sleec_python

Python implementation of "Sparse Local Embeddings for Extreme Multi-label Classification, NIPS, 2015"

main scripts

• main.py: the whole pipeline

variations

list of variations from the original paper:

1. cosine distance to build the kNN graph for embedding learning
2. Alternating Least Sqaure to learn low-rank embedding Z
   • without l1 regularization
   • the paper uses Singular Value Projection
3. ElasticNet to optimize the linear regressor

• without l1 regularization
• the papers uses ADMM with l1

update

• 2017-11-01:
  • for bibtex dataset, achieved p1, p3 and p5 are 0.5964 (0.6532), 0.3455 (0.3973) and 0.2461 (0.2889) ((..) is score by the paper)
• 2017-11-08:
  • embedding is replaced by VX instead of Z
  • added tensorflow version that adds l1 penalty on VX as well
  • kNN predicion with weight
  • parameter tuning on elacstic net and tensorflow version
  • for bibtex, achieved p1 (0.6215), p3 (0.3716) and p5 (0.2697)

todo