Question $2[20$ pts$]$

You are given the stroke classification dataset in 'stroke.csv$\text{'}.$

The continuous features are: age, avg$\_$glucose$\_$level, and bmi. The rest are

discrete features. The label stroke is binary $\{0\frac{,}{b}ar(1)\}.$

NOTE: for each of the following parts show the relevant code snippet and print results.

a$)$ Do the following in the list order:

Read in the dataset. We will treat the entire dataset as training.

Take the discrete features and one$-$hot encode them.

Take the continue features and standard scale them.

Construct X$\_$train by hstacking the two feature matrices above. Also, setup y$\_$train.

Print the shapes to confirm you got this part correctly. The below is my

implementation's shapes.

$($X$\_$train, y$\_$train$)$ shape: $,$

b$)$ Code the function

def logreg$\_$gradient$\_$descent$($X$,$ y$,$ learning$\_$rate$=0\mathrm{.}01,$ max$\_$iters$=1000)$

that implements logistic regression full$-$batch gradient descent. The function returns two

items: the fitted weight vector $($w$\_$hat$)$ and list of iteration log losses $($loglosses$).$ Do

it as follows:

First set np$.$random.seed$(11),$ then initialize w$\_$hat as a sample from a normal

distribution: np$.$random.normal$($loc$=0\mathrm{.}0,$ scale$=1,$ size$=$X$.$shape$[1])$

Setup an empty list for loglosses

Loop for max$\_$iters to do:

Compute prediction probabilities on X using current w$\_$hat.

With predictions, you can compute current iteration log$\_$loss$($y$,$ y$\_$preds$).$

Append this iteration log loss to loglosses list.

Update w$\_$hat using logistic regression gradient update.

After loop return w$\_$hat, loglosses

Just show snippet for your function implementation.