Forecasting with Prophet

Prophet and Key Data Assumptions #

Prophet is a python library released by Facebook’s data science team for time series forecasting that is based on an additive model where non-linear trends are fit with daily, weekly, and yearly seasonality. It is a very nifty tool but it does come with several assumptions that are worth outlining first:

1. Your time series is stationary - meaning the mean and variance do not change over time.
2. Your time series has a linear or logistic trend and that the seasonality is addictive or multiplicative.
3. You have a reasonable amount of historical data. Typically one year of daily data or two years of weekly data.
4. Your time series is not affected by external factors such as economic factors or customer behavior.
5. Your time series is not affected by outliers or missing data. If you have either, they must be handled before your Prophet.

Example #

Below is an example demonstrating the basic process of using Prophet with python to forecast a time series dataset. In this example, we’ll use prophet alongside pandas with the sample data provided by prophet.

Import the necessary libraries. #

import pandas as pd
from prophet import prophet

Import example data. #

data = pd.read_csv('https://raw.githubusercontent.com/facebook/prophet/main/examples/example_wp_log_peyton_manning.csv')

Create a Prophet model. Add additional seasonal parameters, if necessary like weekly or yearly seasonality. #

model = Prophet(seasonality_mode='additive', 
                weekly_seasonality=True, 
                yearly_seasonality=True,
                )

Fit the model to you data. #

model.fit(data)

Make predictions for future dates. In this case, we are forecasting 365 days into the future. #

future = model.make_future_dataframe(periods=365)

Generate the forecast. #

forecast = model.predict(future)

Visualize the generated forecast. #

model.plot(forecast)

Visualize a more detailed forecast components. #

model.plot_components(forecast)