Metrics API¶

anfis_toolbox.metrics ¶

Common metrics utilities for ANFIS Toolbox.

This module provides lightweight, dependency-free metrics that are useful for training and evaluating ANFIS models.

ANFISMetrics ¶

Metrics calculator utilities for ANFIS models.

classification_metrics `staticmethod` ¶

classification_metrics(
    y_true: ArrayLike,
    y_pred: ArrayLike | None = None,
    *,
    y_proba: ArrayLike | None = None,
    logits: ArrayLike | None = None,
) -> dict[str, MetricValue]

Return common classification metrics for encoded targets and predictions.

model_complexity_metrics `staticmethod` ¶

model_complexity_metrics(model: TSKANFIS) -> dict[str, int]

Compute structural statistics for an ANFIS model instance.

regression_metrics `staticmethod` ¶

regression_metrics(
    y_true: ArrayLike, y_pred: ArrayLike
) -> dict[str, MetricValue]

Return a suite of regression metrics for predictions vs. targets.

MetricReport `dataclass` ¶

MetricReport(
    task: Literal["regression", "classification"],
    _values: Mapping[str, MetricValue],
)

Immutable container exposing computed metrics by key or attribute.

getattr ¶

__getattr__(item: str) -> MetricValue

Allow attribute-style access to stored metrics.

getitem ¶

__getitem__(key: str) -> MetricValue

Provide dictionary-style access to metric values.

__post_init__ ¶

__post_init__() -> None

Sanitize stored NumPy scalars/arrays to prevent accidental mutation.

keys ¶

keys() -> Iterable[str]

Expose the metric key iterator from the backing mapping.

to_dict ¶

to_dict() -> dict[str, MetricValue]

Return a shallow copy of the underlying metric mapping.

accuracy ¶

accuracy(y_true: ndarray, y_pred: ndarray) -> float

Compute accuracy from integer/one-hot labels and logits/probabilities.

y_pred can be class indices (n,), logits (n,k), or probabilities (n,k). y_true can be class indices (n,) or one-hot (n,k).

balanced_accuracy_score ¶

balanced_accuracy_score(
    y_true: ndarray, y_pred: ndarray
) -> float

Return the macro-average recall, balancing performance across classes.

classification_entropy ¶

classification_entropy(
    U: ndarray, epsilon: float = 1e-12
) -> float

Classification Entropy (CE). Lower is better (crisper).

Parameters:

Name	Type	Description	Default
`U`	`ndarray`	Membership matrix of shape (n_samples, n_clusters).	required
`epsilon`	`float`	Small constant to avoid log(0).	`1e-12`

Returns:

Type	Description
`float`	CE value as float.

compute_metrics ¶

compute_metrics(
    y_true: ArrayLike,
    *,
    y_pred: ArrayLike | None = None,
    y_proba: ArrayLike | None = None,
    logits: ArrayLike | None = None,
    task: Literal[
        "auto", "regression", "classification"
    ] = "auto",
    metrics: Sequence[str] | None = None,
    custom_metrics: Mapping[str, MetricFn] | None = None,
) -> MetricReport

Compute regression or classification metrics and return a report.

cross_entropy ¶

cross_entropy(
    y_true: ndarray,
    logits: ndarray,
    epsilon: float = _EPSILON,
) -> float

Compute mean cross-entropy from integer labels or one-hot vs logits.

Parameters:

Name	Type	Description	Default
`y_true`	`ndarray`	Array-like of shape (n_samples,) of integer class labels, or one-hot array of shape (n_samples, n_classes).	required
`logits`	`ndarray`	Array-like raw scores, shape (n_samples, n_classes).	required
`epsilon`	`float`	Small constant for numerical stability.	`_EPSILON`

Returns:

Type	Description
`float`	Mean cross-entropy (float).

explained_variance_score ¶

explained_variance_score(
    y_true: ndarray,
    y_pred: ndarray,
    epsilon: float = _EPSILON,
) -> float

Compute the explained variance score for regression predictions.

log_loss ¶

log_loss(
    y_true: ndarray,
    y_prob: ndarray,
    epsilon: float = _EPSILON,
) -> float

Compute mean log loss from integer/one-hot labels and probabilities.

mean_absolute_error ¶

mean_absolute_error(
    y_true: ndarray, y_pred: ndarray
) -> float

Compute the mean absolute error (MAE).

Parameters:

Name	Type	Description	Default
`y_true`	`ndarray`	Array-like of true target values, shape (...,)	required
`y_pred`	`ndarray`	Array-like of predicted values, same shape as y_true	required

Returns:

Type	Description
`float`	The mean of absolute differences over all elements as a float.

