Logistic Regression Cost Function (C1W2L03)

Take the Deep Learning Specialization: bit.ly/3cmtNgK
Check out all our courses: www.deeplearni...
Subscribe to The Batch, our weekly newsletter: www.deeplearni...
Follow us:
Twitter: / deeplearningai_
Facebook: / deeplearninghq
Linkedin: / deeplearningai

[@hunter330]

At 4:15, Andrew should say that we want -ylog(y_hat) "as small as possible" instead of "as big as possible". Source: His notes on Coursera.

[@khaledadrani3184]

+1

[@mahsa5527]

No, we have a minus before log(y~) and this should be the max if we want to make the value of the loss function minimum.

[@mukunthag8760]

He is an extremely good teacher!

[@rameshmaddali6208]

Best video on loss and cost function explanation

[@Morpho32]

At 4:20, he says "we want -log(y hat) to be as big as possible". That's not true, we want to be as small as possible since it's the loss function and that's why we want y_hat to be as big as possible so that actually -log(y_hat) is as small as possible.

[@morancium]

i think you messed up with negative sign

[@sharmakartikeya]

At 4:15, sir mistakenly says that we need to have -log(y_hat) as big as possible. It is the opposite. After all -log(y_hat) is itself the cost function for some y_hat and y = 1. So for - log(y_hat) to be as small as possible (for y = 1) we would want log(y_hat) to be as large as possible. For this we would want y_hat to be as large as possible (Remember the graph of log x, it is a monotonically increasing curve). Remember that y_hat is nothing but the probability (our model produces) of y=1 for some weights and bias, so y_hat can never be greater than 1. So we want our y_hat to be as close to 1 as possible.

[@HabibuMukhandi]

I don't think he mistakenly said it. Remember loss is a bad thing so want to have a big negative loss as possible!

[@zs9510]

Explained very well.

[@1052boon]

설명 감사합니다:)

[@skypickle29]

@:22 we use Wtranspose. Why transpose? The first sample is a column vector. How is W defined? Isn't it a column vector as well in which case you only need to multiply x*W , not W transpose.?

[@papayaspice1155]

explained in a comment on the previous video

[@ramazancesur6143]

this course is very great which beginner level for all in neural network course

[@chitralalawat8106]

Hey! Looks like you're seriously understanding what the speaker is saying.. Could you please help me with this... I want you to join my group so that we all can study further ..!!

[@prafulbs7216]

what is difference Y^(y hat) and y ?

[@nikilkumarr]

y hat is the predicted value (as calculated from the logistic regression model) while y is the actual value (or value of the label). Pretty much, the error is the difference between what the model predicts and the actual value :)

[@uqyge]

great intro course

[@chitralalawat8106]

Hey! Looks like you're seriously understanding what the speaker is saying.. Could you please help me with this... I want you to join my group so that we all can study further ..!!

[@louerleseigneur4532]

Thanks god Thanks sir

[@iammakimadog]

2:27 technically you can still use MSE because MSE is convex

[@andretelfer3678]

I am struggling to understand this too... I hope this comes up in a later video

[@e.galois4940]

Nah, squared error is not convex for sigmoid function bro

[@sandipansarkar9211]

need to make notes

[@prafulbs7216]

could someone explain dimension of weight matrix?

[@papayaspice1155]

explained in a comment on fhet video before this one

[@StarContract]

A neural network is trying to figure out how neural networks work at the moment.

[@doubtunites168]

how is your loss function so far?

[@chitralalawat8106]

Hey! Looks like you're seriously understanding what the speaker is saying.. Could you please help me with this... I want you to join my group so that we all can study further ..!!

[@abhinavkommula4588]

@@chitralalawat8106 stop trolling please

[@chitralalawat8106]

@@abhinavkommula4588 you mind ur own business

[@abhinavkommula4588]

​@@chitralalawat8106 oKay bUdDY mInd tHe tOxIciTy yEaH?

[@meirgoldenberg5638]

At 7:54, Andrew says that the next video would show that logistic regression can be viewed as a small neural net. However, the next video does not seem to fulfill this promise.

[@ashwinilalawat604]

Yes

[@ashwinilalawat604]

No

[@rahulkumar-xl9pt]

Why does the log function need to be large when y is both 0 or 1?

[@kangzheng1390]

rahul kumar they are two different cases.

[@sebastianrodriguezcolina634]

what you want is that your cost function is large when your prediction is far from the true value

[@chitralalawat8106]

Hey! Looks like you're seriously understanding what the speaker is saying.. Could you please help me with this... I want you to join my group so that we all can study further ..!!

[@justin119933]

so as to keep the loss function as small as possible on both scenarios. (check the writing in green on the video at 5:42)

[@zhilinglin4648]

Hi I also have the same question and finally sort it out after reading this post. towardsdatascience.com/optimization-loss-function-under-the-hood-part-ii-d20a239cde11

[@syahirdev3193]

my brain hurts...

[@susnatodhar7197]

can someone tell me what is 'e'?

[@caducoelho2221]

e is the Euler's number.

[@caducoelho2221]

www.mathsisfun.com/numbers/e-eulers-number.html

[@ahasanhabibsajeeb1979]

Why don’t i understand you?