Checks for my numpy multilayer ANN : Feedback required

in the uploaded ‘pdf’
doubt.pdf (71.7 KB)
are Checks wherever I have doubts and thought could be potential hiccups.

The uploaded file has screenshots of output windows, highlighted codes of my inbuilt functions where computations are made & where I have doubts.
PS: I am almost sure, there would be potential bugs in my code, but from a concept perspective checking my understanding of the nuts&bolts.
Any feedback and correction very much appreciated!
Thanks much

Hello @tennis_geek,

Your check 2 wasn’t implemented correctly. Your implementation implies that, for example, all layers’ dj_db are the same, which implies that their difference of being in different layers doesn’t matter at all. I suggest you to audit course 1 of the Deep Learning Specialization (DLS) for how to propagate errors from the output layer back to each hidden layer. If you choose to subscribe to the DLS, you will have access to week 4’s assignment that actually requires you to implement a multi-layer neural network.

Happy new year, and cheers,
Raymond

100% correct, as for a dj/db for each node an error with the same index is added. So, basically node 1 in layer1 & node 1 in layer2 will have identical dj/db sum result. Hope we are on the same note here?

Am still in week 2 of ML Advanced learning algorithms course., hope I can handle the nuances in DLS course 1.

Does this imply, current ML course 2 has some important details pending for discussion? Mostly I followed my previous single hidden node implementation from the scratch.

Happy New Year!

But this should not be the case in a correct implementation of a multi-layer neural network.

Please plan your time. However MLS C2 will NOT guide you through a numpy implementation of a multi-layer neural network. You will only use Tensorflow for that here.

It implies that whatever you learned for a single layer neural network, you can only use it for the output layer of your multi-layer neural network. As for the hidden layers in the middle, you need something else that will be covered in the DLS Course 1.

Raymond

Got it Thanks! Good thing I started implementing multilayer … Will keep it for further conceptual corrections as I progress in my learning.
By 100% correct, I meant about your pointing of the incorrect implementation of dj/db for each layer which I have now. :grinning: Not that my implementation is correct.
I could immediately connect to it only after checking the code. Thanks again!

I see. Good luck with the rest of the course, and perhaps DLS C1 if you will consider it in the future :wink:

Cheers,
Raymond

1 Like