I don’t comment often videos but this one is really good, you explained the math concept really well by doing the examples on the graph etc… Good stuff in this channel !
this video is amazing, good job, im now actually thinking about re visiting the math classes i couldn't take before in order to get better at these machine learning algorithims.
I am subscribing, definitely. You have taught in the most straight forward and explanatory way a concept that other videos made a bit too complicated. Thank you!
The way you explained and implemented was really efficient and easy, it was really helpful as I am just getting started on my ML journey and after implementing this I feel m a step forward. Thank You!!
Thanks man! Awesome video that dumbed it down enough for me. Could you do linear regression of fitting a sphere to points given the sphere radius (or not) like is used in terrestrial laser scanning? If you've never messed with it, you place spheres around what you are scanning, then the software uses the spheres to align all of the different scans at different locations together.
A really insightful video to learn this concept!! I found it very helpful !! Just one suggestion @NeuralNine, please include the files (that you display in the video) in the description of the video. Everything else is amazing and praiseworthy!
Awesome tutorial! Could you please explain why you use gradient descent to minimize squared error instead of using the formula: divide the standard deviation of y values by the standard deviation of x values and then multiply this by the correlation between x and y?
Damn !! I like your video so much, especially the implementation part. Most of the people would just directly use sklearn library and call it a day. But, you really have showed how the code actually runs behind the scene. 👍👍💛💛
hi , really you have the best videos in youtube. Your speaking is also very clear for me although my native language is not English. Thanks you very much it helps me very much , and i am in 17 years old.I want to improve and in my this way you help me.
If you're counting from 0 to n, it means you have (n+1) data points, so should the dividing factor not be 1/(n+1) while calculating mean of squared errors at 4:43?
I implemented it myself and it came out to be more accurate then sckit-learn import numpy as np from sympy import Symbol, solve, diff class LinearRegression: def __init__(self): self.w = {} def fit(self, x, y): for i in np.arange(x.shape[1] + 1): self.w[f"w{i}"] = Symbol(f"w{i}") e = 0 for i in range(len(x)): yp = 0 for j in np.arange(len(self.w)): if j == 0: yp += self.w[f"w{j}"] else: yp += self.w[f"w{j}"] * x[i][j-1] e += (yp-y[i]) ** 2 eq = [] for i in np.arange(len(self.w)): eq.append(diff(e, self.w[f"w{i}"])) w = solve(eq, list(self.w.keys())) for i in np.arange(len(self.w)): self.w[f"w{i}"] = w[self.w[f"w{i}"]] def predict(self, x): def prediction(features): yp = 0 for i in np.arange(len(self.w)): if i == 0: yp += self.w[f"w{i}"] else: yp += self.w[f"w{i}"] * features[i-1] return yp return list(map(prediction, x))
I have a question regarding the expression (-2 / n). Shouldn't it be placed outside the for loop since it is calculated multiple times, even though it's a constant that should be computed once for the gradient?🙂
Why does 1/n become -2/n. As I read in towards data science that it would be 1/n and then the -2 would be after the sum symbol as 2x'i' so like why or how does 1/n become -2/n. Also great video btw, helped me unlock a few skills in the ml skill tree :)
I dont understand the reasoning behind [m * x * b for x in range(20, 80)] why is this the y axis. What does it represent. Why the equation? Also why arent we just reading the found values of x and y? I have understood all the calculations being done but I dont understand the visualization/graph representation part. Please let me know if you see this.
Lmao, I finished summer school for algebra 1 and we just learned this and to calculate it, we used a calculator this is a amazing video the mathematics where correct and your explanation is amazing!
Excellent video, although I will need to scrub up on my mathematics!! I think I am one of the few that loves looking at formulas and wonders; "What in the hell is going on there?" 😆😆
great video NeuralNine I feel this the gradiant descent. is not it ? if It is, is there an implementation for the least square because I feel it is just so random you take some values and you choose the least one from those random values not necessarily the least value you can get
I am getting error AttributeError Traceback (most recent call last) Cell In[10], line 39 37 if i % 50 == 0: 38 print(f"Epoch: {i}") ---> 39 m, b = gradient_descent(m, b, data, L) 41 print(m, b) 43 plt.scatter(data.studytime, data.score, color = "black")
Hello, do you have a video or notes on plotting the linear regression when attributes in the input data points are more than one. Say 10 columns of x, and 1 column of y?? Please respond. Thank you.
