Readings¶
| reading | what it measures |
|---|---|
knn_prediction |
leave-one-out k-NN accuracy against the model's predictions |
correctness_auc |
AUC of the geometric margin as a predictor of the model being right |
confusion_agreement |
Spearman between class proximity and the model's (soft) confusion |
containment |
class directions inside the retained subspace (r = K − 1) |
e_ratio_error_auc |
AUC of the orthogonal-energy ratio as a predictor of an error |
READINGS
module-attribute
¶
READINGS = (
"knn_prediction",
"correctness_auc",
"confusion_agreement",
"containment",
"e_ratio_error_auc",
)
auc
¶
Source code in src\fannar\readings.py
spearman
¶
Source code in src\fannar\readings.py
knn_prediction
¶
Source code in src\fannar\readings.py
class_distance_matrix
¶
Source code in src\fannar\readings.py
geometric_margin
¶
Source code in src\fannar\readings.py
correctness_auc
¶
Source code in src\fannar\readings.py
soft_confusion
¶
Source code in src\fannar\readings.py
count_confusion
¶
Source code in src\fannar\readings.py
confusion_agreement
¶
Source code in src\fannar\readings.py
e_ratio_error_auc
¶
e_ratio_error_auc(space: GramSpace, labels, predictions, r=None) -> float
Source code in src\fannar\readings.py
evaluate
¶
evaluate(
space: GramSpace,
labels,
predictions,
probabilities=None,
k: int = 5,
min_errors: int = 5,
) -> dict