Linear Regression using Gradient Descent in Python - Machine Learning Basics

[@iamdurgeshk]

Man, loved the explanation which I tried multiple times from different sources to understand. Why did you stop making such video? I would love to learn more in Machine Learning. Please start a playlist explaining A-Z of Machine Learning.

[@albertobrody1979]

i know Im randomly asking but does anyone know a trick to get back into an instagram account?? I stupidly forgot my account password. I would love any assistance you can give me

[@romanrandall2106]

@Alberto Brody instablaster ;)

[@iamdurgeshk]

Ahaha, scammers replying to each other

[@albertobrody1979]

@Roman Randall thanks for your reply. I found the site thru google and Im waiting for the hacking stuff now. Seems to take a while so I will get back to you later with my results.

[@albertobrody1979]

@Roman Randall it worked and I now got access to my account again. I'm so happy! Thank you so much you saved my account!

[@dileepn2479]

I took almost a week to find a such informative video to understand gradient descent ...

[@sairasiddique5116]

so fluent clear concepts that were like a big rock in front of me when our lecturer was giving us a lecture and I was like a dump but now I get everything with each detailing you have explained great work

[@AdarshMenon]

Glad it was helpful!

[@sathish77j90]

Really fantastic sir... I am struggling to understand thia concept from 3 days finally now i got it... Thank u very much sir..

[@charlescoult]

Simple and straightforward. Thank you Adarsh.

[@syedniamath5841]

Great Video Adarsh, short and clear to understand and workout.

[@AdarshMenon]

Thank you Syed !!

[@umersalman2506]

This is amazing man! Pls upload more videos on Machine learning algorithms with this type of proper intuitions and implementation

[@AdarshMenon]

Thank you so much ! Yes I am working on more such videos. Thank you for subscribing

[@turtlepedia5149]

sir very very thanks after my job big donation coming to you

[@late_nights]

good explanation. Your 1000th subscriber here

[@AdarshMenon]

Awesome, thank you!

[@modernman00]

You are the best teacher. I recommend this course to anyone studying machine learning at Coursera. Please more video and continue with your teaching approach

[@kalaivananr]

Thank you, Adarsh. You made it very simple.

[@nisarahmed4487]

@Adarsh ... no doubt you have done great job, but can you tell me why you are running the for loop on length of x?? but not on the epochs as you have already told that you are going to use it ?

[@aman5534]

in his git repo he is using epochs as he said

[@kruti2008]

Very well explained, step by step with clarity 👍

[@sayandey1478]

Impressive! That was a very clarifying one, but in the real world situations we don't know how many iterations are suitable for a given problem, moreover, you are not checking whether you have reached your minimum but are concluding from the data visualization. If you could clear that training algorithm to fix learning rate, which I have been looking for so far, could be very helpful.

[@prathamgenius]

Thanks a lot @Adarsh , you helped me to complete ML assignment easily... Cheers 👍

[@AdarshMenon]

Glad to hear that!

[@rashidmehmood5008]

Where did you use epochs in the code?? You declared it but I don’t think, you used it.

[@shubhamgoyal977]

Great video man!! Loved it!! Helped me understand this concept really well. Thanks!!

[@AdarshMenon]

Glad to hear it!

[@adarshsingh401]

plz also make a video on multiple linear regression!!. This one was amazing.

[@igorjaqueira5788]

I believe there is a mistake in the code. First as you pointed out already in other comments, the variable epochs should be used instead of n. But also there should be an error limit, so the loop should run as long as the error is greater than some parameter (epsilon).

[@AdarshMenon]

Yeah that is a good idea when training times are long, and it does not overfit. But yes a good best practice I could have mentioned, thanks !

[@veerasekhar8551]

thanks man :) subscribed expecting you to explain more ML algorithms.

[@AdarshMenon]

More to come!

[@sayonsai8006]

Very easy to understand and implement

[@sathishkumaranandashekaren4313]

Hi , what is the need for adding epochs in the program without epochs also my program is running

[@janhvibhosle3852]

Excellent video sir, please make more videos on machine learning with easier explanation.

[@AdarshMenon]

Thanks Ritesh ! Will sure do !

[@davidthunder9804]

Hey how you did that gradient descent converging animation in colab great video do

[@chetansahu1505]

Please upload more videos, all the model - KNN, Naive Bayes, Decision Tree, Random forest, dimentioanlity reduction techniques, etc all the models including the Depp learning from scratch. please make. Because the way you explain any layman to fool will easily understand. Awesome explanation bro. Big fan with your one video.

[@MOHSINALI-bk2qo]

plz give me two variable example of gradient descent.

[@narottamaswal3978]

I liked and subscribed at this moment 2:47 🙏

[@swapnamondal7161]

Very intuitive video.

[@nano7586]

Best explanation so far.. thanks a lot!

[@tejaskulkarni5795]

Nice Explanation very clear.

[@desavera]

Fantastic ... really solid !!! thanks ...

[@raghavgaur8901]

Sir,where did you use epochs in your code as it is just initialized.

[@AdarshMenon]

Hi epochs were to be used in the for loop instead of n. Please see the code on GitHub (link in description)

[@mohammeddanishreza4902]

Why I am getting m, C value as in array, can u plz help me?

[@neehanthreddy1868]

Could you provide an example with multiple independent variables?

[@jyotshanajha8316]

very well explained. thank you

[@sudeshnasen2731]

What is epoch? Why are we declaring it without using it? Why run the loop for the length of X instead of the epoch?

[@muhammadalikhan5003]

Thank you. It was very useful.

[@bibekanandasahoo3497]

Adarsh can you please make a vidio which explain how the gradient discent work in multiple linear regression ? please !

[@MinaNaghshnejad]

Nice explanation. In gradient descent, the loop shouldn't for loop changed to for i in range(epochs)?

[@AdarshMenon]

Hi yes, you are right. It was a mistake, but it worked because the dataset converged quickly. Thank you for pointing it out !

[@DB-in2mr]

Hi just a note if somebody might get some issue by gathering data from your github data.csv ...there is an error related to " " or also like "Error tokenizing data. C error: Expected ....etc". Apprently ideal is to upload your data in raw format from your git area here ( raw.githubusercontent.com/chasinginfinity/ml-from-scratch/master/03%20Linear%20Regression%20in%202%20minutes/data.csv ) , adding ", header=None" .

[@vinaywaingankar6133]

Great explanation

[@AdarshMenon]

Glad you liked it

[@SaurabhKumar-gj5mp]

One of the best video. Nice explanation man

[@AdarshMenon]

Thank you !

[@aishwaryaravi5249]

this video really helped me! one question though, it gives errors for higher learning rates. why is that so?

[@sreeprakashneelakantan5051]

Very well explained 👌

[@sjback2145]

Great thanks to you. It's nice and clear explanation

[@Christian-mn8dh]

Hey man, love this vid. I tried to build a gradient descent algorithm from scratch too. Why isn't mine working? Why can't you multiply the learning_rate directly by the error? Here's my code: for i in range(4): ypred = m * x + b error = (ypred - y) **2 m = m - (0.001 * error) b = b -(0.001 * error) m = m.sum() b = b.sum() #My 'm' and 'b' values decrease infinitely

[@kaustubhkapare807]

length of training dataset typecasted to float....and that float was given as a parameter to range function....

[@Mason-yj1ip]

This is better than my stupid university lecture. Bro you should take my tuition.

[@ROHANJUNEJABAI]

We repeat this process until our loss function is a very small value or ideally 0 (which means 0 error or 100% accuracy). The value of m and c that we are left with now will be the optimum values. THAT THE 4TH STEP OF GRADIENT DESCENT but according to your code, we are just updating the values till epochs =100 . HOW THIS JUSTIFY THAT WE ARE UPDATING THE VALUES TILL THE LOSS FUNCTION IS MINIMIZED?

[@hirakmondal6174]

Exactly my point. This code is wrong and some people with 0 to none knowledge who are learning are saying this to be GOD LEVEL..smh

[@2020-h2s]

have you implemented gaussian kernel regression from scratch for two independent features dataset.

[@destinedtotrade3214]

I liked & subscribed Thx !