Notes

Inputs are coerced to NumPy arrays with dtype=float.
Broadcasting follows NumPy semantics. If shapes are not compatible for element-wise subtraction, a ValueError will be raised by NumPy.

mean_absolute_percentage_error ¶

mean_absolute_percentage_error(
    y_true: ndarray,
    y_pred: ndarray,
    epsilon: float = 1e-12,
    *,
    ignore_zero_targets: bool = False,
) -> float

Compute the mean absolute percentage error (MAPE) in percent.

MAPE = mean( abs((y_true - y_pred) / max(abs(y_true), epsilon)) ) * 100

Parameters:

Name	Type	Description	Default
`y_true`	`ndarray`	Array-like of true target values.	required
`y_pred`	`ndarray`	Array-like of predicted values, broadcastable to y_true.	required
`epsilon`	`float`	Small constant to avoid division by zero when y_true == 0.	`1e-12`
`ignore_zero_targets`	`bool`	When True, drop samples where \|y_true\| <= epsilon; if all targets are (near) zero, returns `np.inf` to signal undefined percentage.	`False`

Returns:

Type	Description
`float`	MAPE value as a percentage (float).

mean_bias_error ¶

mean_bias_error(y_true: ndarray, y_pred: ndarray) -> float

Compute the mean signed error, positive when predictions overshoot.

mean_squared_error ¶

mean_squared_error(
    y_true: ndarray, y_pred: ndarray
) -> float

Compute the mean squared error (MSE).

Parameters:

Name	Type	Description	Default
`y_true`	`ndarray`	Array-like of true target values, shape (...,)	required
`y_pred`	`ndarray`	Array-like of predicted values, same shape as y_true	required

Returns:

Type	Description
`float`	The mean of squared differences over all elements as a float.

Notes

Inputs are coerced to NumPy arrays with dtype=float.
Broadcasting follows NumPy semantics. If shapes are not compatible for element-wise subtraction, a ValueError will be raised by NumPy.

mean_squared_logarithmic_error ¶

mean_squared_logarithmic_error(
    y_true: ndarray, y_pred: ndarray
) -> float

Compute the mean squared logarithmic error (MSLE).

Requires non-negative inputs. Uses log1p for numerical stability: MSLE = mean( (log1p(y_true) - log1p(y_pred))^2 ).

median_absolute_error ¶

median_absolute_error(
    y_true: ndarray, y_pred: ndarray
) -> float

Return the median absolute deviation between predictions and targets.

partition_coefficient ¶

partition_coefficient(U: ndarray) -> float

Bezdek's Partition Coefficient (PC) in [1/k, 1]. Higher is crisper.

Parameters:

Name	Type	Description	Default
`U`	`ndarray`	Membership matrix of shape (n_samples, n_clusters).	required

Returns:

Type	Description
`float`	PC value as float.

pearson_correlation ¶

pearson_correlation(
    y_true: ndarray,
    y_pred: ndarray,
    epsilon: float = _EPSILON,
) -> float

Compute the Pearson correlation coefficient r.

Returns 0.0 when the standard deviation of either input is ~0 (undefined r).

precision_recall_f1 ¶

precision_recall_f1(
    y_true: ArrayLike,
    y_pred: ArrayLike,
    average: Literal["macro", "micro", "binary"] = "macro",
) -> tuple[float, float, float]

Compute precision, recall, and F1 score with the requested averaging.

quick_evaluate ¶

quick_evaluate(
    model: object,
    X_test: ndarray,
    y_test: ndarray,
    print_results: bool = True,
    task: Literal[
        "auto", "regression", "classification"
    ] = "auto",
) -> dict[str, float]

Evaluate a trained ANFIS model or estimator on test data.

r2_score ¶

r2_score(
    y_true: ndarray,
    y_pred: ndarray,
    epsilon: float = _EPSILON,
) -> float

Compute the coefficient of determination R^2.

R^2 = 1 - SS_res / SS_tot, where SS_res = sum((y - y_hat)^2) and SS_tot = sum((y - mean(y))^2). If SS_tot is ~0 (constant target), returns 1.0 when predictions match the constant target (SS_res ~0), otherwise 0.0.

root_mean_squared_error ¶

root_mean_squared_error(
    y_true: ndarray, y_pred: ndarray
) -> float

Compute the root mean squared error (RMSE).

