ProphetLite

Like Prophet but in numba and all fit with LASSO

A basic comparison vs Prophet:

import pandas as pd
import matplotlib.pyplot as plt
from prophet import Prophet


df = pd.read_csv('https://raw.githubusercontent.com/facebook/prophet/main/examples/example_wp_log_peyton_manning.csv')
m = Prophet(weekly_seasonality=False)
m.fit(df, )
future = m.make_future_dataframe(periods=365)
forecast = m.predict(future)
m.plot(forecast)
m.plot_components(forecast)
plt.show()

Now for ProphetLite

You must pass your data as a numpy array

    y = df['y'].values

Now to build the class and pass the seasonality

from ProphetLite.prophetlite import ProphetLite 


pl = ProphetLite()
fitted = pl.fit(y, [365.25]) #To pass multiple seasonalities just add more nu
predicted = pl.predict(365)

Some Plotting helper methods

    pl.plot()

    pl.plot_components()

Comparison Plots

    plt.plot(np.append(fitted['yhat'], predicted['yhat']), alpha=.3)
    plt.plot(forecast['yhat'], alpha=.3)
    plt.show()

The Trend Components:

    plt.plot(np.append(fitted['trend'], predicted['trend']))
    plt.plot(forecast['trend'])
    plt.show()