modrover.learner#

class modrover.learner.Learner(model_class, obs, main_param, param_specs, weights='weights', get_score=<function get_rmse>)[source]#

Bases: object

Individual learner class for one specific covariate configuration.

Parameters:

  • model_class (type) – Regmod model constructor

  • obs (str) – Name corresponding to the observation column in the data frame

  • main_param (str) – The main parameter we are exploring. This is aligned with the modrover.rover.Rover class.

  • param_specs (dict[str, dict]) – Parameter settings for the regmod model

  • weights (str) – Name corresponding to the weights column in the data frame

  • get_score (Callable) – Function that evaluate the score of of the model

property coef: NDArray | None#

Coefficients in the regmod model.

property vcov: NDArray | None#

Variance-covarianace matrix for the coefficients in the regmod model.

fit(data, holdouts=None, **optimizer_options)[source]#

Fit a set of models on a series of holdout datasets.

This method will fit a model over k folds of the dataset, where k is the length of the provided holdouts list. It is up to the user to decide the train-test splits for each holdout column.

On each fold of the dataset, the trained model will predict out on the validation set and obtain a evaluate. The averaged evaluate across all folds becomes the model’s overall score.

Finally, a model is trained with all data in order to generate the final coefficients.

Parameters:

  • data (DataFrame) – A dataframe containing the training data

  • holdouts (list[str] | None) – Which column names to iterate over for cross validation. If it is None, insample performance score will be used to evaluate the model.

  • **optimizer_options – Extra options for the optimizer.

Return type:

None

predict(data, model=None, return_ui=False, alpha=0.05)[source]#

Generate prediction using regmod model. This function will return predictions for the main_param with given data.

Parameters:

  • data (DataFrame) – A dataset to generate predictions from

  • model (Model | None) – A fitted RegmodModel. If it is None, will use the overall model rather than the cross-validation model.

  • return_ui (bool) – If return_ui=True, a matrix will be returned. The first row is the point prediction, second and thrid rows are the lower and upper bound of the prediction.

  • alpha (float) – When return_ui=True, function will return (1 - alpha) uncertainty interval. By default, alpha=0.05.

Return type:

NDArray

evaluate(data, model=None)[source]#

Given a model and a test set, generate a performance score.

Score is based on the provided evaluation metric, comparing the difference between observed and predicted values.

Parameters:

  • data (DataFrame) – The data set to generate predictions from

  • model (Model | None) – The fitted model to set predictions on. If None will use the overall model rather than the cross-validation model.

Return type:

float