1 of 38

HW2 TA hours

TAs

ntu.mlta@gmail.com

2 of 38

Outline

如何在shell scripts與python檔中接收arguments�
Logistic Regresion�
Generative model

3 of 38

Shell Scripts

What is Linux Shell?

Shell accepts your instruction or commands (usually in English) and pass it to kernel.

4 of 38

Shell Scripts

2. What is Shell Script?

… 把很多指令寫成一個檔案，可以一次做很多事情。

甚至是做判斷式或是迴圈。

5 of 38

Script Tutorial and Example

Shell Script Tutorial: http://linux.vbird.org/linux_basic/0340bashshell-scripts.php

Example:

6 of 38

Passing Arguments

How to pass arguments to your srctipts or .py files?

在terminal或cmd中的輸入與script中變數對應關係：

/path/to/scriptname /path/to/data /path/to/output

$0 $1 $2

2. 在terminal或cmd中的輸入與.py中變數對應關係：

/path/to/pythonfile /path/to/data /path/to/output

sys.argv[0] sys.argv[1] sys.argv[2]

7 of 38

Passing Arguments

script:

python:

8 of 38

Passing Arguments

Let’s run…

9 of 38

PATH

Ex: /Users/MLTA/Desktop/hw2

相對路徑：相對於目前資料夾的路徑。

Ex: ./hw2/logistic.py、../hw1/linear_regression.py

助教會在git clone整個 ML2017/hw2資料夾
並在 ML2017/hw2資料夾執行程式(Ex : hw2_best.sh)
若要讀/存model請用相對路徑
Data的Path助教會用絕對路徑下在hw2_best.sh的argument裡。

10 of 38

Logistic Regression

Implementation:

feature normalization
logistic regression using gradient descent with batch

Code Example:

http://codepad.org/RvIY9BU2

11 of 38

Logistic Regression

If we let batch_size=25...

12 of 38

Logistic Regression

-

13 of 38

Gaussian Distribution-Training(mean)

14 of 38

Gaussian Distribution-Training(sigma)

15 of 38

Gaussian Distribution-Training(sigma)

Shared sigma:

16 of 38

Gaussian Distribution-predict

17 of 38

Gaussian Distribution-sigmoid

使用np.clip() 避免數值太小或太大而overflow

18 of 38

Generative model -Naive Bayes Classifier

Calculate the probability in each category
P(C1 | X) = 連乘P(C1 | X in each category)
log(P(C1 | X) ) = sigma(log(P(C1 | X_i)))
比較 log(P(C1 | X) ) 和 log(P(C2 | X_i))

19 of 38

code

http://codepad.org/wlrT0lw4

20 of 38

HW2 TA hours

TAs

ntu.mlta@gmail.com

21 of 38

Outline

如何在shell scripts與python檔中接收arguments�
Logistic Regresion�
Generative model

22 of 38

Shell Scripts

What is Linux Shell?

Shell accepts your instruction or commands (usually in English) and pass it to kernel.

23 of 38

Shell Scripts

2. What is Shell Script?

… 把很多指令寫成一個檔案，可以一次做很多事情。

甚至是做判斷式或是迴圈。

24 of 38

Script Tutorial and Example

Shell Script Tutorial: http://linux.vbird.org/linux_basic/0340bashshell-scripts.php

Example:

25 of 38

Passing Arguments

How to pass arguments to your srctipts or .py files?

在terminal或cmd中的輸入與script中變數對應關係：

/path/to/scriptname /path/to/data /path/to/output

$0 $1 $2

2. 在terminal或cmd中的輸入與.py中變數對應關係：

/path/to/pythonfile /path/to/data /path/to/output

sys.argv[0] sys.argv[1] sys.argv[2]

26 of 38

Passing Arguments

script:

python:

27 of 38

Passing Arguments

Let’s run…

28 of 38

PATH

Ex: /Users/MLTA/Desktop/hw2

相對路徑：相對於目前資料夾的路徑。

Ex: ./hw2/logistic.py、../hw1/linear_regression.py

助教會在git clone整個 ML2017/hw2資料夾
並在 ML2017/hw2資料夾執行程式(Ex : hw2_best.sh)
若要讀/存model請用相對路徑
Data的Path助教會用絕對路徑下在hw2_best.sh的argument裡。

29 of 38

Logistic Regression

Implementation:

feature normalization
logistic regression using gradient descent with batch

Code Example:

http://codepad.org/RvIY9BU2

30 of 38

Logistic Regression

If we let batch_size=25...

31 of 38

Logistic Regression

-

32 of 38

Gaussian Distribution-Training(mean)

33 of 38

Gaussian Distribution-Training(sigma)

34 of 38

Gaussian Distribution-Training(sigma)

Shared sigma:

35 of 38

Gaussian Distribution-predict

36 of 38

Gaussian Distribution-sigmoid

使用np.clip() 避免數值太小或太大而overflow

37 of 38

Generative model -Naive Bayes Classifier

Calculate the probability in each category
P(C1 | X) = 連乘P(C1 | X in each category)
log(P(C1 | X) ) = sigma(log(P(C1 | X_i)))
比較 log(P(C1 | X) ) 和 log(P(C2 | X_i))

38 of 38

code

http://codepad.org/wlrT0lw4