Local metrics and dataset evaluation

Module contract

These local utility functions support RMSE, AUC, and dataset evaluation. AxoBench introduces and implements the report’s Mean F1, Voltage SERA, and Dynamics SERA protocol; use axosim-evaluate-model for that core metric set. SERA uses squared error; Root-SERA is its square-root presentation.

Source revision: 306a51ed950b. Public export index.

soma_rmse

Source

soma_rmse(prediction: np.ndarray, target: np.ndarray) -> float

Parameter	Type	Default
prediction	np.ndarray	required
target	np.ndarray	required

Returns float.

binary_auc

Source

binary_auc(scores: np.ndarray, labels: np.ndarray) -> float

Parameter	Type	Default
scores	np.ndarray	required
labels	np.ndarray	required

Returns float.

spike_auc

Source

spike_auc(prediction: np.ndarray, target: np.ndarray) -> float

Parameter	Type	Default
prediction	np.ndarray	required
target	np.ndarray	required

Returns float.

evaluate_dataset

Source

evaluate_dataset(model: BranchELM, dataset: ShardedNeuronIODataset, *, batch_size: int=8, device: str='cpu', soma_units: str='millivolts', y_train_soma_scale: float=DEFAULT_Y_TRAIN_SOMA_SCALE, ignore_start: int=0, mask_mode: str='ignore-start', stitch_burn_in: int=150, soma_affine_calibration: bool=False, pin_memory: bool=False, non_blocking: bool=True) -> dict[str, float | int]

Parameter	Type	Default
model	BranchELM	required
dataset	ShardedNeuronIODataset	required
batch_size	int	8
device	str	'cpu'
soma_units	str	'millivolts'
y_train_soma_scale	float	DEFAULT_Y_TRAIN_SOMA_SCALE = 0.1
ignore_start	int	0
mask_mode	str	'ignore-start'
stitch_burn_in	int	150
soma_affine_calibration	bool	False
pin_memory	bool	False
non_blocking	bool	True

batch_size: Examples processed per batch. device: Execution or allocation device. ignore_start: Initial native timesteps excluded by the declared path.

Returns dict[str, float \| int].

write_metrics

Source

write_metrics(metrics: dict[str, float | int], output: str | Path) -> None

Parameter	Type	Default
metrics	dict[str, float | int]	required
output	str | Path	required

Returns None.