CS224N-2020-winter

Information

• Course Website
• Assignments
• Slides
• Videos

TODO

• Assignment 1: Lecture 1 (2020/10/15)
• Assignment 2: Lecture 1-2 (2020/10/29)
• Assignment 3: Lecture 5 (2020/10/30)
• Assignment 4: Lecture 6-8 (2020/11/4)
• Assignment 5: Lecture 11-12

---

Notes

• looperxx.github.io

---

Assignment 1

Co-Occurrence Plot Analysis

GloVe Plot Analysis

Polysemes and homonyms

leaves: ends leaf stems takes leaving grows flowers turns leave goes book: books author novel published memoir wrote written essay biography autobiography keep: keeping kept them sure need putting trying keeps want enough raise: raising raised raises increase interest help reduce boost would rates

• Because for some words, the meaning that used by most times are very similar, while the rarely used meaning of polysemes will not show.

Synonyms happy, cheerful have cosine distance: 0.5172466933727264 Antonyms happy, sad have cosine distance: 0.4040136933326721

Synonyms dinner, supper have cosine distance: 0.5171529948711395 Antonyms dinner, breakfast have cosine distance: 0.2351711392402649

Possible explanation:

• cos(antonyms) > cos(synonyms): Because antonyms may be used in different contexts like cheerful is more formal than happy, so they may have large cosine distance; While for synonyms, happy and sad, they are both not formal and can replace each other to present different meanings, but still have very similar contexts.

---

Assignment 2

---

Assignment 3

Epoch 10 out of 10
100%|██████████| 1848/1848 [01:38<00:00, 18.72it/s]
Average Train Loss: 0.06704435556385166
Evaluating on dev set
1445850it [00:00, 32952250.34it/s]      
- dev UAS: 88.79
New best dev UAS! Saving model.

======================================
TESTING
======================================
Restoring the best model weights found on the dev set
Final evaluation on test set
2919736it [00:00, 49365750.20it/s]      
- test UAS: 89.01
Done!

Assignment 4

Assignment 5

! Caution

When using torch.tensor(), you must manually state dtype and device.

Without state dtype, one error may occur:

RuntimeError: Expected tensor for argument #1 'indices' to have scalar type Long; but got torch.FloatTensor instead (while checking arguments for embedding)

Which means you should state dtype = torch.long.