## Neural Network using BackPropogation in Python

Support me on Patreon: https://www.patreon.com/ajays97

Our Instagram: https://www.instagram.com/geeksnome
Our Blog: http://geeksnome.com

Source

### Comment List

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

It's a very bad practice to name activation function results in terms of z. It creates confusion and slows the learning rate.

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Very simple and excellent one, thank you sir, but LR and Bias concept missing. give us one more with including those.

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

is a hidden layer the same as softmax layer?

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Thank you

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

It's really helpful for me. Thank you

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

A small note that the Weight 1 matrix is actually 2×3 not 3×2.. Since you transposed the matrix implicitly in the _init_ function when you switched the inputs and neurons. That is, for the Weight matrix, the rows become columns and columns become rows (Columns x Rows), and not Rows x Columns like usual. The dot product rule of the matrices says that the the Column (2nd dimension of the matrix) of the first matrix must match the number of Rows (first dimension) of the second matrix. In the example in the video, the input size is 2, weights correspond to the number of neurons in the hidden layer, so the matrix size is 2 inputs by 3 neurons [2][3] then the dot product of [3]1] for the output layer. Notice that 3 of the columns of the first matching the 3 in the rows of the second matrix.

For the output, the second rule of matrix multiplications says that the size of the output is the Rows of the first matrix and the Columns of the second matrix. In the example above, the output matrix will have the size of [2][1], that is, 2 rows (weights), one column (neuron).

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

np.dot(X, self.W1) . how does this work here X is 3×2 and W1 is also 3×2.

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

This was really informative sir, but could you try a larger sample space and make sure the code is not memorizing? cheers

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Thank you ❤ I know the concept but dont able to code the matries now I able

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Why didn’t you optimize the loss function? You did (y-output) shouldn’t it have been (y-output)**2. The theory behind the choice of lost or cost function is fascinating

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

font is too small to read

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Hello Sir, could you explain where the learning rate and the biases are

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

please give an example by considering a dataset

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Sir my question is how to calculate weights regarding fake and real news (titles)

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

When you try to scale this from one hidden layer to n hidden layers, you will realize that his explanation is as good as the dead horse.

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Never thought this would happen! :-}

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

sir why didnt u use learning rate?

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

It was really very helpful!! But can you please explain why did you evaluate output_delta separately?? As I have found in some videos they have considered output_delta=output_error.

• GeeksNome - Hacking, Programming & Technology
December 7, 2020

Nice. It's really helpful to understand the bp.

• GeeksNome - Hacking, Programming & Technology
December 7, 2020