[@harshmankodiya9397]

loved the explanation

[@AdarshMenon]

Glad to hear!

[@navinbondade5365]

I think we you need to add for i in range(iteration)

[@namanjain1164]

The best!! Loved it!!

[@sarthak810]

After plotting the regression line can we apply the liner regression algorithm from sklearn to it

[@agirlfrom_mountains]

Where we r using epiches here and please explain me using max and min how we are are Ploting the gradient decent line???

[@tomtrask_YT]

You really should do something like this to read your csv (the way you do it, it uses the first row as column names) data = pd.read_csv('data.csv', header=0, names=["x_val", "y_val"]) You don't need to assign names but it's "nice" to do that, that way you don't have to use indexes (e.g. use data["x_val"] instead of data.iloc[:, 0])

[@AdarshMenon]

Thanks Tom, yes totally agree with you. Will keep this in mind from now onwards !

[@usamazahid1]

excellent tutorial....kudos

[@praveenbhatt3127]

great video. Can someone tell me how he did that animation of different predicted lines in the starting of the video ?

[@AdarshMenon]

I generated images using opencv for all the values and then used premier pro to generate a gif from the images (lots other gif generating tools available online)

[@gchumbes]

Excellent thanks for sharing

[@tejaskulkarni5795]

Can you also give an example of how to use gradient descent to find the best alpha for ridge and lasso regression please.

[@firstkaransingh]

Good explanation thanks bro ☺️

[@turtlepedia5149]

sir if epochs is used for n then i will go beyond its limit so what will be the interpretetion

[@turtlepedia5149]

sorry sir i got it thanks

[@merohan619]

Is it necessary that gradient descent lines will be straight..? If the scatter plot is in form of tan function then?

[@myentropyishigh8708]

gradient is not a line or something its a technique ,its is used here to find optimum values of slope m and constant c its not a line

[@vivekpuurkayastha1580]

Could you please explain what will be the change in the partial derivative equation for multiple predictors?

[@anthonybeaulamaryj5372]

Is there any way to find the accuracy of this manual implementation?

[@hreedishkakoty6771]

Why is your for loop over len(X) and not epochs ?

[@kanokyusuki]

Thanks a lot! it was a great tutorial.

[@sivaramakrishna2248]

Simply awesome ....

[@sankojushivasai8028]

where did the epochs used?

[@AdarshMenon]

Hi epochs were to be used in the for loop instead of n. I made that mistake in the video. Please see the code on GitHub (link in description)

[@datascientist2958]

Can you continue on same data with stochastic gradient descent and mini batch

[@AdarshMenon]

will definitely add it to my list

[@turtlepedia5149]

what is the use of the itertion variable please tell sir the epochs var u didnt used at all

[@AdarshMenon]

Yes correct, i made a small mistake there, and instead of epochs in the loop I used n. The program worked because it was small values and didnt make much difference

[@SahibzadaIrfanUllahNaqshbandi]

Thank you for the great video. I have a question. If I multiply X with -1 then Gradient descent and linear regression fail to obtain negative slope . Can you please tell me whats the problem?

[@AdarshMenon]

I think you would also need to reverse the signs in the derivatives calculated. Try calculating the derivatives with negative x, use those equations to do the regression

[@JainmiahSk]

To find best-fit line should I include Gradient Descent code or sklearn library automatically do that while include Linear Models?

[@AdarshMenon]

Doing it with sklearn is the easies way

[@JainmiahSk]

@@AdarshMenon To clarify doubts how to contact you? whatsapp or Telegram?

[@JainmiahSk]

@@AdarshMenon Thank you.

[@Saimelodies2512]

can you explain the back propagation derivation of a simple neural network? plZZ???? step by step (using matrix calculus)

[@AdarshMenon]

I made an entire series on it - kzbin.info/aero/PLP3ANEJKF1TwHRDS9sPANOzYaAIfJIuam

[@siddhantdutta612]

I g you forgot to use epoch .. instead you used len(X) int the part where we find the value of m and c .. I may be wrong but i just spotted this

[@AdarshMenon]

You are right, I forgot to use the variable epochs. The code works because, n iterations were enough for the code to converge. Thanks for pointing this out!

