erykml

from shiny import App, render, ui

app_ui = ui.page_fluid(
    ui.h2("Hello Shiny!"),
    ui.input_slider("n", "N", 0, 100, 20),
    ui.output_text_verbatim("txt"),
)


def server(input, output, session):

erykml

score_list = []
for fit_col in ["model_1", "model_2", "model_3"]:
	scores = {
    	"model": fit_col,
    	"train_score": mean_absolute_error(
        	results_df.iloc[:TRAIN_END]["actuals"],
        	results_df.iloc[:TRAIN_END][fit_col]
    	),
    	"test_score": mean_absolute_error(
        	results_df.iloc[TRAIN_END:]["actuals"],

erykml

results_df.plot(title="Comparison of fits using different time-based features",
            	figsize=(16,4),
            	color = ["c", "k", "b", "r"])
plt.axvline(date(2020, 1, 1), c="m", linestyle="--");

erykml

model_3 = LinearRegression().fit(X_3.iloc[:TRAIN_END],
                             	y.iloc[:TRAIN_END])
results_df["model_3"] = model_3.predict(X_3)
results_df[["actuals", "model_3"]].plot(figsize=(16,4),
                                    	title="Fit using RBF features")
plt.axvline(date(2020, 1, 1), c="m", linestyle="--");

erykml

rbf = RepeatingBasisFunction(n_periods=12,
                         	column="day_of_year",
                         	input_range=(1,365),
                         	remainder="drop")
rbf.fit(X)
X_3 = pd.DataFrame(index=X.index,
               	data=rbf.transform(X))
X_3.plot(subplots=True, figsize=(14, 8),
     	sharex=True, title="Radial Basis Functions",
     	legend=False);

erykml

X_2_daily = X_2[["day_sin", "day_cos"]]
model_2 = LinearRegression().fit(X_2_daily.iloc[:TRAIN_END],
                             	y.iloc[:TRAIN_END])
results_df["model_2"] = model_2.predict(X_2_daily)
results_df[["actuals", "model_2"]].plot(figsize=(16,4),
                                    	title="Fit using sine/cosine features")
plt.axvline(date(2020, 1, 1), c="m", linestyle="--");

erykml

X_2 = X.copy()
X_2["month"] = X_2.index.month
X_2["month_sin"] = sin_transformer(12).fit_transform(X_2)["month"]
X_2["month_cos"] = cos_transformer(12).fit_transform(X_2)["month"]
X_2["day_sin"] = sin_transformer(365).fit_transform(X_2)["day_of_year"]
X_2["day_cos"] = cos_transformer(365).fit_transform(X_2)["day_of_year"]
fig, ax = plt.subplots(2, 1, sharex=True, figsize=(16,8))
X_2[["month_sin", "month_cos"]].plot(ax=ax[0])
X_2[["day_sin", "day_cos"]].plot(ax=ax[1])
plt.suptitle("Cyclical encoding with sine/cosine transformation");

erykml

def sin_transformer(period):
	return FunctionTransformer(lambda x: np.sin(x / period * 2 * np.pi))
def cos_transformer(period):
	return FunctionTransformer(lambda x: np.cos(x / period * 2 * np.pi))

erykml

model_1 = LinearRegression().fit(X_1.iloc[:TRAIN_END],
                             	y.iloc[:TRAIN_END])
results_df["model_1"] = model_1.predict(X_1)
results_df[["actuals", "model_1"]].plot(figsize=(16,4),
                                    	title="Fit using month dummies")
plt.axvline(date(2020, 1, 1), c="m", linestyle="--");

erykml

X_1 = pd.DataFrame(
	data=pd.get_dummies(X.index.month, drop_first=True, prefix="month")
)
X_1.index = X.index
X_1