/sci/ - Science & Math » Thread #11126824

68KiB, 634x951, 4D44C4AA00000578-5848489-image-a-95_1529070502211.jpg

View Same Google iqdb SauceNAO

Anonymous Fri 08 Nov 03:43:29 2019 No.11126824 View Reply Original Report

Quoted By: >>11126835 >>11127593

If I'm trying to train a ml classifier but I need more data is okay to use shitty model or heuristics to scale down set of data to label by hand?
I get this will bias the set of data I hand label but it still seems like it could help.

Anonymous

Anonymous Fri 08 Nov 2019 03:47:45 No.11126835 Report

Quoted By: >>11126851

>>11126824
the absolute state of... i donno, the world?

if you have a small dataset then using neural networks is NOT AT ALL the right thing to do. if the dataset is so small you are entertaining the idea of increasing it by hand then just FUCKING MAKE A MODEL/CLASSIFIER YOURSELF!! or just fucking go classify the entire dataset by hand! what the fuck is wrong with people today

Anonymous

Anonymous Fri 08 Nov 2019 03:52:34 No.11126851 Report

Quoted By:

>>11126835
I'm doing transfer learning, I've seen people get good results with fairly small sets of data doing similar and it's text data so it's not so easy to make a "simple" classifier.
>if the dataset is so small you are entertaining the idea of increasing it by hand
It's not small, the set of labelled data is. I want to know if it will do more harm than good to increase labelled set by using biased sample etc.

Anonymous

Anonymous Fri 08 Nov 2019 12:16:40 No.11127593 Report

Quoted By:

>>11126824
>dirty grandpa..

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches