WordVectors

Repo for exploring Word2Vec and Doc2Vec

Setup

1. Install Anaconda

We will be using python 3.5 with Anaconda. Download the installer from

https://www.continuum.io/downloads

and follow installation instructions

2. Create conda virtual environment

conda create -n word2vec_env pip

3. Clone this git repo

git clone https://github.com/abtpst/WordVectors.git

4. Install all packages

Navigate to the cloned location and activate the environment created above

activate word2vec_env

Now install the packages as

pip install -r requirements.txt

If you experience trouble installing scipy or numpy, follow these steps. Note that the environment must be activated and you should run these commands from the cloned location.

1. conda install scipy
2. pip install -r requirements.txt

5. Bug fix for gensim

In the versions of gensim, post 0.10.1, you need to make the following change

Modify the file site-packages/gensim/models/word2vec.py by changing line 1013 to

self.syn0[i] = self.seeded_vector(str(self.index2word[i]) + str(self.seed))

This fix is important as otherwise the word2vec training would fail.

6. Tie this virtual environment to the project

The python interpreter for this project should point to the conda virtual environment we created above. Note that we can use this environment for doc2vec as well.

7. Now you are ready to run the scripts

The src folder contains the packages for exploring Word2Vec and Doc2Vec. The respective Readme files with instructions are also provided.