w******8
发帖数: 59 | 1 I am using Lasso on analyzing gene data. I use 5-CV to find the tuning
parameter lamda but each time I am running it again, a different tuning
parameter will be selected and hence a different final model will be chose.
I know this is from randomness of partitioning the training set. What shall
I do? Shall I just run once and select whatever model produced? This seems
so random and not stable. Thanks for your input! |
a***g
发帖数: 2761 | |
s*r
发帖数: 2757 | 3 把那个 trace画出来，肉眼挑个lambda。
.
shall
【在 w******8 的大作中提到】
: I am using Lasso on analyzing gene data. I use 5-CV to find the tuning
: parameter lamda but each time I am running it again, a different tuning
: parameter will be selected and hence a different final model will be chose.
: I know this is from randomness of partitioning the training set. What shall
: I do? Shall I just run once and select whatever model produced? This seems
: so random and not stable. Thanks for your input!
|
l******0
发帖数: 73 | 4 It seems your data can't be fit to only one model well. How many
observations do you have? Are they from the same strata? |
