(doc) Questions on back-propagation algorithm about module-1 HOT 7 CLOSED

Another typo found in the doc. I just run the task 1.5 and found that the figure in doc is not in proper order. The second one should be Xor curve rather than a hyperplane.

from module-1.

Okay, I think I got most of these changes. I will leave this issue as a reminder to make the backprop docs better.

from module-1.

Right... We could drop that step for simplicity. Speed-wise, implemented with a good hashing algorithm it will make things faster. But I think you make a good point.

from module-1.

Right... We could drop that step for simplicity. Speed-wise, implemented with a good hashing algorithm it will make things faster. But I think you make a good point.

I see, seems that it would be niec to address this in the doc page and leave it as an optional exercise.

from module-1.

Besides, I don't know if I implement the algo right. I need to filter out those Variable with its self.history=None(these Variables are constant that are input for ScalarFunction which got wrapped into Variable), which is not mentioned in the algo description and I struggle to find the bug for a long time.

from module-1.

This is very useful feedback, I will clarify both points in the docs and update the example.

Please keep sending comments or clarifications as you go.

from module-1.

Right. History=None means the variable is a constant or doesn't require a derivative. I will add a property so that is less confusing and update the docs in chain_rule.

We're you able to pass the tests otherwise?

from module-1.