Grigory Sapunov (10) [Avatar] Offline
#1
p.5
In this new coordinate system, the coordinates of our points can be said to be a new
"representation" of our data. And it’s a good one! With this representation, the
black/white classification problem can be expressed as a simple rule: black points are
such that x 0 or "white points are such that x < 0". Our new representation basically
solves the classification problem.

Should be:
black points are such that x > 0

All machine learning algorithms consist of automatically finding such
transformations that turn data into more useful representations for a given task. These
operations could sometimes be coordinate changes, as we just saw, or could be linear
projections (which may destroy information), translations, non-linear operations (such as
select all points such that x 0), etc.

Should be:
select all points such that x > 0
Grigory Sapunov (10) [Avatar] Offline
#2
p.20
The deep learning hardware
startup Nervana Systems was acquired by Intel in 2016 for over $400.

Should be:
for over $400M
Grigory Sapunov (10) [Avatar] Offline
#3
p.51
By the end of this chapter, you will already be able to use neural networks to solve
simple machine problems such as classification or regression over vector data.

Should be:
simple machine learning problems such as classification or regression over vector data.
Grigory Sapunov (10) [Avatar] Offline
#4
p.91
Like hold-out validation, this method doesn’t exempt you from using a distinct validation set for model calibration.


Do we really need a distinct validation set here?

Grigory Sapunov (10) [Avatar] Offline
#5
p.120
Listing 5.9
# Directory with our validation cat pictures
test_cats_dir = os.path.join(test_dir, 'cats')
os.mkdir(test_cats_dir)
# Directory with our validation dog pictures
test_dogs_dir = os.path.join(test_dir, 'dogs')
os.mkdir(test_dogs_dir)

Should be:
# Directory with our test cat pictures
# Directory with our test dog pictures
Grigory Sapunov (10) [Avatar] Offline
#6
p.168

In Listing 6.2 there should be a limit up to maxlen symbols something like in Listing 6.1:

for i, sample in enumerate(samples):
for j, character in list(enumerate(sample))[:max_length]:
index = token_index.get(character)
results[i, j, index] = 1.
Grigory Sapunov (10) [Avatar] Offline
#7
p.220

This error tells you, in essence, that Keras was not able to reach input_1 from the provided output tensor.


Is it correct? It looks like Keras was not able to reach input_1 from the provided input tensor, not the output one.
Grigory Sapunov (10) [Avatar] Offline
#8
p.268
VGG19 is a simple variant of the VGG16 network we introduced in Chapter 5, with three more convolutional layers

It looks strange that the more complex network is called as a simple variant. I understood it as the VGG19 is simpler than VGG16.
Grigory Sapunov (10) [Avatar] Offline
#9
p.286
Listing 8.31
# Produce a 32x32 1-channel feature map
x = layers.Conv2D(channels, 7, activation='tanh', padding='same')(x)

Should be:
# Produce a 32x32 3-channel feature map
Grigory Sapunov (10) [Avatar] Offline
#10
p.289
Figure 8.18 Play the discriminator: in each row, two images were dreamed up by our GAN,
and one image comes from the training set. Can you tell them apart? (answers - real
images: middle, top, bottom, middle)

Should be:
In each column
svenproppert (5) [Avatar] Offline
#11
Figure 6.28 is the same as figure 6.26 although indicated otherwise.