Week 2 Exercise 2 (Understanding the logic case scaranos and output)

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
,
1 
Hi, I have trouble understanding the logic when running the code, below 
One Case with a complete mini batch size only i.e each of 64 examples.   
mini_batch_X = shuffled_X[:,(k*num_complete_minibatches):((k+1)*(num_complete_minibatches))]
    mini_batch_Y = shuffled_Y[:,(k*num_complete_minibatches):((k+1)*(num_complete_minibatches))] 


And other for handling the end case (last mini-batch < mini_batch_size i.e less than 64)
 mini_batch_X = shuffled_X[:,(k)*(m-mini_batch_size*num_complete_minibatches):(k+1)*(m-mini_batch_size*num_complete_minibatches)]
        mini_batch_Y = shuffled_Y[:,(k)*(m-mini_batch_size*num_complete_minibatches):(k+1)*(m-mini_batch_size*num_complete_minibatches)]

After that, here is my output. 
![opt1|690x397](upload://qwGzhEwJ5zjNZxRSeOumI80tGEw.png)
![opt2|690x404](upload://r0gTyDtQI6N2CmM6zbV4i6fRRqG.png)

I want to know how to proceed this properly. (i.e. refer to some examples and relevant slides, etc.). Thank you.
2

opt1
opt11204×693 51 KB

opt2
opt2724×424 23.3 KB

3

Please think through what needs to happen in terms of the indexing to extract the various minibatches. Why is k * num_complete_minibatches relevant in that context? It is the size of the minibatches that is important there, not the number of minibatches, right? So the first one starts at index 0 and includes mini_batch_size elements. So how would you express that as a start:end index range in python?

1 Like
4

opt3
opt31206×698 51.9 KB

I got all the tests passed. Now, I figure out why my shape for the final batch is not matched.

5

Interesting. Note that the size of the last partial batch was correct in the first version that you posted at the beginning of this thread (size = 20). So you fixed the full minibatch case, but now the partial case is broken in the same way that the full version was broken the first time around (size = 2).

That seems like a useful clue as to the nature of the error. :nerd_face: