Errata

When calculating the information gain (IG), the probabilities of the "star" and "dot" instances are written as 0.43 and 0.57, respectively. It should be 0.47 for "star" instances and 0.53 for "dot" instances, as correctly indicated on Figure 3-4.

As a result, the IG will be

IG = 0.99 - [0.47 x 0.39 + 0.53 x 0.79] = 0.39,

instead of 0.37.

Note from the Author or Editor:
Reader is correct. Correction is adopted. Changed by TF.

The last rule should say:

IF (Balance >= 50k) and (Age >= 45) THEN Class = No Write-Off

As it stands, the same condition repeats in line 2 and 3, which is incorrect.

Original: "The results are in Figure 3-17, with a table listing the exact values. As you can see, the first three variables—the house value, the number of leftover minutes, and the number of long calls per month—have a higher information gain than the rest."

Error: "...the number of leftover minutes.." is not the the number two among the highest information gain. The top three are (in descendent order):
1) House
2) Overage <<< This is the right one instead of LEFTOVER (which is No.4)
3) LONG_CALLS_PER_MONTH

Corrected: "The results are in Figure 3-17, with a table listing the exact values. As you can see, the first three variables—the house value, the overcharges per month, and the number of long calls per month—have a higher information gain than the rest."

Note from the Author or Editor:
Reader is correct. Reader's correction adopted.

pp. 84, Figure 4-3: Decision Boundary should be "Age = Balance x -1.5 + 60".

pp.85, Equation 4-1, it should be "1.0 x Age + 1.5 x Balance - 60" for both conditions.

pp. 86, the equation should be "f(x) = 60 - 1.0 x Age - 1.5 x Balance".

Note from the Author or Editor:
Errors 1 and 3 had already been fixed in version 2, but we hadn't caught the second one. Fixed for next version.

The equation reads as follows (in version 2):
(plus) if - 1.0 x Age - 1.5 x Balance + 60 > 0
(dot) if - 1.0 x Age - 1.5 x Balance + 60 <= 0

The errata that is already documented and approved removes the minus sign after the if.
(plus) if 1.0 x Age - 1.5 x Balance + 60 > 0
(dot) if 1.0 x Age - 1.5 x Balance + 60 <= 0

I believe this is not the correct fix. This makes the formula work, but it doesn't flow with the later formula on the page that reads f(x) = 60 - 1.0 x Age - 1.5 x Balance. Nor does it flow with the original formula on page 85:
Age = (-1.5) x Balance + 60
With Algebra you can rewrite the formula as follows:
0 = - Age -1.5 x Balance + 60

I believe the confusion is based around the negative slope. I suggest that you keep the minus sign (-) and flip the comparison operators to read as follows:
(plus) if - 1.0 x Age - 1.5 x Balance + 60 < 0
(dot) if - 1.0 x Age - 1.5 x Balance + 60 >= 0

Note from the Author or Editor:
Thanks for your close reading. I've adopted your suggested change for v.3, and I'll double check the chapter before the next release to make sure it's all consistent. -Tom

The paragrahp states that hinge loss incurs no penalty for an example that is not on the wrong side of the margin and that penalty only becomes positive when an example is on the wrong side of the boundary and beyond the margin.

Similar to the errate already entered about the hinge loss image on page 93, this is a same type of mistake. Hinge loss is also incurred for items on the right side of the decision boundary that are still within the margin.

Note from the Author or Editor:
Thanks. We have remade the diagram completely due to our (repeated) mistakes with it. -Tom

On line number 8, on page 146, it says "Whiskeys with Euclidean distance..." but when we check the results in the table below, the distances are calculated using the Jaccard distance and not the euclidean distance.

Note from the Author or Editor:
Reader is correct -- distance fn had been set to Jaccard. Distance function was changed to Euclidean and table regenerated. Results changed slightly but do not invalidate any points made in the text. New table is:

| Whiskey | Distance |
| Bunnahabhain | 0.00 |
| Glenglassaugh | 3.00 |
| Ardbeg | 3.16 |
| Bruichladdich | 3.16 |
| Tullibardine | 3.32 |
| Aultmore | 3.46 |

The last sentence says:

"So, the leftmost leaf corresponds to the segment of the population with around body and sherry nose, and the whiskeys in this segment are mostly from cluster J."

Instead, the sentence should apply to the RIGHTMOST leaf, as the tree is displayed in the figure, I believe.

Note from the Author or Editor:
Fixed.

Given that model A correctly identifies 30% of the negative examples and model B correctly identifies 30% of the positive examples, they achieve 65% accuracy not 80%. Indeed, the two tables should be:

(7.2) [500 350] (7.3) [150 0]
[ 0 150] [350 500]

Note from the Author or Editor:
Fixed.

I think "In this case, E is the same for all" should be "In this case, p(E) is the same for all"

Note from the Author or Editor:
You are correct, it should be p(E).
Fixed by author.
-Tom

The publication by Mackskassy et al. (2001) is not included in the bibliography section. The same applies to the publications by Mao et al. (2011) and Fawcett & Provost (1999).

Note from the Author or Editor:
All three fixed by author.