@@sairekhaunnam3001 Hey, if there is one attribute the we can plot it on 2D, if two attributes, then in 3D, and for three attribute, we will plot in 4D, which is not possible visually. That's why we restrict ourselves.
Hi sir , it's really helps me to understand linear regression and build it by myself from scratch. Thanks 👍 can you describe all the models from scratch? Like it's
What's the point of printing Epochs if we see nothing on the screen in that regard lol, why is it highly mathematical if we do not even derive anything? I would redo the video with loss_function being in the print, o/w it just hangs in the code for God knows what reason.
The minima of one of the partial derivatives wont necessarily be the minima for the loss function since it also depends on the other variables/features
as of it own this wont be any useful but as we keep learining we can def find some more models which used in predicting some many useful things but as of this right now it is not useful
there is a improvement need to be done sir while discussing about error you are constantly saying the function as mean squared error right! but this is not the error function the function which need to minimize is the cost function. loss or error function measure the error of single training set while cost function is the summation of all the error of the training set.
@NeuralNine How do I test the trained models to give a prediction based on a newly given value/independent variable (a value that is foreign to the training set)? Great video btw!
Please do more mathematical implementations like this please! This really helped me to understand the math behind these algorithms
😭😭
True. Nobody explains like he does.
I don’t comment often videos but this one is really good, you explained the math concept really well by doing the examples on the graph etc… Good stuff in this channel !
I do like all your videos that contain the keywords "mathematical theory" and "from scratch" :). Please do more similar videos. Thank you kindly
same, stuff from scratch is the best stuff for learning
@@samas69420 yeah, and it's really fun
I am always so proud when I am able to code something from scratch
this video is amazing, good job, im now actually thinking about re visiting the math classes i couldn't take before in order to get better at these machine learning algorithims.
AMAZING video!!! I love these videos which teach theory too! Thank youuu!!!
dude Slammer video!! love the fact that you made the math interesting and super easy to understand
I recently stumbled upon this while looking for similar approach via python and I subscribed you now.
Thank you very much for imparting knowledge.
That was just wow! The way you explained it was amazing.
Thank you!
I was searching everywhere and final found what I need. Your video really clears up the fundamentals of creating Linear Regression Model. Thank you
Thanks very much for this!! I am a data analysis student and close to giving up but still hanging on!
@kelvinthomas-pr8sd Thank you! I did then but it still apllies now. I now work but have 2 classes I haven't passed yet. Never give up!
I am subscribing, definitely. You have taught in the most straight forward and explanatory way a concept that other videos made a bit too complicated. Thank you!
The way you explained and implemented was really efficient and easy, it was really helpful as I am just getting started on my ML journey and after implementing this I feel m a step forward. Thank You!!
on 10:00 we need to take derivative of each weight (M) like this, first for M1 then for M2 then M3.... to Mn
Greatly explained in simple words, looking forward to learn more methods from you.
Thank you so much. Im working through problem set for a Neural Analysis class and this really helps. Great video.
Amazing video, looking forward for more implementations from scratch!
Thanks man! Awesome video that dumbed it down enough for me. Could you do linear regression of fitting a sphere to points given the sphere radius (or not) like is used in terrestrial laser scanning? If you've never messed with it, you place spheres around what you are scanning, then the software uses the spheres to align all of the different scans at different locations together.
A really insightful video to learn this concept!! I found it very helpful !!
Just one suggestion @NeuralNine, please include the files (that you display in the video) in the description of the video.
Everything else is amazing and praiseworthy!
Easy to follow and goes straight to the point, thank you!
Awesome tutorial! Could you please explain why you use gradient descent to minimize squared error instead of using the formula: divide the standard deviation of y values by the standard deviation of x values and then multiply this by the correlation between x and y?
Damn !! I like your video so much, especially the implementation part. Most of the people would just directly use sklearn library and call it a day. But, you really have showed how the code actually runs behind the scene. 👍👍💛💛
It would have been nice to compare it to the analytical solution of least squares regression being (Xᵀ•X)⁻¹•(Xᵀ•Y) just to show they're identical
quicker and theoretically more correct for solving this problem!
@@andreagalloni92 could you make a video and share it with everyone please? (theory and python code)
@@JohnMacPherson-hr4yz mmm... Maybe I can :)
I'll try to do it
hi , really you have the best videos in youtube. Your speaking is also very clear for me although my native language is not English. Thanks you very much it helps me very much , and i am in 17 years old.I want to improve and in my this way you help me.
Excellent explanation of the implementation of linear regression, thanks!!
This is a great video! Thank you so much!
Waiting for series of such videos 🤩
Yes, please do more videos like this one. Even Einstein gives it 2 thumbs up! :)
You got a new subscriber... Best in detail explanation ever!!!
I really appreciate the time and the effort to simplify such a thing, Thanks a bunch 🌹🌹🌹🌹
Thank you so much man! Great vid...Keep up the good work!
If you're counting from 0 to n, it means you have (n+1) data points, so should the dividing factor not be 1/(n+1) while calculating mean of squared errors at 4:43?
U are right, it should be like that. He just made an accidental mistake.
Next for logistic regression please!
love the videos, finally
Thank youuu :) Loved the way you explained it.
Bruder, danke dir. Es wollte nicht in meinen Kopf. Dank dir habe ich das verstanden.
Brother! I simply love you after I came across this video!
Amazing simple explanation of linear regression. Please also cover the other techniques in same way
I implemented it myself and it came out to be more accurate then sckit-learn
import numpy as np
from sympy import Symbol, solve, diff
class LinearRegression:
def __init__(self):
self.w = {}
def fit(self, x, y):
for i in np.arange(x.shape[1] + 1):
self.w[f"w{i}"] = Symbol(f"w{i}")
e = 0
for i in range(len(x)):
yp = 0
for j in np.arange(len(self.w)):
if j == 0:
yp += self.w[f"w{j}"]
else:
yp += self.w[f"w{j}"] * x[i][j-1]
e += (yp-y[i]) ** 2
eq = []
for i in np.arange(len(self.w)):
eq.append(diff(e, self.w[f"w{i}"]))
w = solve(eq, list(self.w.keys()))
for i in np.arange(len(self.w)):
self.w[f"w{i}"] = w[self.w[f"w{i}"]]
def predict(self, x):
def prediction(features):
yp = 0
for i in np.arange(len(self.w)):
if i == 0:
yp += self.w[f"w{i}"]
else:
yp += self.w[f"w{i}"] * features[i-1]
return yp
return list(map(prediction, x))
dude did my 6 hours lecture + lab class all in just 24 minutes, bruh. Ez Clap hahah, thanks a lot.
The flow should terminate on global minima right ?? Is this program complete ??
your explanations are really clear and precise, thank you very much♥♥♥
Loved it .......Thank you so much ❤❤
Glad you enjoyed it!
I have a question regarding the expression (-2 / n). Shouldn't it be placed outside the for loop since it is calculated multiple times, even though it's a constant that should be computed once for the gradient?🙂
the math part as awesome! thanks a lot very clear and simple
yoo out all explanations i saw u did the best work this is what i wanted best work man subbed
got a good idea about linear regression thnx bud!!
That was so much easier to understand. thank you.
Excellent explanation. Thanks
Nicely done!
awesome video and dark mode paint rocksss
Awesome representation!!!
Why does 1/n become -2/n. As I read in towards data science that it would be 1/n and then the -2 would be after the sum symbol as 2x'i' so like why or how does 1/n become -2/n. Also great video btw, helped me unlock a few skills in the ml skill tree :)
well done
You’re a life saver thank you soooo much❤❤
Thanks a lot for great explanation!
Great video, very informative
This is amazing!
How can I store the current m and b values for every iteration?
loved it bro u r just amazing
I dont understand the reasoning behind [m * x * b for x in range(20, 80)] why is this the y axis. What does it represent. Why the equation? Also why arent we just reading the found values of x and y? I have understood all the calculations being done but I dont understand the visualization/graph representation part. Please let me know if you see this.
He did it all wrong just ignore it imo
Great video!
Combine theory with practice，and show the calculate procedure，really good👍
Nice video! Could you please provide the dataset that you used
yup
Lmao, I finished summer school for algebra 1 and we just learned this and to calculate it, we used a calculator this is a amazing video the mathematics where correct and your explanation is amazing!
Excellent video, although I will need to scrub up on my mathematics!! I think I am one of the few that loves looking at formulas and wonders; "What in the hell is going on there?" 😆😆
great video NeuralNine
I feel this the gradiant descent. is not it ?
if It is, is there an implementation for the least square
because I feel it is just so random you take some values and you choose the least one from those random values not necessarily the least value you can get
wow, that is exactly what is called simplicity
I am getting error
AttributeError Traceback (most recent call last)
Cell In[10], line 39
37 if i % 50 == 0:
38 print(f"Epoch: {i}")
---> 39 m, b = gradient_descent(m, b, data, L)
41 print(m, b)
43 plt.scatter(data.studytime, data.score, color = "black")
Thank you, very useful and clear.
Why the gradient descent method? The pseudo inverse solves this immediately
Make a video about it and link me, plz
Where can we get the csv file he worked on?
where can we find the data you uploaded. I would be happy to try the code. thanks for making these videos.
bro how does the model created by these mathematical output and how does the predict function works
;
here to learn something that I already know but the video is too entertaining to click off
Hello, do you have a video or notes on plotting the linear regression when attributes in the input data points are more than one. Say 10 columns of x, and 1 column of y??
Please respond.
Thank you.
I too need explanation on this. It will be really helpful
@@sairekhaunnam3001 Hey, if there is one attribute the we can plot it on 2D, if two attributes, then in 3D, and for three attribute, we will plot in 4D, which is not possible visually. That's why we restrict ourselves.
Hi sir , it's really helps me to understand linear regression and build it by myself from scratch. Thanks 👍 can you describe all the models from scratch? Like it's
that's pretty good lecture.. 😍
why did we do a partial derivative?
This was really helpful
amazing crystal clear
this is the real algorithm. great :)
Which software you use for the video ?
where can I get the dataset? This is all new to me and I can't figure out how to get a test script going on my end, I keep getting dead ends
you can just ask chatgpt to generate you a dataset. It will even convert it into csv file
the defined lossfunction is not called anywhere ,why
What's the point of printing Epochs if we see nothing on the screen in that regard lol, why is it highly mathematical if we do not even derive anything? I would redo the video with loss_function being in the print, o/w it just hangs in the code for God knows what reason.
please provide the dataset which you use bcz it will us to follow along with you
Hi, I really like your videos. One question, what is h?
Thank you!
Amazing video!!!!
I love this guy
Why did you decrease the no of epocs ? And doesn't minima can be found with partial derivative =0 ?
The minima of one of the partial derivatives wont necessarily be the minima for the loss function since it also depends on the other variables/features
amazing content
Great explanation
Upload more model implementation plz
Can Anyone tell me some practical usecase or some example where we can implement this? It should be great if anyone can give me full scenario
as of it own this wont be any useful but as we keep learining we can def find some more models which used in predicting some many useful things but as of this right now it is not useful
can you do the same for Support vector machine
Amazing explaining, thanks a lot!
Can you upload the csv file and send me the link, please ?
there is a improvement need to be done sir while discussing about error you are constantly saying the function as mean squared error right! but this is not the error function the function which need to minimize is the cost function. loss or error function measure the error of single training set while cost function is the summation of all the error of the training set.
You sound like tech with Tim😂😂😂
@NeuralNine How do I test the trained models to give a prediction based on a newly given value/independent variable (a value that is foreign to the training set)?
Great video btw!
Yes! This is a good question I hope he answers. If he already has can someone link me please? :)
where is the csv file?
Lstm ,rnn , logistics and more we are expecting more from you
You should not to "think" that it is a best line, you should verify it!