Kaggle Comp: NLP Classification


(James Requa) #1

Here’s an active Kaggle competition that could be a good opportunity to use fastai for NLP classification. The challenge is to classify the author of the text.


(Vikrant Behal) #2

Anyone willing to team up and participate?


(Rob H) #3

Cool. I have a question about training on a small NLP dataset.

Is there a way to increase dropout after a certain number of epochs?

I find that after training a language model on a different dataset, it overfits after awhile. If I increase dropout in the original architecture before any training, then it doesn’t converge optimally. Loss is still decreasing, so I want to train further, and don’t want to overfit.

Here’s what it looks like,

[ 0.       3.93047  3.74086]                               
...                         
[ 10.        3.50935   3.50461]                            
[ 11.        3.42893   3.45722]                             
[ 12.        3.32166   3.42504]

(Lucas Goulart Vazquez) #4

I sure am! I don’t have experience with NLP, but this should be a great learning experience!


(rachana) #5

Very good idea… Let us analyze text. How do you want to go about?


(Sanyam Bhutani) #6

Definitely count me in :innocent:


(Aditya) #7

Me too …


(Gerardo Garcia) #8

I would like to participate too.


(Ankit Goila) #9

Count me in! :slight_smile:


(Vikrant Behal) #10

Great! We’ve 6-7 members. How many of us have prior knowledge of NLP?


(Stephen Lizcano) #11

Hey! I have experience with NLP, can join up if you want. I have primarily used Keras and TF before.


(Setu) #12

count me in as well.


(Vikrant Behal) #13

Many teammates are joining. Let’s form a team 24 hours from now?


(Aditya) #14

A little bit of experience in NLP.

But really interested to learn more…


(Gaurav Desai) #15

I am also interested to join a team. But not much prior knowledge of NLP.


(Vishal Pandey) #16

I am interested too… But not have much experience with NLP…


(Vikrant Behal) #17

Created group! Pls. check your messages.


(James Birchfield) #18

Too late to play along?


(Rahul Pathak) #19

Why dont you guys create a open thread here and discuss. It can help others including me even if we are not participating :slight_smile:


(Vishal Pandey) #20

Not at all… You can join too…