RACER package

Submodules

RACER.RACER module

class RACER.RACER.RACER(alpha=0.9, suppress_warnings=False, benchmark=False)

Bases: object

_bool2str(bool_arr: ndarray) → str

Converts a boolean array to a human-readable string

Args:

bool_arr (np.ndarray): The input boolean array

Returns:

str: Human-readable string output

_closest_match(X: ndarray) → ndarray

Find the closest matching rule to X (This will be extended later)

Args:

X (np.ndarray): Input rule X

Returns:

np.ndarray: Matched rule

_composable(idx1: int, idx2: int) → bool

Returns true if two rules indicated by their indices are composable

Args:

idx1 (int): Index of the first rule idx2 (int): Index of the second rule

Returns:

bool: True if labels match and neither of the rules are covered. False otherwise.

_compose(rule1: ndarray, rule2: ndarray) → ndarray

Composes rule1 with rule2

Args:

rule1 (np.ndarray): The first rule rule2 (np.ndarray): The second rule

Returns:

np.ndarray: The composed rule which is simply the bitwise OR of the two rules

_confusion(rule_if: ndarray, rule_then: ndarray) → Tuple[ndarray, ndarray]

Returns n_correct and n_covered for instances classified by a rule.

Args:

rule_if (np.ndarray): If part of rule (x) rule_then (np.ndarray): Then part of rule (y)

Returns:

Tuple[np.ndarray, np.ndarray]: (n_covered, n_correct)

_covered(X: ndarray, rule_if: ndarray) → ndarray

Returns indices of instances if X that are covered by rule_if. Note that rule covers instance if EITHER of the following holds in a bitwise manner: 1. instance[i] == 0 2. instance[i] == 1 AND rule[i] == 1

Args:

X (np.ndarray): Instances rule_if (np.ndarray): If part of rule (x)

Returns:

np.ndarray: An array containing indices in X that are covered by rule_if

_create_init_rules() → None

Creates an initial set of rules from theinput feature vectors

_finalize_rules() → None

Removes redundant rules to form the final ruleset

_fitness_fn(rule_if: ndarray, rule_then: ndarray) → ndarray

Returns fitness for a given rule according to the RACER paper

Args:

rule_if (np.ndarray): If part of a rule (x) rule_then (np.ndarray): Then part of a rule (y)

Returns:

np.ndarray: Fitness score for the rule as defined in the RACER paper

_generalize_extants() → None

Generalize the extants by flipping every 0 to a 1 and checking if the fitness improves.

_get_majority() → ndarray

Return the majority rule_then from self._y

Returns:

np.ndarray: Majority rule_then

_label_to_int(label: ndarray) → int

Converts dummy label to int

Args:

label (np.ndarray): Label to convert

Returns:

int: Converted label

_process_rules(idx1: int, idx2: int) → None

Process two rules indiciated by their indices

Args:

idx1 (int): Index of the first rule idx2 (int): Index of the second rule

_update_extants(index: int, new_rule_if: ndarray, new_rule_then: ndarray, new_rule_fitness: ndarray)

Remove all rules from current extants that are covered by new_rule. Then append new rule to extants.

Args:

index (int): Index of new_rule new_rule_if (np.ndarray): If part of new_rule (x) new_rule_then (np.ndarray): Then part of new_rule (y) new_rule_fitness (np.ndarray): Fitness of the new_rule

display_rules() → None

Print out the final rules

fit(X: ndarray, y: ndarray) → None

Fits the RACER algorithm on top of input data X and targets y. The code is written in close correlation to the pseudo-code provided in the RACER paper with some slight modifications.

Args:

X (np.ndarray): Features vector y (np.ndarray): Targets vector

predict(X: ndarray, convert_dummies=True) → ndarray

Given input X, predict label using RACER

Args:

X (np.ndarray): Input features vector convert_dummies (bool, optional): Whether to convert dummy labels back to integert format. Defaults to True.

Returns:

np.ndarray: Label as predicted by RACER

score(X_test: ndarray, y_test: ndarray) → float

Returns accuracy on the provided test data.

Args:

X_test (np.ndarray): Test features vector y_test (np.ndarray): Test targets vector

Returns:

float: Accuracy score

RACER.RACER.XNOR(input: ndarray, other: ndarray) → ndarray

Computes the XNOR gate. (semantically the same as input == other)

Args:

input (np.ndarray): Input array other (np.ndarray): Other input array

Returns:

np.ndarray: XNOR(input, other) as an array

RACER.preprocessing module

class RACER.preprocessing.RACERPreprocessor(target: str = 'auto', max_n_bins=32, max_num_splits=32, use_optimal_quantizer=False)

Bases: object

fit(X: DataFrame | ndarray, y: DataFrame | ndarray)

Fits the preprocessor on X and y for downstream transformations.

Args:

X (Union[pd.DataFrame, np.ndarray]): Features vector y (Union[pd.DataFrame, np.ndarray]): Targets vector

fit_transform(X: DataFrame | ndarray, y: DataFrame | ndarray) → Tuple[ndarray, ndarray]

Preprocesses the dataset by replacing nominal vaues with dummy variables. Converts to numpy boolean arrays and returns the dataset. All numerical values are discretized using an optimal binning strategy that employs a decision tree as a preprocessing step.

Args:

X (Union[pd.DataFrame, np.ndarray]): Features matrix y (Union[pd.DataFrame, np.ndarray]): Targets vector

Returns:

Tuple[np.ndarray, np.ndarray]: Transformed features matrix and targets vectors.

fit_transform_pandas(X: DataFrame | ndarray, y: DataFrame | ndarray) → Tuple[ndarray, ndarray]

Preprocesses the dataset by replacing nominal vaues with dummy variables. Converts to numpy boolean arrays and returns the dataset. All numerical values are discretized using an optimal binning strategy that employs a decision tree as a preprocessing step. (This uses the legacy pandas dummy encoder. You can use this to retain total backward compatibility with previous code)

Args:

X (Union[pd.DataFrame, np.ndarray]): Features matrix y (Union[pd.DataFrame, np.ndarray]): Targets vector

Returns:

Tuple[np.ndarray, np.ndarray]: Transformed features matrix and targets vectors.

transform(X: DataFrame | ndarray, y: DataFrame | ndarray) → Tuple[ndarray, ndarray]

Transforms the provided new X and y with previously fitted preprocessor.

Args:

X (Union[pd.DataFrame, np.ndarray]): Features matrix y (Union[pd.DataFrame, np.ndarray]): Targets vector

Returns:

Tuple[np.ndarray, np.ndarray]: Transformed features matrix and targets vectors.

RACER.version module

Module contents