Rohit Ghumare 7e30e4d651 feat(phase-05/05): sentiment analysis ...

Multinomial Naive Bayes from scratch with Laplace smoothing, logistic
regression with L2 regularization, and negation scoping (prefix tokens
after not/no/never with NOT_ until the next punctuation). Demonstrates
the classical canonical text-classification stack.

Pedagogically important: the demo output intentionally shows two
misclassifications on a tiny training set. The reader sees realistic
small-corpus Naive Bayes failure modes tied directly to what bigrams
and larger corpora would fix. This is the teaching point, not a bug.

Ship artifact: sentiment-baseline prompt with explicit refusals for
dropping stopwords on sentiment tasks and for reporting accuracy alone
on imbalanced data.

~75 minutes. Prerequisites phase 2/14 Naive Bayes.

2026-04-22 15:46:12 +01:00

..

sentiment.svg

feat(phase-05/05): sentiment analysis

2026-04-22 15:46:12 +01:00