"Unsupervised Learning "Course 3 week 1 practice lab

Course Q&A Machine Learning Specialization Unsupervised Learning, Recommenders, Reinforcement
1

Hi,
I was trying to solve for the select_threshold function lab question with the following code:

def select_threshold(y_val, p_val):
“”"
Finds the best threshold to use for selecting outliers
based on the results from a validation set (p_val)
and the ground truth (y_val)

Args:
    y_val (ndarray): Ground truth on validation set
    p_val (ndarray): Results on validation set
    
Returns:
    epsilon (float): Threshold chosen 
    F1 (float):      F1 score by choosing epsilon as threshold
""" 

best_epsilon = 0
best_F1 = 0
F1 = 0

step_size = (max(p_val) - min(p_val)) / 1000

for epsilon in np.arange(min(p_val), max(p_val), step_size):

    ### START CODE HERE ### 
    tp = 0
    fp = 0
    fn = 0
    rec = 0
    prec = 0
    for i in range(len(p_val)):
        if(p_val[i]< epsilon):
            p_val[i] = 1
        else:
            p_val[i] = 0
        if p_val[i]== 1 & y_val[i] == 1:
            tp+=1
        elif p_val[i] == 1 & y_val[i] == 0:
            fp+=1
        elif p_val[i] == 0 & y_val[i] == 1:
            fn+= 1
            
    rec = tp/(tp+fn)
    prec = tp/(tp+fp)
    
    F1= 2*prec*rec/(prec+rec)
 
    
    ### END CODE HERE ### 
    
    if F1 > best_F1:
        best_F1 = F1
        best_epsilon = epsilon
    
return best_epsilon, best_F1

however I am getting the following error:

ZeroDivisionError Traceback (most recent call last)
in
1 p_val = multivariate_gaussian(X_val, mu, var)
----> 2 epsilon, F1 = select_threshold(y_val, p_val)
3
4 print('Best epsilon found using cross-validation: e' epsilon)
5 print('Best F1 on Cross Validation Set: f' F1)

in select_threshold(y_val, p_val)
42 fn+= 1
43
—> 44 rec = tp/(tp+fn)
45 prec = tp/(tp+fp)
46

ZeroDivisionError: division by zero

What am I doing wrong? (I know it has to be solved by vectorisation, I was just trying to solve it with a for loop)

2

Hi @fulorianarendra5

First of all, you cannot share your code here, try to address your problem through error message and etc.

Now, the problem happens when tp + fn or tp + fp is zero. To fix it, use and instead of & for logical conditions as & is used for bitwise AND operations.

Btw, it is a good practice to avoid modifying p_val in place by creating a new variable for predictions!

Hope it helps! Feel free to ask if you need further assistance.

1 Like
3

Hey Thanks for the help. Had been stuck for a long time on the same!!. Will keep your instruction in mind for sure from next time.!! Thanks again :slight_smile:

2 Likes
4

You’re welcome! Happy to help and good luck :raised_hands:

1 Like