CS224n: Natural Language Processing with Deep Learning Assignments Winter, 2019
- Python 3.5+
- Pytorch
-
Count-Based Word Vectors: Co-Occurrence Word Embeddings and Matrix, SVD for its dimentional reduction.
-
Cosine Similarity and Distance: searching Polysemous Words, Synonyms and Analogies, Bias.
- Sigmoid function, softmax, negative sampling loss and gradient functions' implementations.
- SGD implementation.
- Training word vectors, and later applying them to a simple sentiment analysis task using Stanford Sentiment Treebank (SST) dataset.
Dependency parsing:
- Transition-based parsing (Stack + buffer)
- Malt parser performed by dense Neural Net and Word Embedding
Model is based on:
- LSTM
- Attention
Model is based on previous assignment. Differences:
- Character-level embedding for encoder
- Character-level decoding for words
YouTube playlist with lectures:
CS224n official website: