|
pulsatrix 1.1.0
|
What the caller has: the model, its inputs and its data. More...
#include <explanation_planner.hpp>
Public Attributes | |
| ModelAccess | access = ModelAccess::BlackBox |
| The most the model gives, from a prediction function only (BlackBox) to an LRP rule at every layer (Relevance). Each level includes the ones before it. | |
| InputKind | input = InputKind::Any |
| What the model takes. Any means unknown: explainers for one kind of input are then kept, with that kind as a condition. | |
| bool | can_reinitialize = false |
| A pulsatrix Module whose weights can be redrawn and restored, for the randomization test and the null-model baseline. | |
| bool | has_dataset = false |
| Many inputs, not just the one being explained: global explainers need them, and so does aggregating explanations over a dataset. | |
| bool | has_labels = false |
| The dataset's true labels, and a slice for each input (a group, a data source), for the per-slice error check (KS-3). | |
| bool | has_base_model = false |
| The model this one was fine-tuned from, for explainers that compare the two. | |
| int64_t | features = 0 |
| How many features each input has, or 0 if unknown; explainers with a limit (ExplainerCapabilities::max_features) are excluded above it. | |
What the caller has: the model, its inputs and its data.
| ModelAccess pulsatrix::ModelTraits::access = ModelAccess::BlackBox |
The most the model gives, from a prediction function only (BlackBox) to an LRP rule at every layer (Relevance). Each level includes the ones before it.
| bool pulsatrix::ModelTraits::can_reinitialize = false |
A pulsatrix Module whose weights can be redrawn and restored, for the randomization test and the null-model baseline.
| int64_t pulsatrix::ModelTraits::features = 0 |
How many features each input has, or 0 if unknown; explainers with a limit (ExplainerCapabilities::max_features) are excluded above it.
| bool pulsatrix::ModelTraits::has_base_model = false |
The model this one was fine-tuned from, for explainers that compare the two.
| bool pulsatrix::ModelTraits::has_dataset = false |
Many inputs, not just the one being explained: global explainers need them, and so does aggregating explanations over a dataset.
| bool pulsatrix::ModelTraits::has_labels = false |
The dataset's true labels, and a slice for each input (a group, a data source), for the per-slice error check (KS-3).
| InputKind pulsatrix::ModelTraits::input = InputKind::Any |
What the model takes. Any means unknown: explainers for one kind of input are then kept, with that kind as a condition.