TODO (2 week deadline) - Githubissues

bakerwho / weboftruth

Torch + bash scripts for training a web of truth (literal objective incontrovertible truth, because it obviously exists)

5 stars 1 forks source link

TODO (2 week deadline) #2

Open bakerwho opened 4 years ago

bakerwho commented 4 years ago

Build Python modules to:

[x] negatively sample facts (each fact is 3 numbers) from SVO dataset (n_neg_facts, sampling_params?)
[x] compile datasets using specified numbers/ratios of true & false facts
[x] adapt torch-kge or other code to run on our datasets
[ ] build  visualizations (PCA) of embeddings 
[ ] build Word2Vec/glove embeddings with BOW for comparison
[ ] build linear classifier + MLP to classify true/false based on embeddings

For June 1 Monday:

[x] Create a reduced dataset (50-50 T/F) and get it pre-processed enough to run TransE, TransH, HolE
[ ] MAYBE start building visualization code?

Spacy-NOW & WikiHumans:

[ ] List of named entities, or (better) (head, , tail) triples.

bakerwho commented 4 years ago

For June 1 Monday:

[x] Create a reduced dataset (50-50 T/F) and get it pre-processed enough to run TransE, TransH, HolE
[ ] MAYBE start building visualization code?