Errata

My compter environment: Windows 7, python 2.7, libsvm-3.16
I didn't know whether python 3 or a different verion libsvm will produce the same problem.
---------------------------------------------------------------------------------------
When run this code on page 181, libSVM will produce an error, said
TypeError: [can only concatenate list (not "tuple") to list]
at the sentence
xi = [0] + xi
in libsvm function gen_svm_nodearray()
The above sentence requires xi at the right hand should also be a list.
However, the zip(x,y) at line 6 from bottom on page 181 will create
a list of TUPLE, not a list of LIST. This is the reason for the error.

This problem can be solved just as defining function "predict" as
follow:

def predict(x,y,model=m):
dsample =[]
for i in range(len(x)):
dsample = dsample + [[x[i],y[i]]]
return array(svm_predict([0]*len(x),dsample,model)[0])

P.S.
Just as the same as Mr. aikawa pointed out, the [-1,1] should be
[1,-1] at the line 4 from bottom.

My compter environment: Windows 7, python 2.7, libsvm-3.16
I didn't know whether python 3 or a different verion libsvm will produce the same problem.
---------------------------------------------------------------------------------------
When run this code on page 181, libSVM will produce an error, said
TypeError: [can only concatenate list (not "tuple") to list]
at the sentence
xi = [0] + xi
in libsvm function gen_svm_nodearray()
The above sentence requires xi at the right hand should also be a list.
However, the zip(x,y) at line 6 from bottom on page 181 will create
a list of TUPLE, not a list of LIST. This is the reason for the error.

This problem can be solved just as defining function "predict" as
follow:

def predict(x,y,model=m):
dsample =[]
for i in range(len(x)):
dsample = dsample + [[x[i],y[i]]]
return array(svm_predict([0]*len(x),dsample,model)[0])

P.S.
Just as the same as Mr. aikawa pointed out, the [-1,1] should be
[1,-1] at the line 4 from bottom.

The images of gradient magnitude are inverted in Figure 1-10 (d) and Figure 1-11 (bottom).

Figure 1-12 seems inverted, too.

Clarification to using the imlist as the list containing the filenames of images.
If imlist is obtained from a simple directory glob (e.g. imlist = glob.glob(r'/home/computer_vision/ukbench/*.jpg'),
then imlist must be sorted in order for this part of the vocabulary to work properly.

When continuing on page 158, the ordering of imlist must correspond properly with the database index or else you would a erroneous accuracy report.

Therefore, running imlist.sort() will solve the problem.

For N sample data with d dimensions, the dot operation
dot(x, dot(S,x.T))
will produce a N x N array.
If N is very large,

y = exp(-0.5*diag(dot(x,dot(S,x.T)))

will cause a memory error.
In fact in my computer environment (Windows 7, memory 2.00GB), the code in the book produced a memory error. I think somebody will be the same.

Since only the diagonal is needed, compute every elements in the array is not required.
The possibility of memory error can be reduced by replace sentence

y = exp(-0.5*diag(dot(x,dot(S,x.T)))

by following code

y = zeros((n,1))
for i in range(n):
for j in range(d):
y[i] += 1.0/v[j]*x[i,j]*x[i,j]
y[i] = exp(-0.5*y[i])

Somebody can write a more compact code.

Over.