Metrics
Metrics for evaluating neural network performance.
-
nabla.nn.utils.metrics.accuracy(predictions, targets)[source]
Compute classification accuracy.
- Parameters:
predictions (Array) – Model predictions - either logits/probabilities [batch_size, num_classes]
or class indices [batch_size]
targets (Array) – True labels - either one-hot [batch_size, num_classes] or indices [batch_size]
- Returns:
Scalar accuracy value between 0 and 1
- Return type:
Array
-
nabla.nn.utils.metrics.top_k_accuracy(predictions, targets, k=5)[source]
Compute top-k classification accuracy.
- Parameters:
predictions (Array) – Model predictions (logits or probabilities) [batch_size, num_classes]
targets (Array) – True labels [batch_size] (sparse format)
k (int) – Number of top predictions to consider
- Returns:
Scalar top-k accuracy value between 0 and 1
- Return type:
Array
-
nabla.nn.utils.metrics.precision(predictions, targets, num_classes, class_idx=0)[source]
Compute precision for a specific class.
Precision = TP / (TP + FP)
- Parameters:
predictions (Array) – Model predictions (logits) [batch_size, num_classes]
targets (Array) – True labels (sparse) [batch_size]
num_classes (int) – Total number of classes
class_idx (int) – Class index to compute precision for
- Returns:
Scalar precision value for the specified class
- Return type:
Array
-
nabla.nn.utils.metrics.recall(predictions, targets, num_classes, class_idx=0)[source]
Compute recall for a specific class.
Recall = TP / (TP + FN)
- Parameters:
predictions (Array) – Model predictions (logits) [batch_size, num_classes]
targets (Array) – True labels (sparse) [batch_size]
num_classes (int) – Total number of classes
class_idx (int) – Class index to compute recall for
- Returns:
Scalar recall value for the specified class
- Return type:
Array
-
nabla.nn.utils.metrics.f1_score(predictions, targets, num_classes, class_idx=0)[source]
Compute F1 score for a specific class.
F1 = 2 * (precision * recall) / (precision + recall)
- Parameters:
predictions (Array) – Model predictions (logits) [batch_size, num_classes]
targets (Array) – True labels (sparse) [batch_size]
num_classes (int) – Total number of classes
class_idx (int) – Class index to compute F1 score for
- Returns:
Scalar F1 score for the specified class
- Return type:
Array
-
nabla.nn.utils.metrics.mean_squared_error_metric(predictions, targets)[source]
Compute MSE metric for regression tasks.
- Parameters:
predictions (Array) – Model predictions [batch_size, …]
targets (Array) – True targets [batch_size, …]
- Returns:
Scalar MSE value
- Return type:
Array
-
nabla.nn.utils.metrics.mean_absolute_error_metric(predictions, targets)[source]
Compute MAE metric for regression tasks.
- Parameters:
predictions (Array) – Model predictions [batch_size, …]
targets (Array) – True targets [batch_size, …]
- Returns:
Scalar MAE value
- Return type:
Array
-
nabla.nn.utils.metrics.r_squared(predictions, targets)[source]
Compute R-squared (coefficient of determination) for regression tasks.
R² = 1 - (SS_res / SS_tot)
where SS_res = Σ(y_true - y_pred)² and SS_tot = Σ(y_true - y_mean)²
- Parameters:
predictions (Array) – Model predictions [batch_size, …]
targets (Array) – True targets [batch_size, …]
- Returns:
Scalar R² value
- Return type:
Array
-
nabla.nn.utils.metrics.pearson_correlation(predictions, targets)[source]
Compute Pearson correlation coefficient.
- Parameters:
predictions (Array) – Model predictions [batch_size, …]
targets (Array) – True targets [batch_size, …]
- Returns:
Scalar correlation coefficient
- Return type:
Array
