Multiclass Classification Data flow

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
1

I have a general conceptual misunderstanding of how data inflows (and outflows).
I will pull examples from the last graded assignment to make it easier to illustrate.

If each row in the X input matrix is a category (eg, the number 1 – from final project), when does the model iterate over the second category (second row) of X.
Second: are w1..w25 in layer[1] apply to all categories after the model has been trained?

if my misundertanding is with matrix operations, do:
all 5000x400 values in X go to UNIT1 in layer[1] TO estimate w1, b and
AND all 5000x400 values in X go to UNIT2 in layer[1] to estimate w2, b2
AND all 5000x400 values in X go to UNIT2 in layer[1] to estimate w3, b3…

Thanks for helping me understand order of operation.
SB

2

I’m going to assume you’re working on the identification of 10 handwritten digits.

X contains all of the examples and the original features (5000 x 400).

It passes through the first layer. What comes out are 5000 examples, but only 25 new features. These are the new features that the hidden layer units learned. So that layer provides (5000 x 25) to the next layer.

The same reduction in the number of features happens at the output layer, but there we have 10 outputs (one corresponding to each label). So the output layer creates (5000 x 10) at the output.

3

Thank you for the clarification TMosh.

S

1 Like