Kind of discouraged-- error on final lab

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
,
1

Frustrating lab

This part:

Information Gain from splitting the root on brown cap:  0.034851554559677034
Information Gain from splitting the root on tapering stalk shape:  0.12451124978365313
Information Gain from splitting the root on solitary:  0.2780719051126377

works just fine. I’m matching all three of those.

There seems to be some test immediately following those, though, and that’s where I’m struggling. Here’s a sample of my debugging output.

NODE INDICES GIVEN: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
Data selected for feature 2: [1 1 0 0 1 1 0 1 0 0]
my left 1,   [0, 1, 4, 5, 7]
my right 1,  [2, 3, 6, 8, 9]
Left indices:  [0, 1, 4, 5, 7]
Right indices:  [2, 3, 6, 8, 9]
snode =1.0, sleft=0.7219280948873623, sright=0.7219280948873623, wleft=0.5, wright=0.5
Information gain: 0.2780719051126377
Information Gain from splitting the root on solitary:  0.2780719051126377
NODE INDICES GIVEN: [0, 1, 2, 3, 4]
Data selected for feature 0: [1 1 1 0 0]
my left 1,   [0, 1, 2]
my right 1,  [3, 4]
Left indices:  [0, 1, 2]
Right indices:  [3, 4]
snode =0.0, sleft=0.0, sright=0.0, wleft=0.6, wright=0.4
Information gain: 0.0
NODE INDICES GIVEN: [0, 1, 2, 3, 4]
Data selected for feature 0: [1 1 1 0 0]
my left 1,   [0, 1, 2]
my right 1,  [3, 4]
Left indices:  [0, 1, 2]
Right indices:  [3, 4]
snode =0.0, sleft=0.0, sright=0.0, wleft=0.6, wright=0.4
Information gain: 0.0
NODE INDICES GIVEN: [0, 1, 2, 3, 4]
Data selected for feature 0: [1 1 1 0 0]
my left 1,   [0, 1, 2]
my right 1,  [3, 4]
Left indices:  [0, 1, 2]
Right indices:  [3, 4]
snode =0.9709505944546686, sleft=0.9182958340544896, sright=1.0, wleft=0.6, wright=0.4
Information gain: 0.01997309402197489
NODE INDICES GIVEN: [0, 1, 2, 3, 4]
Data selected for feature 1: [0 0 0 0 1]

That’s up to where it crashes

and the error I get

TypeError                                 Traceback (most recent call last)
<ipython-input-103-2a50df79e115> in <module>
      9 
     10 # UNIT TESTS
---> 11 compute_information_gain_test(compute_information_gain)

~/work/public_tests.py in compute_information_gain_test(target)
    105     assert np.isclose(result, 0.019973, atol=1e-6), f"Wrong information gain. Expected {0.019973} got: {result}"
    106 
--> 107     result = target(X, y, node_indexes, 1)
    108     assert np.isclose(result, 0.170951, atol=1e-6), f"Wrong information gain. Expected {0.170951} got: {result}"
    109 

<ipython-input-102-3bf83ca050ed> in compute_information_gain(X, y, node_indices, feature)
     18     """    
     19     # Split dataset
---> 20     left_indices, right_indices = split_dataset(X, node_indices, feature)
     21 
     22     # Some useful variables

<ipython-input-100-8fc02917f9b6> in split_dataset(X, node_indices, feature)
     53 
     54     #this step converts them to a list, because the QA tests fail if its not a list
---> 55     left_indices=list(left_indices) #this should be a list
     56     right_indices=list(right_indices) #this should be a list
     57     print("my left 1,  ",  left_indices)

TypeError: iteration over a 0-d array

I have found many errors very helpful and made a lot of improvements based on these errors over the course of the last few hours. That’s how I’ve gotten to the point where I’m nearly able to start section 4.4!

However, I’m kind of discouraged by this one. Is it really saying that the method I used, using numpy, for “split_dataset” is inherently unacceptable to the autograder, and I have to go back and redo everything!!! :exploding_head:

Please help, if you have any suggestions!

Thanks,
Steven

2

Your code in lines 55 and 56 is not necessary.

The task in split_dataset() is just to iterate through node_indices (using a for-loop), and add an element to the left_indices or right_indices by using the .append() method.

3

Thank you! I figured it out!

I really appreciate you helping me understand what was intended, and I think I now understand what motivates it when I see that later it is used in a recursive function. I was wondering, why not numpy, before that?