Ch05 lab bug produces incorrect bootstrap estimate SE(B_0)

[johnlevidy]

ISLP_labs/Ch05-resample-lab.ipynb

Line 860 in 4854ffe

"def boot_SE(func,\n",

This function as written has a bug. Non positional indexing on data frames will result in behavior different than positional sampling. When I run the notebook as is, I get

intercept 0.717654 horsepower 0.006041

To fix this, we can just write:

def boot_SE_fixed(func, D, n=None, B=1000, seed=0):
    rng = np.random.default_rng(seed)
    n = len(D) if n is None else n

    first_, second_ = 0, 0

    for _ in range(B):
        positions = rng.choice(
            len(D),
            size=n,
            replace=True,
        )
        value = func(D, positions)
        first_ += value
        second_ += value**2

    return np.sqrt(second_ / B - (first_ / B)**2)

( here we rng.choice on the length of D )

And

def boot_OLS_fixed(model_matrix, response, D, idx):
    D_ = D.iloc[idx]
    Y_ = D_[response]
    X_ = clone(model_matrix).fit_transform(D_)
    return sm.OLS(Y_, X_).fit().params

here we use iloc instead of loc.

hp_func_fixed = partial(boot_OLS_fixed, MS(['horsepower']), 'mpg')

Finally, we correctly wrap it and:

hp_se = boot_SE_fixed(hp_func_fixed,
                Auto,
                B=1000,
                seed=0)
hp_se

produces the desired results, making it consistent with the markdown below.