mikekeith52

df2 = df.copy()
df2['year'] = df2.reset_index().DATE.dt.year
df2['month'] = df2.reset_index().DATE.dt.month
df2['pct_chg'] = df2['HOUSTNSA'].pct_change()

avg_chg_month = df2.groupby('month')['pct_chg'].mean().reset_index()
avg_chg_month['pct_chg_std'] = df2.groupby('month')['pct_chg'].std().values
avg_chg_month['pct_chg_2016'] = df2.loc[df2['year'] == 2016,'pct_chg'].values
avg_chg_month['pct_chg_2020'] = df2.loc[df2['year'] == 2020,'pct_chg'].values
avg_chg_month['pct_chg_2021'] = df2.loc[df2['year'] == 2021,'pct_chg'].values

mikekeith52

def prepare_fcst(f, test_length=0.1, fcst_length=120):
    """ adds all variables and sets the test length/forecast length in the object
    
    Args:
        f (Forecaster): the Forecaster object.
        test_length (int or float): the test length as a size or proportion.
        fcst_length (int): the forecast horizon.
        
    Returns:
        (Forecaster) the processed object.

mikekeith52

pd.options.display.max_rows = None
all_xvars = f.get_regressor_names()
final_dropped = pd.DataFrame({"Var": all_xvars})
for i, v in f.export("model_summaries").iterrows():
    model = v["ModelNickname"]
    Xvars = v["Xvars"]
    dropped_vars = [x for x in f.get_regressor_names() if x not in Xvars]
    if not dropped_vars:
        continue
    tmp_dropped = pd.DataFrame(

mikekeith52

selected_model = "gbt"
hp = results[selected_model][3]
f.set_estimator(selected_model)
f.manual_forecast(**hp, Xvars="all", call_me=selected_model + "_all_vars")
f.manual_forecast(
    **hp, Xvars=lasso_reduced_vars, call_me=selected_model + "_l1_reduced_vars"
)
f.manual_forecast(
    **hp, Xvars=mlr_reduced_vars, call_me=selected_model + "pfi-mlr_reduced_vars"
)

mikekeith52

critical_pval = 0.05
print('-'*100)
print('Augmented Dickey-Fuller results:')
stat, pval, _, _, _, _ = f.adf_test(full_res=True)
print('the test-stat value is: {:.2f}'.format(stat))
print('the p-value is {:.4f}'.format(pval))
print('the series is {}'.format('stationary' if pval < critical_pval else 'not stationary'))
print('-'*100)

mikekeith52

f.add_sklearn_estimator(StackingRegressor,'stacking')
f.set_estimator('stacking')
results = f.export('model_summaries')
estimators = [
    ('knn',
     KNeighborsRegressor(**results.loc[results['ModelNickname'] == 'knn','HyperParams'].values[0])),
    ('xgboost',
     XGBRegressor(**results.loc[results['ModelNickname'] == 'xgboost','HyperParams'].values[0])),
    ('lightgbm',
     LGBMRegressor(**results.loc[results['ModelNickname'] == 'lightgbm','HyperParams'].values[0])),

mikekeith52

from sklearn.linear_model import LinearRegression
from sklearn.linear_model import ElasticNet
from sklearn.neural_network import MLPRegressor
from sklearn.neighbors import KNeighborsRegressor
from xgboost import XGBRegressor

results = mvf.export_model_summaries()

estimators = [
    ("mlr", LinearRegression()),

mikekeith52

f.set_validation_length(12)
grid = {
    'order':[(1,1,1),(1,1,0),(0,1,1)],
    'seasonal_order':[(2,1,1,12),(1,1,1,12),(2,1,0,12),(0,1,0,12)]
}

f.ingest_grid(grid)
f.tune()
f.auto_forecast(call_me='arima4')

mikekeith52

from scalecast.auxmodels import auto_arima

auto_arima(
    f,
    start_P=1,
    start_q=1,
    max_p=6,
    max_q=6,
    m=12,
    seasonal=True,

mikekeith52

# EDA
f.plot_acf()
plt.show()
f.plot_pacf()
plt.show()
f.seasonal_decompose().plot()
plt.show()
stat, pval, _, _, _, _ = f.adf_test(full_res=True)
print(stat)
print(pval)