coco.py (Apurva) - Coco JSON parser and ID mapping
embedder.py (Lashika) - Text tokenization and the glove vectorizer
model.py (Shanya and Evgeny) - The actual model (pytorch) + save/load
train.py (Sriram) - Training loop + gradient descent steps
database.py (Nihanth) - Matrix search on DB and image loader
main.py (Whoever done) - Actual runnable final script
Trained weights are in data/W_embed.npy (512 to 200). Re-run train.py to regenerate
Libraries: numpy mygrad gensim requests pillow torch torchvision