Advanced Prophet Usage

You can scan through fbprophet docs and find many options how to tweak your model. Some of that functionality is moved to initialization stage to be compatible with Sklearn API. We will showcase the parts that were moved to initialization, but you can also look for other model parameters that could help fine-tuning your model

from hcrystalball.wrappers import ProphetWrapper

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
plt.style.use('seaborn')
plt.rcParams['figure.figsize'] = [12, 6]

from hcrystalball.utils import generate_tsdata
X, y = generate_tsdata(n_dates=365*2)

ProphetWrapper?

Advanced Holidays

For holidays, we are able to define instead of single boolean attribute distribution around given day. We define lower_window, upper_window and prior_scales

extra_holidays = {
    'Whit Monday':{'lower_window': 2, 'upper_window':2, 'prior_scale': 10},
    'Good Friday':{'lower_window': 1, 'upper_window':1, 'prior_scale': 30}
}

Unusual Seasonalities

extra_seasonalities = [
    {
        'name':'bi-weekly',
        'period': 14.,
        'fourier_order': 5,
        'prior_scale': 15.0,
        'mode': None
    },
    {
        'name':'bi-yearly',
        'period': 365*2.,
        'fourier_order': 5,
        'prior_scale': 5.0,
        'mode': None
    },
]

Exogenous Variables

extra_regressors = ['trend_line']
X['trend_line'] = np.arange(len(X))

prophet = ProphetWrapper(
    extra_holidays=extra_holidays,
    extra_seasonalities=extra_seasonalities,
    extra_regressors=extra_regressors,
    name='prophet_extra',
)

(prophet.fit(X[:-10], y[:-10])
        .predict(X[-10:])
        .merge(y, left_index=True, right_index=True, how='outer')
        .tail(50)
        .plot()
);

Compared to non-tweaked model, this shows our fine-tuning helped to increase performance significantly

(ProphetWrapper().fit(X[:-10], y[:-10])
                 .predict(X[-10:])
                 .merge(y, left_index=True, right_index=True, how='outer')
                 .tail(50)
                 .plot()
);

Full Prophet Output

If you need, you can also pass full_prophet_output and get rich predict output

(ProphetWrapper(full_prophet_output=True, conf_int=True)
     .fit(X[:-10], y[:-10])
     .predict(X[-10:])
)

	trend	prophet_lower	prophet_upper	trend_lower	trend_upper	additive_terms	additive_terms_lower	additive_terms_upper	weekly	weekly_lower	weekly_upper	multiplicative_terms	multiplicative_terms_lower	multiplicative_terms_upper	prophet
2018-12-22	6.143320	0.749968	10.803790	6.143320	6.143320	-0.035883	-0.035883	-0.035883	-0.035883	-0.035883	-0.035883	0.0	0.0	0.0	6.107438
2018-12-23	6.140754	0.862639	11.400232	6.140754	6.140754	-0.032846	-0.032846	-0.032846	-0.032846	-0.032846	-0.032846	0.0	0.0	0.0	6.107908
2018-12-24	6.138188	0.828253	11.574750	6.138188	6.138188	0.025082	0.025082	0.025082	0.025082	0.025082	0.025082	0.0	0.0	0.0	6.163270
2018-12-25	6.135621	0.540491	11.316455	6.135621	6.135621	-0.020201	-0.020201	-0.020201	-0.020201	-0.020201	-0.020201	0.0	0.0	0.0	6.115420
2018-12-26	6.133055	0.751823	11.142846	6.133055	6.133055	0.018654	0.018654	0.018654	0.018654	0.018654	0.018654	0.0	0.0	0.0	6.151709
2018-12-27	6.130489	0.923553	11.641867	6.130489	6.130489	0.030489	0.030489	0.030489	0.030489	0.030489	0.030489	0.0	0.0	0.0	6.160978
2018-12-28	6.127923	1.076292	11.500066	6.127916	6.127946	0.014705	0.014705	0.014705	0.014705	0.014705	0.014705	0.0	0.0	0.0	6.142628
2018-12-29	6.125356	0.927312	11.153359	6.125302	6.125440	-0.035883	-0.035883	-0.035883	-0.035883	-0.035883	-0.035883	0.0	0.0	0.0	6.089474
2018-12-30	6.122790	0.873291	11.383074	6.122696	6.122977	-0.032846	-0.032846	-0.032846	-0.032846	-0.032846	-0.032846	0.0	0.0	0.0	6.089944
2018-12-31	6.120224	0.832878	11.464349	6.120045	6.120502	0.025082	0.025082	0.025082	0.025082	0.025082	0.025082	0.0	0.0	0.0	6.145306