Add an infobox for leaky relu

when defining the covariance matrix, it should be should be 1/(n-1) rather than 1/n

Chapter 17: Synthesizing New Data with Generative Adversarial Networks
Chapter 18: Using Reinforcement Learning for Decision Making in Complex Environments

Ch 18: Using Reinforcement Learning for Decision Making in Complex Environments

High resolution image synthesis and semantic manipulation with conditional GANs. https://arxiv.org/abs/1711.11585

17. Generative Adversarial Networks for Synthesizing New Data
18. Reinforcement Learning for Decision Making in Complex Environments

Pg. 84, second to last line:

... using the `logical_or` function ....

should probably be

... using the `logical_xor` function ....

cartpole/main.py

gridworld/gridworld_env.py

When we get updated proofs, we should check page 19 again

1

2

3

4

5

6

7

8

9

10

11

12

13

Hey @vmirly I noticed that your notebooks are named ch14-notebook.ipynb etc. In contrast, all the previous chapters are named ch12.ipynb etc. Which convention should we adopt? I don't have a strong preference, but we should be consistent.

Original:

plt.text(10.2, -0.5,
         s='OD280/OD315 of diluted wines',
         ha='center', va='center', fontsize=12)

plt.tight_layout()
#plt.savefig('images/07_11.png', dpi=300, bbox_inches='tight')
plt.show()

Change:

plt.tight_layout()

plt.text(0, -0.2,
         s='OD280/OD315 of diluted wines',
         ha='center', va='center', fontsize=12,
         transform=axarr[1].transAxes)

Change

zipped_mnist = [f for f in os.listdir('./') if f.endswith('ubyte.gz')]

to

zipped_mnist = [f for f in os.listdir() if f.endswith('ubyte.gz')]

for windows compatibility

In gridworld.py, line 181, we have. Please fix that.

(3nd edition) -> (3rd edition)

4. Building Good Training Sets – Data Preprocessing -> 4. Building Good Training Datasets – Data Preprocessing

1

2

3

4

5

6

7

Part 2

8

9

hiya,

on chapter 15 part 1

page 526 on book
function conv1d:

for i in range(0, int(len(x)/s),s)

this is a quick fix, only works with current example, if you augment padding by one it wont work,

id recommend this

for i in range(0, int((len(x_padded) - len(w_rot) / s)) + 1, s):

s = os.path.join('https://archive.ics.uci.edu', 'ml',
                 'machine-learning-databases', 'iris','iris.data')

on windows gives: https://archive.ics.uci.edu\ml\machine-learning-databases\iris\iris.data
which does not work. I would do a pull request and suggest replacing it with:

s = 'https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data'

but, I don't know if there is a reason for doing the join.

1

2

3

4

5

6

Did you comment out the random seed by accident? Note that the results change.

7

8

9

10

11

#12

13

Part 2/3

14

Part 3/3

15

16

Ch 17: Synthesizing New Data with Generative Adversarial Networks

Pg. 75, third line:

Notice that the columns all sum up to one

should probably be

Notice that the rows all sum up to one

Change

import pandas as pd

df = pd.read_csv('https://archive.ics.uci.edu/ml/'
        'machine-learning-databases/iris/iris.data',
                 header=None,
                 encoding='utf-8')
df.tail()

to

import os
import pandas as pd


s = os.path.join('https://archive.ics.uci.edu', 'ml',
                 'machine-learning-databases', 'iris','iris.data')
print('URL:', s)

df = pd.read_csv(s,
                 header=None,
                 encoding='utf-8')

df.tail()

In chapter 13, in

• the notebook
• the PDF

change

images -> cat_doc_images

Red = old, green = new

Note that the cell with the code output is always shown above the cell with the code for some reason.

1

2

3

4

5

6

7

8

9

10

PART 2

11

12

13

(also change the figure here)

14

For some reason i got different results, pls double check

5th row from the bottom
x^{(i)} \epsilon \mathbb{R}^{150 x 1}
I think it should be

x_{j} \epsilon \mathbb{R}^{150 x 1}

(I used https://www.codecogs.com/latex/eqneditor.php to check my latex)

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

modify the sentence that says that we used the same architecture for W-DCGAN and DCGAN

however, it does not make the original data set normally distributed

standardization does make the data set normally distributed.

book errata p 682

the lower sum doesn't need the "c" over the sum.

plt.ylim([0.8, 1.0])
plt.tight_layout()
# plt.savefig('images/06_06.png', dpi=300)
plt.show()

change to

plt.ylim([0.8, 1.03])
plt.tight_layout()
# plt.savefig('images/06_06.png', dpi=300)
plt.show()

Original:

plt.text(-3.5, -5., 
         s='Sepal width [standardized]', 
         ha='center', va='center', fontsize=12)
plt.text(-12.5, 4.5, 
         s='Petal length [standardized]', 
         ha='center', va='center', 
         fontsize=12, rotation=90)

#plt.savefig('images/07_05', dpi=300)
plt.show()

Change:

plt.tight_layout()

plt.text(1, -0.2,
         s='Sepal width [standardized]',
         ha='center', va='center', fontsize=12,
         transform=axarr[1, 0].transAxes)

plt.text(-0.12, 0,
         s='Petal length [standardized]',
         ha='center', va='center',
         fontsize=12, rotation=90,
         transform=axarr[0, 0].transAxes)

with the initialized of the Perceptron class , the book mentioned the parameter "n_iter", the i cannot found in the code, and the code of 2nd has it. so i think it should be a issue