Unofficial Lecture 9 Notes

[RamboFast]

Hi

In the previous post you say this:

@timlee

p = x[y==1].sum(0)+1 - Number of times it shows in positive corpus

p = x[y==0].sum(0)+1 - Number of times it shows in negative corpus

r = np.log((p/p.sum())/(q/q.sum())) - summing logs is a lot easier to do, otherwise you might run into float errors

b = np.log(len(p)/len(q)) - this is our naive bays

x=trn_term_doc
y=trn_y

p = x[y==1].sum(0)+1
q = x[y==0].sum(0)+1
r = np.log((p/p.sum())/(q/q.sum()))
b = np.log(len(p)/len(q))

Here’s our predictions from Naive Bayes ~ 80% positive

pre_preds = val_term_doc @ r.T + b
preds = pre_preds.T>0
(preds==val_y).mean()

I see how you apply the log-count-ratio to a dataset which has 2 labels (positive and negative). But I do not understand how to apply log-count-ratio to data which has more than 2 labels. Maybe someone can explain this to me?

I searched the complete internet and I could not find any fastai example that uses Naive Bayes with a dataset which contains multiple labels. Maybe you have some links for me?

→ I have the same question asked here but with much more detail. Unfortunatlly I never got any response Maybe you can help with that.