issues
search
juanmc2005
/
SimilarityLearning
Similarity Learning applied to Speaker Verification and Semantic Textual Similarity
12
stars
7
forks
source link
Newest Changes - From 21 June to 5 July
#33
Closed
juanmc2005
closed
5 years ago
juanmc2005
commented
5 years ago
Changelog
General
Added file logs and improved logging
Added experiment IDs. Logs for an experiment are stored in a separate folder
Training information plots are created automatically based on logs as soon as the training is finished or cancelled
Removed
main.py
script. Each task is trained using its own script
Script arguments are dumped to a configuration file inside the experiment folder
New script parameters
General refactoring
STS
Triplet loss for SemEval
Bugfix:
SemEval augmented pairs considered positive and negative at the same time
t-SNE sentence embedding visualization improved using nearest neighbors
Speaker Verification
Speaker distance historgram plots
Training accuracy calculation as a plugin
Added additional test evaluation in speaker verification for debugging purposes
Support for VoxCeleb2 (although no
development
partition available)
t-SNE visualization for speaker embeddings
Changelog
General
main.py
script. Each task is trained using its own scriptSTS
Speaker Verification
development
partition available)