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.
- torch >= 1.4
- numpy >= 1.18
- matplotlib
- tqdm
- nltk
- gensim
python main.py
tensorboard --logdir <PATH_TO_TENSORBOARD_EVENTS_FILE>
NOTE: By default, PATH_TO_TENSORBOARD_EVENTS_FILE is set to SUMMARY_DIR in config.py
python test.py
war | india | crime | guitar | movies | desert | physics | religion | football | computer |
---|---|---|---|---|---|---|---|---|---|
fight | europe | despite | band | movie | region | theory | religious | team | program |
battle | central | help | play | series | along | mathematics | christian | win | systems |
army | western | seek | record | show | western | mathematical | regard | sport | available |
force | indian | challenge | piece | film | southern | study | tradition | club | design |
ally | part | fail | star | feature | plain | science | christianity | league | information |
Check out my blog post on word2vec here.