Working with Prophet#

Custom dashboard displaying metadata logged with Prophet

Prophet is a popular time-series forecasting library.

With the Neptune–Prophet integration, you can keep track of parameters, forecast data frames, residual diagnostic charts, cross-validation folds, and other metadata while training models with Prophet.

See in Neptune Code examples

API reference ≫ Prophet
neptune-prophet repository on GitHub
Prophet documentation
Prophet on GitHub

Before you start#

Set up Neptune. Instructions:
(optional) To log the plots as interactive HTML visualizations, you need to have the Plotly graphing library installed.
- You can enable logging as interactive HTML by setting the log_interactive parameter to True where applicable.
- Otherwise, the plots are recorded as static images by default.

Installing the Neptune-Prophet integration#

pip install -U neptune-prophet

Logging options#

In order to track the metadata to Neptune, you need to initialize a Neptune run.

import neptune

run = neptune.init_run()

Haven't registered?

If you haven't registered or set up your credentials, you can log anonymously to the public project:

run = neptune.init_run(
    api_token=neptune.ANONYMOUS_API_TOKEN,
    project="common/fbprophet-integration",
)

You can then use the run object to track metadata with the functions listed below.

Logging Prophet summary#

create_summary()

You can log all relevant metadata at once with this function, such as forecast plots and residual diagnostics.

Parameters

Parameter	Description
`model`	Fitted Prophet model object.
`df` (optional)	The dataset that was used for making the forecast. If provided, additional plots will be recorded.
`fcst` (optional)	Forecast returned by Prophet. If not provided, it'll be calculated using the `df` data.
`log_charts`	Additionally save the diagnostic plots. Defaults to `True`.
`log_interactive`	Save the plots as interactive HTML files. Requires the Plotly library. Defaults to `False`.

Returns

Dictionary with all the plots.

Example

import pandas as pd
from prophet import Prophet
import neptune
import neptune.integrations.prophet as npt_utils

run = neptune.init_run()

dataset = pd.read_csv(
    "https://raw.githubusercontent.com/facebook/prophet/main/examples/example_wp_log_peyton_manning.csv"
)
model = Prophet()
model.fit(dataset)

run["prophet_summary"] = npt_utils.create_summary(
    model=model, df=dataset, log_interactive=True
)

Getting the model configuration#

To extract the configuration from the Prophet model object, use the get_model_config() function:

run = neptune.init_run()
dataset = ...
model = Prophet()
model.fit(dataset)

run["model_config"] = npt_utils.get_model_config(model)

Getting the serialized model#

To serialize the Prophet model, you can use the get_serialized_model() function:

run = neptune.init_run()
dataset = ...
model = Prophet()
model.fit(dataset)

run["model"] = npt_utils.get_serialized_model(model)

Getting forecast components#

To get the forecast components (such as trend and monthly seasonality) produced by Prophet, use the get_forecast_components() function:

run = neptune.init_run()
dataset = ...
model = Prophet()
model.fit(dataset)
predicted = model.predict(dataset)

run["forecast_components"] = npt_utils.get_forecast_components(model, predicted)

Logging forecast plots#

To save Prophet plots (such as forecast components, changepoints, forecast with prediction interval), use the create_forecast_plots() function:

model.fit(dataset)
predicted = model.predict(dataset)

run["forecast_plots"] = npt_utils.create_forecast_plots(model, predicted)

Logging residual forecast plots#

To save additional diagnostic plots (such as histogram of residuals, ACF plot, QQ-plot for residuals) to Neptune, use the create_residual_diagnostics_plots() function:

model.fit(dataset)
predicted = model.predict(dataset)

run["residual_diagnostics_plot"] = npt_utils.create_residual_diagnostics_plots(
    predicted, dataset.y
)