66, near top: U is not square, so has orthonormal columns, but is not orthogonal

85, line above (3.70): change "it's generalized inverse" to "a generalized inverse". Also note that in (3.71) P_m is not an orthogonal projection operator

89, line -5: should be N > p for the first case

161, paragraph above section 5.6: it says that "The EPE and CV curves have a similar shape, but the entire CV curve is above the EPE curve." Change "above" to "below"

213, (6.31): denominator index should be different from i

301, line -10: Test error for GAM should be 5.5%, not 5.3%

326, line -6: Test error for GAM should be 5.5%, not 5.3%

574, (14,104): need a constraint such as Y'Y=I ow trivial solution Y=0

579, line -2: bold L should be a bold I

585, Ex 14.24(b): need to assume the sum of all the x_ij is 1 (or change the definition of c)

654, line -3: "overall in this example"

Thanks to

Sheridan Grant

Allan Hansen

Wen Jie Koh

Dietmar Maringer

Scott Powers

Shankar Prawesh

Jonathan Taylor

Ya Wei

Christopher Yeh

Alexander Young

Arif Zaman

310, second paragraph, line +3. Then the "expected" training error .... Its a random rule

428, footnote: norm is incorrectly formatted

536, last paragraph: It should read: "For each of these first two principal components u_{i1}d_1 and u_{i2}d_2"

577, last paragraph: It should read "...than those from the..."

583, Exercise 14.19: The equation cited should be 14.91 instead of 14.92

585, Exercise 14.23b: The variable u_{i,j} should be replaced by x_{i,j}

588-589, paragraph 1 (clarification): "This means the bias of bagged trees is the same as that of individual (bootstrap) trees"

597 (clarification). Add "See Exercise 15.3" to caption

603: Change indicator notation from 1 to I

Thanks to

Camilo Lamus

Thomas Lengauer

Arif Zaman

360, Equation (10.38) has errors. In the gradient, there should not be "m"s, the last 1 should

be a K, and the gradient sould be evaluated at bold f_{m-1}

478, Should be squared radius between 22.4 and 40

568, "Step 2(b) in Algorithm 14.3 requires optimizing (14.91) with respect to A...", not (14.92)

Thanks to

Zachary Grinspan

Dennis Leung

Hrvoje Stojic

Syndey Wong

A Japanese translation of the ESL appeared in 2014. One of the editors, Yoshihisa Ijiri, kindly supplied us with a long list of candidate additional japanese_errata.pdf. Most of these had not be caught, and all but a few were real errata, and these have been corrected as well.

56, f(x) should be a vector

91, equation (3.81) changed

97, Ex 3.7, added some detail on prior

105, bullet - this is true whether or not the fitted values sum to 1 (Laurell Stell)

109, eq 4.11 changed to have the more traditional formula

133, the Wolfe dual (4.52) should also include equality constraint of Eq. (4.51).

135-136 ex 4.2 changed. There were mistakes before, and misleading def for Sigma_B

159, Fig 5.9 - the figure was incorrect- variances were plotted rather than SE

224, line 11 - the transpose needs to be an inverse for h(x_0)

236, (7.45) 2 missing in denom of third expression on right

"Huffmann coding" should have only one "n".

357, eq 10.32 should have a 1/2 on right hand side

358, below (10.34), extra ) and should be ^T

359, 6 lines into Section 10.2.2 - need a close brace and a transpose

386, Exercise 10.8(b): The denominator in the last line should be exponentiated.

439, Bullet point 4 from the top is erroneous. The truth is on page 111: LDA or QDA was among

the top three classifiers for 10 of the 22 datasets in the STATLOG project. LDA was among the

top three classifiers for 7 of the 22 datasets.

457, Expression (12.66) should read max \theta^T Y^T S Y\theta.

520, figure labels changed

541, Item 1 should needs a transpose on the right X_\ell

559, (14.78) change s_ to a_, make subscripts of S all ell

647, The first line of Exercise 17.5 should refer to Section 17.3.1, not Section 17.4.1.

648, The reference (17.28) should not be repeated in Exercise 17.11.

653, line 10 delete "absolute"

661, the methods of Section 18.3 ....

