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

[1]:

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

[2]:

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

[3]:

from hcrystalball.wrappers import ProphetWrapper

[4]:

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

[5]:

extra_holidays = {
    'Christmas Day':{'lower_window': -2, 'upper_window':2, 'prior_scale': 20},
#     'Good Friday':{'lower_window': -1, 'upper_window':1, 'prior_scale': 30}
}

Unusual Seasonalities

[6]:

extra_seasonalities = [
    {
        'name':'bi-weekly',
        'period': 14.,
        'fourier_order': 5,
        'prior_scale': 10.0,
        'mode': None
    },
    {
        'name':'bi-yearly',
        'period': 365*2.,
        'fourier_order': 5,
        'prior_scale': 5.0,
        'mode': None
    },
]

Exogenous Variables

[7]:

from sklearn.pipeline import Pipeline
from hcrystalball.feature_extraction import HolidayTransformer

[8]:

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

[9]:

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

[10]:

pipeline = Pipeline([
    ('holidays_de', HolidayTransformer(country_code = 'DE')),
    ('model', prophet)
])
pipeline_extra = Pipeline([
    ('holidays_de', HolidayTransformer(country_code = 'DE')),
    ('model', prophet_extra)
])

[11]:

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

prds.plot(title=f"MAE:{(prds['target']-prds['prophet']).abs().mean()}");

[11]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f70ff64eb50>
../../../_images/examples_tutorial_wrappers_06_advanced_prophet_14_1.png 
[12]:

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

prds_extra.plot(title=f"MAE:{(prds_extra['target']-prds_extra['prophet_extra']).abs().mean()}");

[12]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f70ff0e3fd0>
../../../_images/examples_tutorial_wrappers_06_advanced_prophet_15_1.png

Compared to non-tweaked model, we are able to better catch the series dynamics, but don’t win against roughly average predictions

[13]:

prds = (ProphetWrapper().fit(X[:-10], y[:-10])
                 .predict(X[-10:])
                 .merge(y, left_index=True, right_index=True, how='outer')
                 .tail(50)
)
prds.plot(title=f"MAE:{(prds['target']-prds['prophet']).abs().mean()}");

[13]:

<matplotlib.axes._subplots.AxesSubplot at 0x7f70ff10f350>
../../../_images/examples_tutorial_wrappers_06_advanced_prophet_17_1.png

Full Prophet Output

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

[14]:

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

[14]:
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 6.165281 1.143956 11.732651 6.165281 6.165281 -0.011450 -0.011450 -0.011450 -0.011450 -0.011450 -0.011450 0.0 0.0 0.0 6.153831
2018-12-23 6.162873 0.883810 11.388751 6.162873 6.162873 -0.082384 -0.082384 -0.082384 -0.082384 -0.082384 -0.082384 0.0 0.0 0.0 6.080489
2018-12-24 6.160466 1.085740 11.712883 6.160466 6.160466 0.051958 0.051958 0.051958 0.051958 0.051958 0.051958 0.0 0.0 0.0 6.212424
2018-12-25 6.158058 0.532724 11.514059 6.158058 6.158058 -0.033260 -0.033260 -0.033260 -0.033260 -0.033260 -0.033260 0.0 0.0 0.0 6.124798
2018-12-26 6.155651 0.677829 11.364090 6.155651 6.155651 0.063844 0.063844 0.063844 0.063844 0.063844 0.063844 0.0 0.0 0.0 6.219495
2018-12-27 6.153243 0.973931 11.163730 6.153243 6.153243 -0.022048 -0.022048 -0.022048 -0.022048 -0.022048 -0.022048 0.0 0.0 0.0 6.131196
2018-12-28 6.150836 0.952671 11.307701 6.150822 6.150859 0.033340 0.033340 0.033340 0.033340 0.033340 0.033340 0.0 0.0 0.0 6.184176
2018-12-29 6.148428 0.629671 11.264196 6.148379 6.148517 -0.011450 -0.011450 -0.011450 -0.011450 -0.011450 -0.011450 0.0 0.0 0.0 6.136979
2018-12-30 6.146021 0.827071 11.679368 6.145887 6.146201 -0.082384 -0.082384 -0.082384 -0.082384 -0.082384 -0.082384 0.0 0.0 0.0 6.063636
2018-12-31 6.143613 0.957509 11.614246 6.143412 6.143908 0.051958 0.051958 0.051958 0.051958 0.051958 0.051958 0.0 0.0 0.0 6.195571