|
pulsatrix 1.1.0
|
A probe on the real task beside the same probe on the control task. More...
#include <probe_controls.hpp>

Public Attributes | |
| ProbeResult | task |
| ProbeResult | control |
| double | selectivity = 0 |
task.accuracy - control.accuracy, Hewitt and Liang's selectivity. | |
| double | balanced_selectivity = 0 |
| The same with balanced accuracy, for imbalanced labels. | |
| int64_t | train_types = 0 |
| How many token types the training rows hold. Few types make the control easy to learn; with no more types than dimensions a linear probe can learn any control. | |
A probe on the real task beside the same probe on the control task.
| double pulsatrix::ProbeSelectivity::balanced_selectivity = 0 |
The same with balanced accuracy, for imbalanced labels.
| ProbeResult pulsatrix::ProbeSelectivity::control |
| double pulsatrix::ProbeSelectivity::selectivity = 0 |
task.accuracy - control.accuracy, Hewitt and Liang's selectivity.
| ProbeResult pulsatrix::ProbeSelectivity::task |
| int64_t pulsatrix::ProbeSelectivity::train_types = 0 |
How many token types the training rows hold. Few types make the control easy to learn; with no more types than dimensions a linear probe can learn any control.