search

daxiongshu / Grasp-and-Lift

4 stars 2 forks source link

code updated #1

Closed daxiongshu closed 9 years ago

daxiongshu commented 9 years ago

Hi, I just update the code for the submission stack1sub.csv

It seems that the normal base model + normal stacking may just work fine.

SudalaiRajkumar commented 9 years ago

Thanks Jiwei. I have uploaded a cross_val.py file into SRK/ToRun/. I have merged both your and my feature creation idea in this code and then ran LDA on the whole set of features with three diff subsampling.

Can you please run this script in your machine and let me know the cv scores. Sorry, my 8gig machine cant handle the array which this code will produce. :(

Here the CV methodology is leave one series out at a time. I have seen the CV almost same as LB so far using this CV method. Thank you.

SudalaiRajkumar commented 9 years ago

Btw, how can I call you? Jiwei or Carl or rcarson or daxiongshu? Which one do you like :) I guess Jiwei Liu is the original name :)

SudalaiRajkumar commented 9 years ago

Please remove few electrodes from "low band pass feature creation" if the size of the resulting array is too big to handle for the system. Thank you.

daxiongshu commented 9 years ago

You could call me carl : )

Can I run the validation in my old way, which is using series 7 and 8 of each subject to validate? I think It will be much faster to test. The score is consistently lower LB. I think run the leave one serious out cv later : )

Best.

SudalaiRajkumar commented 9 years ago

Great sure. Thanks Carl :)