Skip-Gram-Model-PyTorch

PyTorch implementation of the word2vec (skip-gram model) and visualization of the trained embeddings using TSNE !

My TensorFlow implemntation of Skip-Gram Model can be found here.

Requirements

• torch >= 1.4
• numpy >= 1.18
• matplotlib
• tqdm
• nltk
• gensim

Training

python main.py

Visualizing real-time training loss in Tensorboard

tensorboard --logdir <PATH_TO_TENSORBOARD_EVENTS_FILE>

NOTE: By default, PATH_TO_TENSORBOARD_EVENTS_FILE is set to SUMMARY_DIR in config.py

Testing

python test.py

Inference

war	india	crime	guitar	movies	desert	physics	religion	football	computer
invasion	provinces	will	bass	movie	shore	mathematics	judaism	baseball	digital
soviet	pakistan	prosecution	drum	albums	hilly	mathematical	islam	championship	computers
troop	mainland	accusations	solo	songs	plateau	chemistry	religions	basketball	software
army	asian	provoke	quartet	cartoon	basin	theoretical	religious	coach	electronic
ally	colonial	prosecute	vocals	animate	highlands	analysis	jewish	wrestler	interface

Blog-Post

Check out my blog post on word2vec here.