Calculate hi value in Chapter 5 and plot margin in chapter9 about statlearning-notebooks

[sujitpal]

Thanks for the kind words.

For the first issue (Chapter 5), the h(i) is calculated by Equation 3.37 which goes something like this (my mathjax foo is non-existent unfortunately):
h(i) = (1/n) + (x(i) - xmean)^2 / sum((x(i) - xmean)^2)
which I attempted to replicate faithfully. Can you tell me what expression you meant?

For Chapter 9, I think the expression that SVR uses for a straight line is y = mx - c (rather than the more common y = mx + c). That is why c = reg.intercept_ / reg.coef_[0][1]. The SVM tutorial also uses the same thing, notice: yy = a * xx - (clf.intercept_[0]) / w[1]

[JuDa-hku]

Thank you for your help.
For the first issue, you just miss the power 2 in the code.

For the second one, SVR uses y = mx-c which is different with the SVM (y = mx+c). By using SVM, we may have a unified way to write the code.

ys = [m * x + c for x in xs]
plt.plot(xs, ys, color='black', linewidth=2.5)
# support margins
sup = reg.support_vectors_[0]
ys_down = [m * x + (sup[1] - m * sup[0]) for x in xs]
sup = reg.support_vectors_[-1]
ys_up = [m * x + (sup[1] - m * sup[0]) for x in xs]

[sujitpal]

Ah, I see the problem in Chapter 5 now, thanks for catching, fixed. Also on Chapter 9, we can do the unification by fixing ys = mx - c (I had an error there) and subtracting the support vectors in both cases similar to your example above. Also realized that the paths to the data files were incorrect following the move from the old repository to the new one, fixed that too. Closing this, thanks again for reporting.