Word embedding evaluation

And vector represented word embedding can be evaluated for different tasks. E.g. word similarity, word analogy, doc representation etc.

Evaluation tasks that we support

1. word similarity

Use human labelled similarity score between select word, compare similarity score generated by embedding vector. The closer to human similarity score, the closer to human judgement.

2. doc classification

Represent doc using a combination of word embedding. The easiest way is to average (too naive). Can use RNN to generate better representation. Use representation and doc embedding to classify.

How to use

First install required

pip install --editable .

Next run via command line

WordEmbEval --out_dir=output/eval --emb_path=/tmp/glove.twitter.27B.100d.txt --csv_separator=" " --quoting=3