Word2Vec as a service

Why this project ?

Word2Vec is a popular model family used in datascience. To be meaningful, it requires a model of millions of words. Loading those models in a normal laptop can take a hit on ram usage. Why not providing a word2vec implementation trained on the GoogleNews-vectors-negative300 dataset as a service ? Just hit the /word2vec?word=something of this service ad you will get your word vector

Build and run it

sudo docker compose up -d --build

Just run it

sudo docker compose up 
sudo docker compose stop <id of container>

You can send GET requests to http://localhost:/word2vec?word=something and you will get the array of the word you selected.

You can also send POST requests to http://localhost:/words2vec

For example:

import requests

import requests, json 
vec_l = None
try:
    r = requests.post('http://triads-dl.dbmi.pitt.edu:18000/words2vec', json={'words':words})
    if r.status_code == 200:
        print('Vectors retrieved. Loading...')
        vec_l = json.loads(r.text)
except KeyError:
    print(f'ERROR - could not retrieve vectors - status code: {r.status_code}')

word2Ind = {}
M = []
i = 0
for w in words:
    if i == len(vec_l["data"]):
        break 
    M.append(vec_l["data"][i][1])
    word2Ind[w] = i
    i += 1