#include <featurizer_metrics.hpp>
|
| double | train_fraction = 0.5 |
| | The share of inputs used to fit the probe and pick the main features; the rest are scored.
|
| |
| uint64_t | seed = 0 |
| |
| double | min_f1_gain = 0.03 |
| | Main features are added, best first, while each raises F1 by at least this much (SAEBench's k-sparse probing threshold).
|
| |
| int64_t | max_main_features = 10 |
| |
| double | min_cosine = 0.025 |
| | An absorbing feature's decoder direction must have at least this cosine with the probe direction (SAEBench: 0.025).
|
| |
| double | min_share = 0.4 |
| | And absorbing features must carry at least this share of the input's projection onto the probe direction (SAEBench: 0.4).
|
| |
| int64_t | probe_epochs = 30 |
| |
| float | probe_learning_rate = 1e-2f |
| |
| int64_t | batch = 4096 |
| |
◆ batch
| int64_t pulsatrix::AbsorptionOptions::batch = 4096 |
◆ max_main_features
| int64_t pulsatrix::AbsorptionOptions::max_main_features = 10 |
◆ min_cosine
| double pulsatrix::AbsorptionOptions::min_cosine = 0.025 |
An absorbing feature's decoder direction must have at least this cosine with the probe direction (SAEBench: 0.025).
◆ min_f1_gain
| double pulsatrix::AbsorptionOptions::min_f1_gain = 0.03 |
Main features are added, best first, while each raises F1 by at least this much (SAEBench's k-sparse probing threshold).
◆ min_share
| double pulsatrix::AbsorptionOptions::min_share = 0.4 |
And absorbing features must carry at least this share of the input's projection onto the probe direction (SAEBench: 0.4).
◆ probe_epochs
| int64_t pulsatrix::AbsorptionOptions::probe_epochs = 30 |
◆ probe_learning_rate
| float pulsatrix::AbsorptionOptions::probe_learning_rate = 1e-2f |
◆ seed
| uint64_t pulsatrix::AbsorptionOptions::seed = 0 |
◆ train_fraction
| double pulsatrix::AbsorptionOptions::train_fraction = 0.5 |
The share of inputs used to fit the probe and pick the main features; the rest are scored.
The documentation for this struct was generated from the following file: