Linear Regression with Python in 60 Seconds

Learn about Linear Regression in python in this short!
#python #datascience #shorts

[@jaehuilim7467]

An excellent illustration of Overfitting

[@zabeardybeardy232]

and of a regresion with one variable... Notice how the plot literally tells you that views correlate with the fact that tomorrow is another day.

[@Leisure4682]

Isn't it underfitting? The model is too simplistic for the data points. A second degree equation will work a lot better.

[@starship9629]

Your comment proves you have no idea about machine learning. Estimates for linear regression are by definition unbiased on linear data. However in non-linear data like this, linear regression has high bias because it can only capture linear decision boundaries. Ao the model is in-fact underfittting. Also this comment and the uploader of this video assumes these model are good at extrapolating which is not their intended use at all. In this example, the uploader is trying to model a time-series with a simple linear regression which will only work if the decision boundary is linear which it is not.

[@starship9629]

​@@zabeardybeardy232Yes. Linear models can not fit on non-linear data hence they have low complexity. Low complexity = underfitting

[@jaehuilim7467]

@@starship9629 lol...I guess you are missing the point of this video. Because the provider made this video to demonstrate "Linear regression", we all know that we are only focusing on the general trend in this data. Then, the fitted least squares (I assume!) regression line should pierce through the data points on the plot. To me, a line with a slightly lower slope seems to be more appropriate regression line.

[@asdeEducation]

I hope I remember come back to this video in October

[@robmulla]

Haha. Set a reminder! Hopefully we can beat the forecast!

[@kartikgoswami]

Its November

[@nirjalkumarmahato330]

It's January

[@nirjalkumarmahato330]

It's February 🙃

[@alirezouali3119]

It’s march

[@ferrychandra9934]

Now do this without libraries, except Matplotlib and Pandas 😂

[@David-hl9iv]

With Linear regression you don't necessarily need to get a line. By choosing different functions as your basis elements for the regression, you can fit any curve to the data, be it exponential or a sine curve. Linear Regression looks to minimize the distance of expected outputs at x to outputs that can be represented by the linear combinations of the functions that you are given, at that point.

[@paulantoni1934]

Yeah but then we could just go for polynomial regression.

[@spazeasterix6723]

Although correct in concept, exponentials are rarely practically used. Probably because exponentials are much more sensitive than even polynomial functions and are therefore much more susceptible to overfitting. But as I said, technically, you're implication is correct.

[@1_lloyd_146]

Btw, I rlly love this type of video, so make more videos of linear and nonlinear regression plsssss

[@robmulla]

Thanks for the feedback. Will do.

[@Kashi_Mochibean]

I just found this randomly. But it's the most wonderfully-subtly-obvious and nerdiest way I've ever been asked for views and subscriptions. So I'm hitting the button

[@giraffa-analytics]

Linear regression is a bad choice for this kind of data. This is a time series problem. cumulative sums introduce autocorrelation between datapoints, which violates the assumption of iid residuals of linear regression. Code ok, reasoning not.

[@starship9629]

Exactly

[@NicholasRenotte]

Rob that keyboard cam is 🔥

[@robmulla]

Thanks Nick. This was my old setup where I had an old webcam precariously attached to a mic stand. It would fall over about once an hour 😂 I need to get your opinion on cameras. Recently invested in better lighting but feel like I’m doing something wrong. Your videos are top notch.

[@NicholasRenotte]

@@robmulla LOL i've been there, the ol' jerry rigged camera setup 😂. Elgato Multi-Mount arms are a lifesaver!!

[@thespicycabbage]

you should be using time series

[@Malzie31]

Love these!! I’ve seen two and they are excellent ideas and useful

[@robmulla]

Thanks for the feedback. My first attempts at making shorts. It’s really hard to keep it under a minute.

[@broxxad]

Ey, you should only expect logarithmic regression with your awesome contents.

[@robmulla]

Haha. Thanks!

[@Ligress]

Exponential

[@h44r96]

Well said 👍

[@broxxad]

@@Ligress logarithmic sounds more realistic.

[@Ligress]

@@broxxad Fine, by the way Do you work in Data feild

[@boimchartis]

Nice video bro👍👍

[@robmulla]

Glad you liked it

[@halkkoi1538]

Wow you are so amazing bro

[@robmulla]

Thanks Halk!

[@CaribouDataScience]

You misspelled Ggplot2 😅

[@robmulla]

No, I think I got it right. 😉

[@kadourkadouri3505]

don't you think a nonlinear least squares model would better fit this data ?

[@robmulla]

Great question. It may be!

[@starship9629]

​@@robmullaOf course it is. Your data is clearly non-linear by looking at it. Yet you're trying to fit it with a linear model. And even worse, you are treating it as a time-series which makes no sense as linear regression models can not capture temporal dependencies. You should stop making videos on topics you have no idea about and present yourself as an authority in areas you have no idea about. It's very unprofessional and misleading.

[@yunuskaratepe]

I don't understand the idea behind the keyboard cam but cool video :)

[@bashar9200]

Thank you for such wonderful content!! I learn a lot from your codes

[@robmulla]

Thanks for watching

[@eron1633]

I'm kinda new in python and programming,and which interface did you use in this video?)

[@robmulla]

I'm using jupyter lab. I have an intro video that would be perfect for you here: kzbin.info/www/bejne/a6HJYZKYpbOVodk

[@zabeardybeardy232]

Yes, your views correlate with the fact that the next day rolls over....

[@hasijasanskar]

October -> September :)

[@shivayshakti6575]

👍

[@robmulla]

🙌

[@Satyam1010-N]

Reading regression analysis books 😅 .

[@1_lloyd_146]

Good work, but im curious what is day_count, is it the total of days or sth? Also, I rll want to have the dataset of this if it doesnt bother you

[@robmulla]

Thanks. I don’t have this dataset public but I do have another video on exploratory data analysis where I go over a similar dataset which I make public. Check that one out. I think day count is just a increasing number since the day the channel started.

[@1_lloyd_146]

@@robmulla Tysm! That’s super helpful

[@jensenholm2361]

Wow that’s the most linear dataset I’ve ever seen

[@robmulla]

Not any more :D

[@gustavomaia2252]

Now you have learned that the skicit learn developers know how to do a linear regression 😊

[@viewsfromthechris7810]

Looks like you got way more than the prediction lol!

[@robmulla]

Yep :D

[@bhargavapothakamuri4218]

I am getting value error when I run on my 2D CSV, with 2 columns “date” and “value”, during model.fit(x,y)

[@starship9629]

Pro-tip: Don't learn statistics from this guy. And use OLS instead. That will resolve your problem

[@akankshashrivastava100]

Do you teach the same?

[@Belawithnoalcohol]

This is October, so how is it so far?

[@SayoojGarrix]

83k. So regression miserably failed 😂

[@user-kp9si1km6g]

How to write gta 6 on python, just import gta6 lib

[@robmulla]

Pretty much. Import and add 1.

[@primary4075]

Nah.... You forgot normality test

[@lukas8385]

How to linear regression: step 1 import linear regression -.-

[@robmulla]

Yep! You’ve got the hang of it!

[@gokayfem]

did you get 100k views?

[@robmulla]

Yes I did! As of today (Oct 20) I'm at 387k views. Thanks for watching.

[@bharathraj221]

Filenotfounderror😢

[@starship9629]

Well do you have his dataset on your computer?

[@starship9629]

You need to take this video down. You're misleading people