Cost function is also called as Loss function, both are synonyms. Division by m or 2m is interchangeable. What we are really concerned about it a model which is producing least error , not the value of the loss function directly. Cost functions can be of 3-types , among them is a regression, which again has 3 types , that is Mean Error , Mean squared error , mean absolute error . Why so many ? Cause a data set may have negative/positive errors , taking mean directly may cancel out +/- errors , and taking a square directly can be a bit troublesome if you have some outliers . In these videos , Andrew can be seen using all three in regression based . Note :it's not the only required param for concluding a model isn't good.

I know right

I am pretty sure coursera copied from his content.

he is coursera

@@GelsYT hahaha true..

wrong dude, the other guy wants to give you UBI. This guy wants to give you OVA

X is 4x5 and X(transpose) is 5x4. Therefore, X(transpose)*X = 5x4 * 4x5 which results in a 5x5 matrix, which has an inverse.

Sending you the bill...

@@namangupta8609 Did you receive the bill? Or you’ll be the only youtuber watching this video

learning rate

@@ofathy1981 alpha is the learning rate in gradient descent .... theta is a parameter like weights in NN

@@ofathy1981 Theta does not represent the learning rate, instead it represents the parameters of the model (e.g. the weights). So P(y | x; θ) translates to English as "The probability of *y* given *x* , parameterized by *θ* ".

That the parameters of logistic classifier which is trained separately for each case