def test_scipy_optimizer(seed):
np.random.seed(seed)
num_obs = 20
df = pd.DataFrame({
"obs": np.random.randn(num_obs),
"cov0": np.random.randn(num_obs),
"cov1": np.random.randn(num_obs)
})
data = Data(col_obs="obs", col_covs=["cov0", "cov1"], df=df)
spline_specs = SplineSpecs(knots=np.linspace(0.0, 1.0, 5),
degree=3,
knots_type="rel_domain")
var_cov0 = Variable(name="cov0")
var_cov1 = SplineVariable(name="cov1", spline_specs=spline_specs)
model = GaussianModel(
data, param_specs={"mu": {
"variables": [var_cov0, var_cov1]
}})
coefs = scipy_optimize(model)
tr_coef = np.linalg.solve(
(model.mat[0].T * model.data.weights).dot(model.mat[0]),
(model.mat[0].T * model.data.weights).dot(model.data.obs))
assert np.allclose(coefs, tr_coef)
def get_mortality_pattern_model(df: DataFrame,
col_time: str = "time_start",
units_per_year: int = 12,
knots: np.ndarray = np.arange(2010, 2021),
smooth_order: int = 1) -> ExcessMortalityModel:
seas_spline_specs = SplineSpecs(knots=np.linspace(0.0, 1.0, 5),
degree=3,
knots_type="rel_domain")
time_knots = get_time_knots(df.time.min(), df.time.max(), knots)
time_spline_specs = SplineSpecs(knots=time_knots,
degree=1,
knots_type="abs")
seas_var = SplineVariable(col_time, spline_specs=seas_spline_specs)
time_var = SplineVariable("time", spline_specs=time_spline_specs)
variables = [
SeasonalityModelVariables([seas_var], col_time, smooth_order),
TimeModelVariables([time_var])
]
return ExcessMortalityModel(df, variables)
def get_mortality_pattern_model(df: DataFrame,
col_time: str = "time_start",
units_per_year: int = 12,
knots_per_year: float = 0.5,
tail_size: int = 18,
smooth_order: int = 1) -> ExcessMortalityModel:
"""
Define one mortality pattern model
"""
seas_spline_specs = SplineSpecs(knots=np.linspace(0.0, 1.0, 5),
degree=3,
knots_type="rel_domain")
time_knots = get_time_knots(df.time.min(), df.time.max(), units_per_year,
knots_per_year, tail_size)
time_spline_specs = SplineSpecs(knots=time_knots,
degree=1,
knots_type="abs")
seas_var = SplineVariable(col_time, spline_specs=seas_spline_specs)
time_var = SplineVariable("time", spline_specs=time_spline_specs)
variables = [
SeasonalityModelVariables([seas_var], col_time, smooth_order),
TimeModelVariables([time_var])
]
return ExcessMortalityModel(df, variables)
def var_cov1(spline_gprior, spline_uprior, spline_specs):
return SplineVariable(name="cov1",
spline_specs=spline_specs,
priors=[spline_gprior, spline_uprior])
def main():
# process input data
df = pd.read_csv(data_path)
df_2020 = df[[
"location_name", "year_x", "week", "pop_2020", "death_rate_2020"
]].copy()
df_2019 = df[[
"location_name", "year_y", "week", "pop_2019", "death_rate_2019"
]].copy()
df_2020 = df_2020.rename(
columns={
"year_x": "year",
"pop_2020": "population",
"death_rate_2020": "death_rate"
})
df_2019 = df_2019.rename(
columns={
"year_y": "year",
"pop_2019": "population",
"death_rate_2019": "death_rate"
})
df = pd.concat([df_2019, df_2020])
df["deaths"] = df["death_rate"] * df["population"] / 100000
time_ub = {k: v - time_start + 1 for k, v in time_end.items()}
locations = df.location_name.unique()
data = {k: {} for k in time_end.keys()}
for location in locations:
df_loc = df.loc[df.location_name == location]
df_loc = df_loc.sort_values(["year", "week"])
df_loc = add_time(df_loc, "year", "week", time_start)
for k in time_end.keys():
data[k][location] = df_loc.loc[df_loc.time < time_ub[k]].copy()
data[k][location]["offset_0"] = np.log(
data[k][location].population)
# create models
models = {}
for location, d in data["fit"].items():
seas_var = SplineVariable(time_unit, spline_specs=seas_spline_specs)
time_var = SplineVariable("time", spline_specs=time_spline_specs)
variables = [
SeasonalityModelVariables([seas_var], time_unit, smooth_order=1),
TimeModelVariables([time_var])
]
models[location] = ExcessMortalityModel(d, variables)
# run models
results = {}
for location, model in models.items():
model.run_models()
d_pred = model.predict(data["pred"][location],
col_pred="mortality_pattern")
results[location] = d_pred
# save results
if not results_folder.exists():
results_folder.mkdir()
df_result = pd.concat(results.values())
df_result.to_csv(results_folder / "prediction.csv", index=False)
# plot results
for location, result in results.items():
ax, axs = plot_data(result, "week", "year")
plt.delaxes(axs[1])
ax = plot_model(ax, result, "mortality_pattern", color="#008080")
ax.set_title(location, loc="left")
ax.legend()
plt.savefig(results_folder / f"{location}.pdf", bbox_inches="tight")
plt.close("all")
return models, results
def spline_variable():
return SplineVariable(name=COL_COVS[0],
spline_specs=SplineSpecs(knots=np.linspace(
0.0, 1.0, 5),
degree=3))
# result folder
results_path = Path("./examples/results_debug")
# define all variables
intercept_variable = Variable("intercept")
idr_spline_specs = SplineSpecs(knots=np.linspace(0.0, 1.0, 5),
degree=2,
knots_type="rel_domain",
include_first_basis=False)
idr_variable = SplineVariable("idr_lagged",
spline_specs=idr_spline_specs,
priors=[
SplineUniformPrior(order=1,
lb=-np.inf,
ub=0.0),
SplineGaussianPrior(order=1,
mean=0.0,
sd=1e-4,
domain_lb=0.4,
domain_ub=1.0)
])
time_spline_specs = SplineSpecs(knots=np.linspace(0.0, 1.0, 10),
degree=2,
knots_type="rel_domain",
include_first_basis=False,
r_linear=True)
time_variable = SplineVariable("time_id",
spline_specs=time_spline_specs,
priors=[
SplineGaussianPrior(order=1,