Skip to content

Instantly share code, notes, and snippets.

@lgalke
Last active February 22, 2022 23:46
Show Gist options
  • Save lgalke/febaaa1313d9c11f3bc8240defed8390 to your computer and use it in GitHub Desktop.
Save lgalke/febaaa1313d9c11f3bc8240defed8390 to your computer and use it in GitHub Desktop.
Word Embedding Postprocessing: All but the top
"""
All-but-the-Top: Simple and Effective Postprocessing for Word Representations
Paper: https://arxiv.org/abs/1702.01417
Last Updated: Fri 15 Nov 2019 11:47:00 AM CET
**Prior version had serious issues, please excuse any inconveniences.**
"""
import numpy as np
from sklearn.decomposition import PCA
def all_but_the_top(v, D):
"""
Arguments:
:v: word vectors of shape (n_words, n_dimensions)
:D: number of principal components to subtract
"""
# 1. Subtract mean vector
v_tilde = v - np.mean(v, axis=0)
# 2. Compute the first `D` principal components
# on centered embedding vectors
u = PCA(n_components=D).fit(v_tilde).components_ # [D, emb_size]
# Subtract first `D` principal components
# [vocab_size, emb_size] @ [emb_size, D] @ [D, emb_size] -> [vocab_size, emb_size]
return v_tilde - (v @ u.T @ u)
@lgalke
Copy link
Author

lgalke commented Nov 15, 2019

@gojomo I just double checked with the paper itself and it should be v_tilde - (v @ u.T @ u) instead of v - (v_tilde @....

Applying PCA to centered / non-centered versions should not make a difference. The important thing is to subtract the mean from the embeddings to match the paper. v_tilde holds the centered version in this gist so now we return the embeddings - mean - first D principal components.

@s1998: I think this last point is also not considered in your implementation. In line 11 you do not subtract from the centered version. Am I missing something?

@s1998
Copy link

s1998 commented Nov 18, 2019

@lgalke You are correct, it should be mean centered embeddings that I need to subtract from. I'll fix that, thanks.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment