adapt.metrics.reverse_validation
- adapt.metrics.reverse_validation(model, Xs, ys, Xt, **fit_params)[source]
Reverse validation.
The reverse validation score is computed as a source error by inversing the role of the source and the target domains. A clone of the model is trained to adapt from the target to the source using the model target predictions as pseudo target labels. Then the final score is computed between the source prediction of the cloned model and the groundtruth.
- Parameters
- modelBaseAdapt instance
Adaptation model
- Xsarray
Source input array
- ysarray
Source output array
- Xtarray
Target input array
- fit_paramskey, value arguments
Parameters for the fit method of the cloned model.
- Returns
- scorefloat
See also
domain_classifierDANN
References
- 1
[1] Y. Ganin, E. Ustinova, H. Ajakan, P. Germain, H. Larochelle, F. Laviolette, M. Marchand, and V. Lempitsky. “Domain-adversarial training of neural networks”. In JMLR, 2016.