Generated by Bing Image Creator: Photo of dandelions and violets blooming all over a field
This post is following of the above post. In the previous post, I used R's 'tree' package. In this post, I will use R's 'glmnet' package for classification.
First, Ioad 'glmnet' package.
Since 'glmnet' require a matrix object, I will make a matices for use of 'glmnet'.
I also add squared, cubed and interaction variables.
Then, I will divide 'mtx' into two matrices, one is for training, the other is for testing.
Okay, let's use 'glmnet' package. First, I use cv.glmnet() function to find the best lambda.
I set alpha = 1, so it is LASSO regression.
Let's plot the result.
cvfit_lasso$lambda.min shows the best lambda.
I use the best lambda to train tne model.
Let's see beta coefficients.
v2, v3 and tv are not included in this estimate.
Let's predict using 'fit_lasso'.
Let's make a contingency table.
So, this lasso regression predict (124 + 39) / (124 + 39 + 66 + 41) = 60% only.
So, in this case, tree model is better than lasso regression.
That's it. Thank you!
Next post is
To read the post from the 1st post,