694, eq. (18.55): Left absolute value bracket for last term has to be lifted to the numerator.

698, line 6: Proofs of result (18.53), not (18.20)

717, last line - should be "Relaxed Lasso"

Thanks to

David Epstein

Andre Ferrari

Thomas Lengauer

Steven Lyuee

Edward Malthouse

Kajsa Mollersen

Scott Powers

Ludi Rehak

Laurell Stell

Jonathan Taylor

Can Yang

Peng Yu

Qin Zeng

23, Curse picture - change d to p

61, lcweight -> lweight

71, changed first entry in table 3.4

72, missing parenthesis

74, f_k needs a hat

89, Dantzig selector change s to t

98, put 1/2 in exercise 3.27

126, l-6 "indentify"

127, Friedman reference changed

165, changed thin-plate spline def to simpler equivalent form

275, extra . in EM formula

279, word missing

297, "," after V

321, extra ","

339, punctuation

348, extra space removed

353, space included

360, missing "to"

426, missing period

428, \alpha_0 should be \beta_0

441, funny braces changed in line 5

489, greater "than"

541, A should be bold

544, grammar change

578, matrix L

579, Bibliography typset wrong near end

581, R_\ell the R should be bold

588, space added and Case change

601, "depends" line 7

612, Figure 16.2 added a parenthesis about relative to error of constant model

618, capital M

642, line 15 "maximization"

649, extra brace

657, 2 lines below 18.11 subscript wrong

661, put 1/2 in 18.18

661, reference changed

668-669, reference Leslie changed

679, (2) and (3) to (2a) and (2b)

684-685, "Treatment" vs "Sensitivity"; formula for se changed

688, punctuation in figure changed

708-709, reference fixed

716, reference Leslie et al corrected

These errors were found from January 2010 to December 2010 by

Moritz Gschwandtner

Iain Johnstone

Mehrdad Jafarr Mamaghani

Nelson Ray

Kinh Tieu

Enrique Tobis

Can Yang

Tian Zheng

A big thank you to Thomas Lengauer and Stephen Milborrow for their extensive errata reports.

page 103, line -3: put a ^T on x

page 104, equation (4.5): as in line 103. Since y_i is a column vector (length K), care is needed in constructing the fitted values

page 215, last line: colors were incorrect

page 216, figure 6.17: the fitted mixture was plotted incorrectly, using the variances as standard deviations!

page 257, line 2 of "Bibliographic Notes": Allen reference changed from 1977 to 1974.

page 283. line -8: eleven trees are shown.

page 349, equation (10.23): error fixed in second line

page 504, 2 lines below (14.22): changed "variables" to "observations". In fact, the observation vectors are standardized as if they were variables.

page 570-571, lines -1 and +1: First edition had a square-root of S_M, which is removed here. Text still referred to this, so was altered.

page 584, line +1: since "it" is easy.

page 587, line+1: parens around (section 8.7).

page 588, line 2(a) in Algorithm 16.1: upside-down hat on \alpha.

page 649, line -1: Var(epsilon) is 1; needed sigma^2 instead.

page 658, fourth paragraph: Line 3 of Table 18.1, not Line 2.

page 661, equation (18.19): beta has two indices

page 716, Leslie reference: "protein", not "pretein"

page 686, line above (18.42): would like "to" test...

page 700: see page 257.

These errors were found from July 2009 to December 2009 by

Theodore Alexandrov

Jacob Bien

Francesco Dinuzzo

David Goldberg

Diego Kuonen

Yan Li

Phielipp Mariano

Stephen Milborrow

Changyi Park

Patrick Perry

Jason Rennie

Daniela Witten

page 26, 27 " colors are orange and blue in figure 2.9

page 245, line -5 "left" -> "top"

page 509, section 14.3.6. There are errors in this subsection relating k-means clustering to the total distance measure 14.28. See the explanation here.

page 571, line +1 "that" -. "than"

page 687, line-2. Reference to figure should be "Figure 18.19 shows..."

These errors were found from February 2009 to July 2009 by

Dave DeBarr

Thomas Lengauer

Ilana Belitskaya-Levy

Kinh Tieu

Daniella Witten