Here is a pdf file of the corrected pages for the 4th printing
Here follows details of the corrections
Page 7, 4th line of How this Book is Organized :
It should be Chapters 3 and 4 instead of 2 and 4
Page 17, Figure 2.4: "... bigger green and red squares..."
Page 23, 2 lines below (2.24): 5000 should be 500
Page 26, line 1: N(0,1) should be N(0,\sigma^2); line 3: X should be bold
Page 26, before equation (2.28): Need to assume that E(X)=0
Page 43, line -3: "is nonsingular" -> "has full column rank"
Page 47, line 8: change p_1-p_2 to p_1-p_0
Page 47, equation (3.14): change v_j to v_j^\half
Page 52, line -2: (3.6) should be (3.27)
Page 64, equation (3.51): change s to t
Page 68, Algorithm 3.2, item 3: The expression for the coefficients is
incorrect, since it does not account for the normalization of the x_j
at each stage; no expression is given now.
Page 68, line -7: "that with" -> "that"
Page 91, first bullet: ^s on the \mu_k's on the right hand side
Page 96, line -2: \beta_1 and \beta_{K-1} should be ^T
Page 105, last line of section 4.4: "in inappropriately" -> "inappropriately"
Page 107, line -1: strike "each of"
Page 109, equation (4.45): remove the min in front of this equation; we
now preced this equation with a phrase ", to be minimized
w.r.t. $\beta$ and $\beta_0$,"; the + should be a - in front of the sum
Page 110, line -8: ... via \alpha_i>0 ...
Page 112, Ex 4.3, line +4: change B to B^T
Page 120, line + 19: there are FOUR interior knots, so they must be at
the 20, 40, 60 and 80th percentiles of x
Page 120, footnote: omit "s" in "defaults"
Page 123: made the vertical scales the same
Page 129, equation (5.14): ^t should be ^T
Page 135, Figure 5.9: Top left, legend should be EPE, not PSE
Page 136, line +8: ) missing in bias expression
Page 138, line -3: H should be h
Page 139, line -5: ... can "be" generalized ...
Page 147, Equation 5.56: subscript on x should be j
Page 152, line +1: \psi(x)=\phi(2x)-\phi(2x-1)
Page 154, line +10: remove "For example,"
Page 155, lines 5 and 7: "left" and "right" should be "upper" and
"lower"
Page 157, Ex 5.2: There are several changes to the text of this
question, best seen by viewing the new page 157 (pdf)
Page 158, equation (5.71): missing )
Page 160, equation (5.76): missing (x) on left
Page 161, equation (5.77): missing (x) on right; added some text in
opening paragraph to clarify the basis B_{i,m} for m smaller than M; line +13:
Exercise 5.2 explores these issues. Here is the corrected page
161 (pdf)
Page 163, line +7: change (N+4) x N to N x (N+4)
Page 186, equation (6.28)(first line):\xi_j
Page 191, Ex 6.2: for all j in {1,2, ...,k}
Page 198, Equation (7.13): replace "x" by "x_0" on RHS; in addition,
the statement is only true in an average sense. So we need to put an
E_{x_0} in front of each of the expressions
Page 204 equation (7.25): "loglik" should be divided by N
Page 206, equation (7.31):
A factor of \sigma^2_\epsilon is missing from the 2nd term within the
brackets.
Page 209, line +10: "additive" -> "multiplicative"
Page 210, Section 7.9, second paragraph: Some confusion bewteen p
dimensional x and 1 dimensional x. Initially x is p dimensional, but
then we specialize to p=1.
Page 212, line +8: included (SRM) in parenthesis
Page 223, Exercise 7.3b: Use absolute values on both sides
Page 223, Exercise 7.4: Include an E_y on the RHS of first equation,
and a ^ on f in the second.
Page 230, line -3: "set of all $\theta_0$"
Page 232, equation (8.23): dZ should be d\theta
Page 233, equations (8.27) and (8.28): 1/\tau should be \sigma^2/\tau
in both the means and variances; line -5: \theta' should be \beta'
Page 235, equation (8.30): the mean is not z, but z/(1+1/\tau)
Page 238-239, equation (8.40): There is a piece missing from the full
log-likelihood; add +\sum_{i=1}^N [(1-\Delta_i)\log\pi + \Delta_i
\log(1-\pi)]
Page 244, before Equation (8.50): Change . to :
Page 250, line -5: "Viewed in this way"
Page 262, line 6: "outlined in Section 4.4"
Page 262, line-10: replace PEOPLE/yee by ~yee
Page 265, line -4: "significantly" -> "significant"; line -2:
"indicate" -> "indicates"
Page 271, equation (9.17): minus sign missing for cross entropy
Page 272, line -15: Subscripts of second L should be k'k; line +9 "in
terms of squared"
Page 275, line -4: overall error rate showN...
Page 276, table 9.3: Overall error is 9.3%
Page 277, line -17: the numerator of the ratio shoul be L_{01}
Page 287, line -3: change crl-tot to CAPTOT
Page 295, Section 9.7, line 15: "M is a reasonable fraction of N".
Page 301, line +8 (not counting box for Algorithm 10.1):
It should be line 2d (instead of line 2c).
