explainable_rl.data_handler package

Submodules

explainable_rl.data_handler.data_handler module

class DataHandler(hyperparam_dict, dataset, test_dataset=None)

Bases: object

Data Handler which stores and preprocesses data needed for training.

__init__(hyperparam_dict, dataset, test_dataset=None)

Initialise the DataHandler.

Parameters

• hyperparam_dict (dict) – Dictionary of hyperparameters.

• dataset (pd.DataFrame) – Training dataset.

• test_dataset (pd.DataFrame) – Test dataset.

_filter_data()

Filter the dataset.

_fit_standard_scalars()

Train the sklearn MinMaxScaler and store one per column.

_get_labels(label_dict)

Get the labels from the label dictionary.

Parameters

label_dict (dict) – The label dictionary.

Returns

The labels.

Return type

list

_inverse_transform_col(col_name: str)

Reverse the normalisation of one column of the dataset.

Parameters

col_name (str) – The column name.

_transform_col(col_name: str)

Normalise one column of the dataset.

Parameters

col_name (str) – The column name.

get_action_labels()

Get the action labels.

Returns

Action labels.

Return type

list

get_actions(split='train')

Get the actions taken in the dataset.

Parameters

split (str) – Specifies train or test split.

Returns

Actions.

Return type

pd.DataFrame

get_rewards(split='train')

Get the rewards taken in the dataset.

Parameters

split (str) – Specifies train or test split.

Returns

The rewards.

Return type

pd.DataFrame

get_states(split='train')

Get the states taken in the dataset.

Parameters

split (str) – Specifies train or test split.

Returns

The states.

Return type

pd.DataFrame

normalise_dataset(cols_to_norm=None)

Normalise the dataset to centre with mean zero and variance one.

Parameters

cols_to_norm (list) – The column names that need normalising.

prepare_data_for_engine(cols_to_normalise=None)

Prepare the data to be given to the engine.

Parameters

cols_to_normalise (list) – List of columns to normalise.

preprocess_data(normalisation=True, columns_to_normalise=None)

Preprocess data into state, action and reward spaces.

Preprocessing applies shuffling, normalisation (if selected) and splits the dataset into states, actions and rewards.

Parameters

• normalisation (bool) – True if normalisation is to be applied.

• columns_to_normalise (list) – Columns on which to apply normalisation. If left empty all columns will be normalised.

reverse_norm()

Reverse the normalising of the dataset.

Module contents