ProphetModelTrainer¶

This is a Python class that inherits from BaseModelTrainer and is used to train and predict values using the Prophet model. The class provides methods for fitting the model, making predictions, applying prediction bounds, processing data, and generating plots.

Configuration¶

Required Configuration¶

ProphetModelTrainer contains the following required configurations:

time_index: column representing the time index of the model.
model_target: column representing the model target
forecast_frequency: The frequency of the forecast. I.E. how often predictions are made: D for days, H for hour, etc. See the prophet documentation for details.
forecast_period: Length of time of the forecast. This is an integer value of forecast_frequency units.
local_dir: local directory to use to write forecasts.

Optional Configuration¶

ProphetModelTrainer contains the following required configurations:

country_holidays: Country holiday code to use, if any.
regressor_cols: List of columns that should be included as additional regressors, if any.
target_min: minimum value allowed for predictions, if any (i.e. 0)
target_max: maximum value allowed for predictions, if any.

Default Configuration¶

ProphetModelTrainer contains no default configuration.

Methods¶

fit¶

Fits the Prophet model to the training data.

def fit(self, data, *args, **kwargs)

Arguments:

data (dict): Dictionary containing the training data with columns 'X_train' and 'y_train'.

Returns:

model: A trained Prophet model object.

predict¶

Predicts values for train, validation, and test data.

def predict(self, data, *args, **kwargs)

Arguments:

data (dict): A dictionary containing training, validation, and test data with columns 'X_train', 'y_train', 'X_valid', 'y_valid', 'X_test', 'y_test'.

Returns:

dict: A dictionary containing predictions for train, validation, and test sets.

_apply_prediction_bounds¶

Applies min/max target values to the predicted values.

def _apply_prediction_bounds(self, data, target_min=None, target_max=None)

Arguments:

data (pd.DataFrame): DataFrame containing the predicted values.
target_min (float, optional): Minimum permissible value for the predicted values.
target_max (float, optional): Maximum permissible value for the predicted values.

Returns:

pd.DataFrame: DataFrame containing the predicted values within the range of [target_min, target_max].

predict_df¶

Predicts values for the given DataFrame.

def predict_df(self, df, *args, **kwargs) -> pd.DataFrame

Arguments:

df (pd.DataFrame): DataFrame containing the data to be processed.

Returns:

pd.DataFrame: The predicted values.

_process_data¶

Processes data as per the requirements of the Prophet model.

def _process_data(self, data) -> pd.DataFrame

Arguments:

data (pd.DataFrame): DataFrame containing the dataset to be processed.

Returns:

pd.DataFrame: The processed dataset.

get_artifacts¶

Generates plots for visualizing the forecasted data.

get_artifacts(self, id, *args, **kwargs)

Arguments:

id (str): ID for the model being used for generating plots.

Returns:

dict: A dictionary containing the paths of the saved plots.

Usage Example¶

from lolpop.component import ProphetModelTrainer

... #create data 

# Create an instance of the ProphetModelTrainer class
trainer = ProphetModelTrainer(problem_type='timeseries', params={'seasonality_mode': 'multiplicative'})

# Fit the model to the training data
model = trainer.fit(data={'X_train': train_X, 'y_train': train_y})

# Make predictions for the test data
predictions = trainer.predict(data={'X_test': test_X, 'y_test': test_y})