Skip to main content

Multivariate forecasting using Facebook Prophet

Project description

Multi Prophet

Build Status PyPI version License: MIT

Multi Prophet is a procedure for forecasting time series data for multipe dependent variables based on Facebook Prophet package. If you have no prior experience with Facebook Prophet, check out their docs.

Multi Prophet does not train a single model with many outputs, it just wraps Facebook Prophet interface to support configuration and controll over multiple models. Multi Prophet has a very similar interface as Facebook Prophet.

The main difference is that return values of each method is a dictionary where each dependent value is a key, and the value is the return value of the linked Facebook Prophet model.

If Prophet return value is a data frame, then MultiProphet return value will be:

{"dependent_variable1": df1, "dependent_variable2": df2}

Installation

Multi Prophet is on PyPi. pip install multi-prophet

Getting started

Creating a basic model is almost the same as creating a Prophet model:

Prophet

# dataframe needs to have columns ds and y
from fbprophet import Prophet

m = Prophet()
m.fit(df)

future = m.create_future_dataframe(df)
forecast = m.predict(future)
m.plot(forecast)

Multi Prophet

# dataframe needs to have column ds, and it has y1 and y2 as dependent variables
from multi_prophet import MultiProphet

m = MultiProphet(columns=["y1", "y2"])
m.fit(df)

future = m.create_future_dataframe(df)
forecast = m.predict(future)
m.plot(forecast)

Adding country holidays

Prophet

m.add_country_holidays(country_name="US")

Multi Prophet

# For all dependent variables
m.add_country_holidays("US")

# For selected dependent variables
m.add_country_holidays("US", columns=["y1"])

Adding seasonality

Prophet

m.add_seasonality(name="monthly", period=30.5, fourier_order=5)

Multi Prophet

# For all dependent variables
m.add_seasonality(name="monthly", period=30.5, fourier_order=5)

# For selected dependent variables
m.add_seasonality(name="monthly", period=30.5, fourier_order=5, columns=["y1"])

Adding regressors

Prophet

m.add_regressor("Matchday")

Multi Prophet

# For all dependent variables
m.add_regressor("Matchday")

# For selected dependent variables
m.add_regressor("Matchday", columns=["y"])

Ploting results

Prophet

# Prophet
m.plot(forecast)
m.plot_components(forecast)

# With Plotly
from fbprophet.plot import plot_plotly, plot_components_plotly
import plotly.offline as py
py.init_notebook_mode()

fig = plot_plotly(m, forecast)
py.iplot(fig)

fig = plot_components_plotly(m, forecast)
py.iplot(fig)

Multi Prophet

m.plot(forecast)
m.plot_components(forecast)

# With Plotly
figures = m.plot(forecast, plotly=True)
for fig in figures.values():
    fig.show()

# or access by key
figures["y1"].show()

figures = m.plot_components(forecast, plotly=True)
for fig in figures.values():
    fig.show()

# or access by key
figures["y1"].show()

Facebook Prophet model configuration

Facebook Prophet supports a lot of configuration through kwargs. There are two ways to do it with Multi Prophet:

  1. Through kwargs just as with Facebook Prophet
    • Prophet
m = Prophet(growth="logistic")
m.fit(self.df, algorithm="Newton")
m.make_future_dataframe(7, freq="H")
m.add_regressor("Matchday", prior_scale=10)
* Multi Prophet
m = MultiProphet(columns=["y1", "y2"], growth="logistic")
m.fit(self.df, algorithm="Newton")
m.make_future_dataframe(7, freq="H")
m.add_regressor("Matchday", prior_scale=10)
  1. Configuration through constructor
# Same configuration for each dependent variable
m = MultiProphet(columns=["y1", "y2"],
                 growth="logistic",
                 weekly_seasonality=True,
                 n_changepoints=50)

# Different configuration for each model
config = {
    "y1": {"growth": "linear", "daily_seasonality": True},
    "y2": {"growth": "logistic", "weekly_seasonality": True}
}
m = MultiProphet(columns=["y1", "y2"], config=config)

# Adding regressors (dataframe has columns c1 and c2)
regressors = {
    "y1": [
        {"name": "c1", "prior_scale": 0.5},
        { "name": "c2", "prior_scale": 0.3}
    ],
    "y2": [{"name": "c2", "prior_scale": 0.3}]
}
m = MultiProphet(columns=["y1", "y2"], regressors=regressors)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

multi-prophet-1.1.1.tar.gz (5.4 kB view hashes)

Uploaded source

Built Distribution

multi_prophet-1.1.1-py3-none-any.whl (5.5 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page