Breaking News

SME

0 0

The structure of this package deal has been designed by Xavier Glorot (https://github.com/glorotxa), with some contributions from Antoine Bordes (https://www.hds.utc.fr/~bordesan).

Replace (Nov 13): the code for Translating Embeddings (see https://everest.hds.utc.fr/doku.php?id=en:transe) has been included together with a brand new model for Freebase (FB15k).

  1. Overview

This package deal proposes scripts utilizing Theano to carry out coaching and analysis on a number of datasets of the fashions:

  • Structured Embeddings (SE) outlined in (Bordes et al., AAAI 2011);
  • Semantic Matching Vitality (SME_lin & SME_bil) outlined in (Bordes et al., MLJ 2013);
  • Translating Embeddings (TransE) outlined in (Bordes et al., NIPS 2013).
  • TATEC outlined in (Garcia-Duran et al., ECML14, arxiv15).

Please discuss with the next pages for extra particulars and references:

Content material of the package deal:

  • mannequin.py : incorporates the lessons and features to create the totally different fashions and Theano features (coaching, analysis…).
  • {dataset}_exp.py : incorporates an experiment perform to coach all of the totally different fashions on a given dataset.
  • The info/ folder incorporates the information information for the educational scripts.
  • within the {dataset}/ folders:
    • {dataset}_parse.py : parses and creates knowledge information for the coaching script of a given dataset.
    • {dataset}_evaluation.py : incorporates analysis features for a given dataset.
    • {dataset}_{model_name}.py : runs one of the best hyperparameters experiment for a given dataset and a given mannequin.
    • {dataset}_{model_name}.out : output we obtained on our machines for a given dataset and a given mannequin utilizing the script above.
    • {dataset}_test.py : carry out fast runs for small fashions of all sorts to check the scripts.

The datasets presently accessible are:

  1. third Celebration Libraries

That you must set up Theano to make use of these scripts. It additionally requires: Python >= 2.4, Numpy >=1.5.0, Scipy>=0.8. The experiment scripts are suitable with Jobman however this library shouldn’t be obligatory.

  1. Set up

Put the script folder in your PYTHONPATH.

  1. Knowledge Recordsdata Creation

Put absolutely the path of the downloaded dataset (from: https://everest.hds.utc.fr/doku.php?id=en:smemlj12 or https://everest.hds.utc.fr/doku.php?id=en:transe) at the start of the {dataset}_parse.py script and run it (the SME folder must be your present listing). Observe: Working Tensor_parse.py generates knowledge for each Kinhsips, UMLS & Nations.

  1. Coaching and Evaluating a Mannequin

Merely run the corresponding {dataset}_{model_name}.py file (the SME/{dataset}/ folder must be your present listing) to launch a coaching. When it is over, working {dataset}_evaluation.py with the trail to the best_valid_model.pkl of the realized mannequin runs the analysis on the check set

  1. Citing

If you happen to use this code, you could possibly present the hyperlink to the github web page: https://github.com/glorotxa/SME . Additionally, relying on the mannequin used, it is best to cite both the paper on Structured Embeddings (Bordes et al., AAAI 2011), on Semantic Matching Vitality (Bordes et al., MLJ 2013) or on Translating Embeddings (Bordes et al., NIPS 2013).

  1. References
  • (Garcia-Duran et al., arxiv 15) Combining Two And Three-Method Embeddings Fashions for Hyperlink Prediction in Data Bases Alberto Garcia-Duran, Antoine Bordes, Nicolas Usunier and Yves Grandvalet. http://arxiv.org/abs/1506.00999
  • (Bordes et al., NIPS 2013) Translating Embeddings for Modeling Multi-relational Knowledge (2013). Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston and Oksana Yakhnenko. In Proceedings of Neural Info Processing Programs (NIPS 26), Lake Taho, NV, USA. Dec. 2013.
  • (Bordes et al., MLJ 2013) A Semantic Matching Vitality Perform for Studying with Multi-relational Knowledge (2013). Antoine Bordes, Xavier Glorot, Jason Weston, and Yoshua Bengio. in Machine Studying. Springer, DOI: 10.1007/s10994-013-5363-6, Could 2013
  • (Bordes et al., AAAI 2011) Studying Structured Embeddings of Data Bases (2011). Antoine Bordes, Jason Weston, Ronan Collobert and Yoshua Bengio. in Proceedings of the twenty fifth Convention on Synthetic Intelligence (AAAI), AAAI Press.

Source

Happy
Happy
0 %
Sad
Sad
0 %
Excited
Excited
0 %
Sleepy
Sleepy
0 %
Angry
Angry
0 %
Surprise
Surprise
0 %

Average Rating

5 Star
0%
4 Star
0%
3 Star
0%
2 Star
0%
1 Star
0%

Leave a Reply

Your email address will not be published. Required fields are marked *