Page 302, line +5: "by almost a factor of four" rather than "by more
than a factor of 8".
Page 305, unumbered equation before (10.9): G(x_i)
Page 306, (10.15) and above: all G should be G_m
Page 307, line -6: exponent is garbled
Page 312, line +5: f_m
Page 314, line -12: "The left panel of" should be removed.
Page 314, line -9: "spam" should be changed to "3d".
Page 316, line+3: "for the full MART model (with J=6 terminal node trees)"
Page 320: double use of f_m as the function as well as the
updates. Replace some f_m by h_m; see corrected page 320 (pdf)
Page 321, line -11: As noted in Section 10.9 ...
Page 326, (10.40): Take the hat of \gamma; line -6: Figure 10.12.1 does not exist; should be Figure
10.11; line -3: benefits ... are
Page 332, line -13: change x_j to X_\ell
Page 334, line+8: should be \tilde{f}_S and \bar{f}_S
Page 335, line +11: changed STATLIB to StatLib
Page 338, lower caption: "...interaction..."
Page 349, equation (11.3): extra )
Page 363, line+18: 0,1,2,...,9
Page 373, line +7: change ||C|| to C
Page 374, line+9 (one below equation 12.9): change "maximize" to
"minimize".
Equation (12.8): errors in the condition; remove second x_i and change
b to \beta_0
Page 375, first paragraph: many "hats" ^ are missing from the \alpha_i
and \xi_i
Page 377, line -2: we solve for y_if(x_i)=1
Page 381, table 12.1: remove ( before 4.51; change F to f in table
headings
Page 382, line -9: need f''_j^2 in the penalties
Page 382, line -4: any convex loss
Page 384, lines 11-12: replace "L_1" to "piecewise-linear"; change
"smaller" to "greater"
Page 385, Table 12.2: The Bayes error is incorrect. The correct value is
0.029 (based on a simple Chi-squared computation). Our results for
MARS were incorrect; we had inadvertantly used an additive MARS model,
rather than the full-interaction model. This has been corrected. We redid the
simulations in table 12.2 using 50 rather than 5 simulations, to avoid
any uncertainties due to sampling fluctuations. The
conclusions remain the same, although the statement at the bottom of
the page about the additive model and MARS being close to the Bayes
error is no longer applicable, and has been modified. We hence include
here the modified version of page 385
Page 386, equation (12.35): change t to r
Page 387, lines 7 and 14: "hats" on the \alpha_i
Page 388, line 8: ... this has THE same form
Page 392, line +1: remove "a" from "a restrictions"
Page 396, line -8:
Instead of "the bibliographic notes" it should be the Computational Considerations section on p. 405.
Page 397, equation (12.56) and one below: \Omega should be \lambda\Omega
Page 401, line +6:
Instead of "the bibliographic notes" it should be the Computational
Considerations on p. 405.
Page 403: In the caption "kower" should be "lower".
Page 414, algorithm 13.1, line 3: decreasing ... towards zero.
Page 427, line below equation (13.7): v_d should be v_p
Page 438, line+1: Change "From Bayes theorem" to "Conditioning"
Page 439, line -13: \nu_1
Page 440, equation (14.2): comma before and after equation
Page 444, line -3: The first and fifth levels of occupation are
represented in the rules diagram 14.2 (lower)
Page 449, line -1: missing )
Page 452: in two places, "household status" should change to
"householder status"
Page 453, line 15: "apartment" should be "type of home"
Page 456, 2 lines below (14.22): Change "=" to "\propto"
Page 462, Equation (14.31): Need an N_k between the two summation
signs, where N_k=\sum_{i=1}^N I(C(i)=k). Also in C* equation, and
(14.33)
Page 463, lines 12 and 14: E- and M- should be interchanged
Page 468, line 10: generalizationS; line 13: repeated "of the"
Page 472, lines 14 & 15: green and blue, not solid and broken; line 7:
curves iS largest
Page 473, Figure 14.11: The right-hand plot needs to be translated
vertically so it starts at zero, to correspond to the modified version
of the left-hand plot.
Page 475, line 14: replace d_{ii'} by C_{ii'}
Page 477, line 1: "if the all of the" - strike the first "the"; line
-4: g_k should be p_k
Page 482, line 9: the points are (0,1,0), (0,0,1) and (1,0,0)
Page 495, Equation (14.59): subscripts a_{*p} should be a_{*q}
Page 496, line -1: with with
Page 500, middle: "and normalized linear ..." -> "a normalized linear ..."
Pages 507-508: Minor typos in (c). Also changed x to X throughout.
Page 512: "to appear" changed to updated reference
Page 513, reference Friedman (1994a)" "sneighbour" changed to
"neighbor"; "to appear" updated
Page 513: in Friedman et al (1977) algorithm is incorrectly spelled
Page 514: The page numbers for the Friedman, Hastie, and Tibshirani
(2000) reference should be 337-407 (instead of 337-307).
Page 521: "to appear" changed to updated reference
Page 533, last index entry: "Within-class"
In most of the exercise sections the right headline was incorrect;
this has been changed throughout.