Score normalizations
1 unresolved thread
1 unresolved thread
This is a long MR with the following features:
-
Proper checkpointing of scores (BioAlgCheckpointWrapper). Still in pickle, though -
Moved the score-writer outside of the VanillaBiometricsPipeline -
Simplified the scoring by BioAlgorithm
. Also made once SampleSet, dump one score. -
Implemented a new pipeline that does Z, T and ZT score normalizations. Also created wrappers for Checkpointing and Dask this pipeline. Such pipeline has to wrap a VanillaBiometricsPipeline to work. -
Adapted the CLI command, bob pipelines vanilla-biometrics
to work with ZNorm. Now, if you want to ztnorm a pipeline, just dobob pipelines vanilla-biometrics --ztnorm
Don't know if someone is willing to review this one.
Edited by Tiago de Freitas Pereira
Merge request reports
Activity
added 1 commit
- f1473028 - Implemented ZT-Normalization. More tests should be implemented.
added 1 commit
- ed187896 - Got it right the paralellization of ZT-Norm computation
added 1 commit
- dd511881 - Implemented ZTNorm for the CheckpointWrapper
added 1 commit
- dc5c2528 - Changed the standard deviation function to get the same output as bob.learn.em.ztnorm
- Resolved by Tiago de Freitas Pereira
278 # Axi1 1=TNORM 279 if axis == 1: 280 return np.sqrt( 281 np.sum( 282 ( 283 norm_base_scores 284 - np.tile( 285 mu.reshape(norm_base_scores.shape[0], 1), 286 (1, norm_base_scores.shape[1]), 287 ) 288 ) 289 ** 2, 290 axis=1, 291 ) 292 / (norm_base_scores.shape[1] - 1) 293 ) We could finally close this one
Closes #76 (closed)
assigned to @tiago.pereira
mentioned in commit b72b8022
mentioned in issue #105 (closed)
mentioned in issue #76 (closed)
Please register or sign in to reply