Exercise doubt

[manohar1989]

Hi Daniel,

I am trying to complete the exercises in skimlit project
You asked to test by adding 1_of_11 feature rather than 2 different features. My doubt is, can i go with one hot encoding here, it looks very odd to go that way since number of possibilities are a little more than depth 15 and 20 like earlier?

Shall I go with use since this is also english?

PS.I tried including 1 of 11 in the train sentence itself like (in both token and char embedings):
'1 of 9 : it is not clear whether these patients would benefit from antifungal treatment .',
this got 88% accuracy right now, it looks better than different features.

[mrdbourke]

Hey Manohar,

You asked to test by adding 1_of_11 feature rather than 2 different features. My doubt is, can i go with one hot encoding here, it looks very odd to go that way since number of possibilities are a little more than depth 15 and 20 like earlier?
Shall I go with use since this is also english?

I'd try it out and see how it goes, though it seems you achieved better results using the full embedding style:

PS.I tried including 1 of 11 in the train sentence itself like (in both token and char embedings):
'1 of 9 : it is not clear whether these patients would benefit from antifungal treatment .',
this got 88% accuracy right now, it looks better than different features.

This is really cool to see. That's a great boost from what it was before.

I'd keep going the way you are experimenting.

If you find a better way, I'd love to hear about it, I could even update the exercises to reflect your methods.

[manohar1989]

I tried a score aproach as well.
I think line number and total_lines are related hence instead of taking 2 inputs i did "score = line_number/total_lines"
This also is giving same result almost 87% on val data and 86% on test data.
something like:

[mrdbourke]

This looks really cool. Thank you for sharing.

So many different ways to engineer features.