Errata

This comment relates to Second Edition Second Release:

On page 85, the mnist Bunch is constructed with:
mnist = fetch_openml('mnist_784', version=1)

On page 86, the data and target sets are generated with:
X, y = mnist["data"], mnist["target"]

With Scikit v1.2.0, X and y are not ndarrays (like implied by the text), but Pandas DataFrame and Series. This makes the image generation code on page 86 fail. Or one retypes X and y to ndarrays or one changes the mnist construction to:
mnist = fetch_openml('mnist_784', version=1, as_frame=False).

With the latter, X and y are generated as ndarrays.

In this equation, the bias vector is shown as `b`which has size n-neurons. However, this calculation is being made for a mini-batch. The other terms in the summation will have a dimension of m x n-neurons, similar to output Yt. Also, the previous equation implies that b is a column vector. So shouldn't `b` be changed to matrix B of same dimension where each row = b-transposed?

It is not clear how Rand % error and Strat % error columns are calculated?
For example, in a 1st row Overall - Random = 0.039826 - 0.040213 = 0.000387, not 0.973236 % as shown in the table

in the test it shows the following code with printed output, while the output needs .toarray() to be shown:

cat_encoder.transform(df_test) should be cat_encoder.transform(df_test).toarray()


cat_encoder.transform(df_test_unknown) should be cat_encoder.transform(df_test_unknown).toarray()

On the following line, it throw out an error
>> df_output = pd.DataFrame(cat_encoder.transform(df_test_unknown),
... columns=cat_encoder.get_feature_names_out(),
... index=df_test_unknown.index)


changing to the following, problem is fixed.
>> df_output = pd.DataFrame(cat_encoder.transform(df_test_unknown).toarray(),
... columns=cat_encoder.get_feature_names_out(),
... index=df_test_unknown.index)

In Figure 9-10 the axis of the plots in the upper row seems not to be aligned with the axis of the lower row. I assume the same data as in Figure 9-9 is used: k=4 (upper right) seems to be worse than k=3 (upper left).

'x' is not defined, it should be 'a' Considering the context. So the right way is to
change 'x.initializer.run(session=sess)' to 'a.initializer.run(session=sess)'

I believe the value of the Lagrangian is inverse to the one shown.

Actual:
Lagrangian(w, b, alpha) = 1/2 SuperSum - sum(alpha)

Expected:
Lagrangian(w, b, alpha) = -1/2 SuperSum + sum(alpha)

It probably doesn't matter much, since all we're gonna do with this is (1) calculate the partial derivatives with regards to alpha, and (2) put these equal to zero (not shown), so the sign plays no role.