ML4

Mini-Project 4 - Task 1

Alexandre Banon; Vincent Delmas; Michael Haaf

Paper :

Sanjeev Arora, Yingyu Liang, and Tengyu Ma. 2017. A simple but tough-to-beat baseline for sentence embed-dings. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings.

Files :

<Sentence_embedding.py> contains our reproduction of the algorithm 1 from the paper (Arora et al.,2017).

<function.py> and <frequence.py> contain useful functions for the main algorithm.

<figure_2_a.py> contains a script to produce the figure 2 of our report.

<environment.py> check if the work space is well organized, and shares the link to dowload the datasets.

Python libraries :

numpy, time, sys, os, math, nltk, codecs, pandas, csv, matplotlib

Links for Dowloading the data :

STS (2012 to 2015) and SICK 2014

GloVe embedding pre-trained vectors

PSL embedding pre-trained vectors

enwiki database from Wikipedia articles

Parameters of the function [Algo] :

[a] A number, parameter of the weight factor of the method proposed by the paper

[task] Possible values : "STS 2012", "STS 2013", "STS 2014", "STS 2015", "SICK 2014"

[methode] Possible values : "WR", "avg", "bin", "g", "h"

[word_embedding] Possible values : "GloVe", "PSL"