[Bmi212_c-news] Notes from yesterday's meeting with CSFoo

[Daniel Newburger]

Hi folks,

Sorry I forgot to send this last night.

-Dan

Lasso linear regression (L1) to select features while making the model
>
imposes a sparsity penalty
>
>
> Elastic net (L1 & L2 at the same time)
>

> 1.  reduce features (maybe to 10,000), then use logistic reg. or SVM
>

> -can use softmax or multiclass SVM across all 8 classes (7 diseases and
> control)
>
-for lasso (or another algorithm that can select features), need an
> algorithm that does not require all data be in memory
>

> SO:
>

> Our pipeline:
>
    1.  Select SNPs feature vector using p-value and odds-ratio
>
    2.  use softmax or SVM to learn across all disease classes
>
    3.  repeat for binomial (early vs. late), for age of onset on disease.
> What about regression?
>

>
Learning theory = method of analyzing your learning method to see if have
> enough data
>

> libraries:
>
svmperf
>
svmlight
>
libsum
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://simtk.org/pipermail/bmi212_c-news/attachments/20091017/283f804b/attachment.html