info[self.START]
end_date = info[self.END]
population = info[self.N]
trend = Trend(self.clean_df,
population,
self.country,
province=self.province,
start_date=start_date,
end_date=end_date)
trend.analyse()
df = trend.result()
# Get min value for vline
vlines.append(df[self.R].min())
# Rename the columns
phase = phase.replace("0th", self.INITIAL)
df = df.rename({f"{self.S}{self.P}": f"{phase}{self.P}"}, axis=1)
df = df.rename({f"{self.S}{self.A}": f"{phase}{self.A}"}, axis=1)
df = df.rename({f"{self.R}": f"{phase}_{self.R}"}, axis=1)
df_list.append(df)
if self.n_points == 0:
comp_df = pd.concat(df_list, axis=1)
else:
comp_df = pd.concat(df_list[1:], axis=1)
comp_df[self.R] = comp_df.fillna(
0).loc[:, comp_df.columns.str.endswith(self.R)].sum(axis=1)
comp_df[f"{self.S}{self.A}"] = comp_df.fillna(
0).loc[:, comp_df.columns.str.endswith(self.A)].sum(axis=1)
comp_df = comp_df.apply(
lambda x: pd.to_numeric(x, errors="coerce", downcast="integer"),
axis=0)
# Show figure
if not show_figure:
return phase_series
pred_cols = comp_df.loc[:, comp_df.columns.str.
endswith(self.P)].columns.tolist()
if len(pred_cols) == 1:
title = f"{self.area}: S-R trend without change points"
else:
change_str = ", ".join(self.change_dates)
title = f"{self.area}: S-R trend changed on {change_str}"
Trend.show_with_many(result_df=comp_df,
predicted_cols=pred_cols,
title=title,
vlines=vlines[2:],
filename=filename)
return phase_series