This is simply the square root of mean_squared_error.

softmax ¶

softmax(logits: ndarray, axis: int = -1) -> np.ndarray

Compute a numerically stable softmax along a given axis.

symmetric_mean_absolute_percentage_error ¶

symmetric_mean_absolute_percentage_error(
    y_true: ndarray,
    y_pred: ndarray,
    epsilon: float = _EPSILON,
) -> float

Compute the symmetric mean absolute percentage error (SMAPE) in percent.

Parameters:

Name	Type	Description	Default
`y_true`	`ndarray`	Array-like of true target values.	required
`y_pred`	`ndarray`	Array-like of predicted values, broadcastable to y_true.	required
`epsilon`	`float`	Small constant added to denominator to avoid division by zero.	`_EPSILON`

Returns:

Type	Description
`float`	SMAPE value as a percentage (float).

xie_beni_index ¶

xie_beni_index(
    X: ndarray,
    U: ndarray,
    C: ndarray,
    m: float = 2.0,
    epsilon: float = 1e-12,
) -> float

Xie-Beni index (XB). Lower is better.

XB = sum_i sum_k u_ik^m ||x_i - v_k||^2 / (n * min_{p!=q} ||v_p - v_q||^2)

Parameters:

Name	Type	Description	Default
`X`	`ndarray`	Data array, shape (n_samples, n_features) or (n_samples,).	required
`U`	`ndarray`	Membership matrix, shape (n_samples, n_clusters).	required
`C`	`ndarray`	Cluster centers, shape (n_clusters, n_features).	required
`m`	`float`	Fuzzifier (>1).	`2.0`
`epsilon`	`float`	Small constant to avoid division by zero.	`1e-12`

Returns:

Type	Description
`float`	XB value as float (np.inf when centers < 2).

This module provides comprehensive metrics for evaluating ANFIS models across regression, classification, and clustering tasks.

Regression Metrics¶

Functions for evaluating regression model performance:

mean_squared_error() - Mean squared error
mean_absolute_error() - Mean absolute error
root_mean_squared_error() - Root mean squared error
mean_absolute_percentage_error() - Mean absolute percentage error
symmetric_mean_absolute_percentage_error() - Symmetric MAPE
r2_score() - Coefficient of determination
explained_variance_score() - Explained variance of predictions
median_absolute_error() - Median absolute error (robust to outliers)
mean_bias_error() - Mean prediction bias
pearson_correlation() - Pearson correlation coefficient
mean_squared_logarithmic_error() - Mean squared logarithmic error

Classification Metrics¶

Functions for evaluating classification model performance:

softmax() - Numerically stable softmax
cross_entropy() - Cross-entropy loss
log_loss() - Log loss
accuracy() - Classification accuracy
balanced_accuracy_score() - Macro average of per-class recall
precision_recall_f1() - Precision, recall, and F1 with macro/micro/binary averaging

Clustering Validation¶

Functions for evaluating fuzzy clustering quality:

partition_coefficient() - Bezdek's partition coefficient
classification_entropy() - Classification entropy
xie_beni_index() - Xie-Beni validity index

Metric Reports & Automation¶

compute_metrics() - One-stop helper that infers the task (regression vs. classification) and returns a MetricReport
MetricReport - Read-only container with attribute/dict-style access and a .to_dict() export

Metrics API¶

anfis_toolbox.metrics ¶

ANFISMetrics ¶

classification_metrics staticmethod ¶

model_complexity_metrics staticmethod ¶

regression_metrics staticmethod ¶

MetricReport dataclass ¶

__getattr__ ¶

__getitem__ ¶

__post_init__ ¶

keys ¶

to_dict ¶

accuracy ¶

balanced_accuracy_score ¶

classification_entropy ¶

compute_metrics ¶

cross_entropy ¶

explained_variance_score ¶

log_loss ¶

mean_absolute_error ¶

mean_absolute_percentage_error ¶

mean_bias_error ¶

mean_squared_error ¶

mean_squared_logarithmic_error ¶

median_absolute_error ¶

partition_coefficient ¶

pearson_correlation ¶

precision_recall_f1 ¶

quick_evaluate ¶

r2_score ¶

root_mean_squared_error ¶

softmax ¶

symmetric_mean_absolute_percentage_error ¶

xie_beni_index ¶

Regression Metrics¶

Classification Metrics¶

Clustering Validation¶

Metric Reports & Automation¶

classification_metrics `staticmethod` ¶

model_complexity_metrics `staticmethod` ¶

regression_metrics `staticmethod` ¶

MetricReport `dataclass` ¶

getattr ¶

getitem ¶