[@siddhantdutta612]

@@AdarshMenon yaa exactly😅 ... Btw the video was very helpful .. good job.👍 Yaa and one more thing.. it would be even better if you make videos on multivariate regression too...

[@NavneetKaur-ps8bn]

Why am I getting the values of m and c as nan and nan respectively. and also (FloatingPointError: overflow encountered in double_scalar) this error . Also i tried by fixing the values of learning rate and epochs but it won't work .

[@AdarshMenon]

Inside the loop, you can try printing the values of m and c, to see exactly at which iteration it is overflowing. If it is happening from the first iteration itself, maybe a problem with the dataset, otherwise could be learning rate/epoch values

[@tejaltatiwar4682]

Amazing

[@vihaangoyal5134]

NICE AND LEARNATE VIDEO

[@swaroopsasikumar]

Great Video.

[@AdarshMenon]

Thanks!

[@shivangchauhan8609]

Good one, thanks

[@himanshusoni9876]

If the data set have multiple features than gradient descent is applicable or not??? please reply ..... Thank you

[@AdarshMenon]

yes can be applied towardsdatascience.com/machine-learning-bit-by-bit-multivariate-gradient-descent-e198fdd0df85

[@nikhil7129]

sir can we use it when we have multiple independent features

[@AdarshMenon]

Yes you can, use sklearn because it is easier using that

[@ACC861]

If you use a different learning rate.. the results will vary right ?

[@AdarshMenon]

Technically yes, but for such a small dataset I dont think there will be a huge difference. The learning rate determines how fast the function converges, so it is important to experiment with values and choose a suitable one for your use case

[@turtlepedia5149]

Sir I am currently into the Andrew NG course on coursera... Sir after i complete this can you suggest me some advance level course that u can follow after thus course.. As I want to get mor better in ml

[@AdarshMenon]

To get better at ml, after doing the courses and understanding the basics, I would suggest you try and solve real problems. Kaggle would be a good place to start.

[@bhavyanavuluri9177]

Hi , I am getting output as NAN . why?

[@balubalaji9956]

Thanks Adarsh for making me understand 😅

[@ManjeetSingh-ly3fk]

Pls share a code for automatic updation of learning rate in this linear regression with scratch.

[@Viratfanpage369]

pls give the code for the graph that was in the starting of the video

[@Yzyou11]

Thank you bro 👍🙏

[@ahmednabil2181]

6 for i in range(len(x)): ----> 7 y_pred = m*x + c 8 dm = (-2/len(x))*sum(x*(y - y_pred)) 9 dc = (-2/len(x))*sum(y - y_pred) it gives me an error in line number 7 telling me that can only concatenate list (not "int") to list how do i fix that?

[@navinbondade5365]

you did not use iteration or epochs in this code, what is use of it

[@vrstudios9978]

Hii Adarsh if possible can you tell how to present output as animation gif with moving line for every iteration or If anyone knows let me know

[@nikhil7129]

i think their should be any condition to stop when we find optimal value?

[@AdarshMenon]

you can adjust he value of epochs and stop where the loss is approaching 0

[@riyazbagban9190]

where did you use epochs

[@IndiCEO]

Thanks man.........

[@jayaprakashkumar4573]

The output after printing m and c is "nan nan " for me. Plzz help

[@christiankentorasmussen7492]

Super good video :)

[@suritchakraborty9055]

Why am I getting the values of m and c as nan and nan respectively. I didn't have a CSV data set but instead I created two lists and converted them into panda series data type. I don't know where am I going wrong

[@AdarshMenon]

i think you should use the csv or some other real world data. If the data is already the optimum value, gradient descent wont work. you can also experiment with other vales of learning rate and epochs. Maybe also try printing the value of m and c to see how many times the loop run before they become nan

[@suritchakraborty9055]

@@AdarshMenon it worked out needed to fix the learning rate as my data was small thanks a lot

[@sathishkumaranandashekaren4313]

Hi your program doesnt show iteratioins and not clear about global minima

[@kassa1384]

nice video

[@AdarshMenon]

Thanks