/gdp

gdp is generating distributed representation code sets written by pytorch. This code sets is including skip gram and cbow.

Primary LanguagePythonMIT LicenseMIT

gdp

gdp is generating distributed representation code sets written by pytorch.

This code sets is including skip gram and cbow.

Installation

Dependencies

gdp requires:

  • python 3.6.6
  • pytorch-cpu 1.0.0
  • numpy 1.15.4
  • tqdm 4.28.1

User installation

You can install gdp running the following commands.

pip install git+https://github.com/RottenFruits/gdp

Example

skip gram

This is example that run simple skip gram.

from gdp import distributed_representation as dr
from gdp import corpus as cp

data = [
    'he is a king',
    'she is a queen',
    'he is a man',
    'she is a woman',
    'warsaw is poland capital',
    'berlin is germany capital',
    'paris is france capital',
]

corpus = cp.Corpus(data = data, mode = "a", max_vocabulary_size = 5000, max_line = 0, 
                   minimum_freq = 0)

window_size = 1
embedding_dims = 30
batch_size = 128

dr_sg = dr.DistributedRepresentation(corpus, embedding_dims, window_size, batch_size, 
                                       model_type = "skip-gram", ns = 0, trace = True)
dr_sg.train(num_epochs = 101, learning_rate = 0.05)

skip gram with negative sampling

If you want to use negative sampling is this.

dr_sgns = dr.DistributedRepresentation(corpus, embedding_dims, window_size, batch_size, 
                                       model_type = "skip-gram", ns = 1, negative_samples = 5, trace = True)
dr_sgns.train(num_epochs = 101, learning_rate = 0.05)

etc

If you want to use cbow architecture, you should replace model_type "skip-gram" to "cbow".

And more example code is in example directory, please check it too.

Distributed representations

gdp inclues:

  • skipgram
  • skipgram with negative sampling
  • cbow
  • cbow with negative sampling

Reference