/sci/ - Science & Math » Thread #13639495

11KiB, 400x265, linear_regression.png

View Same Google iqdb SauceNAO

Anonymous Tue 14 Sep 11:44:06 2021 No.13639495 View Reply Original Report

Quoted By: >>13639504 >>13639507 >>13639508 >>13639565 >>13639566 >>13639580 >>13639603 >>13639608

Why is linear regression such a well-studied phenomenon while the rest of the regression techniques are given less attention? Why do we expect that most real-world data will have linear relationships while that is seldom the case? All my statistic classes so far have been mainly linear regression for the most part and the rest are skimmed through. What is the reason for that?

Anonymous

Anonymous Tue 14 Sep 2021 11:49:16 No.13639504 Report

Quoted By:

>>13639495
It's simple.

Anonymous

Anonymous Tue 14 Sep 2021 11:52:05 No.13639507 Report

Quoted By:

>>13639495
Many physical models used in chemistry/biology can be linearized by algebraic manipulation and I guess they were the primary users of statistics before CS/ML was a thing

Anonymous

Anonymous Tue 14 Sep 2021 11:52:14 No.13639508 Report

Quoted By:

>>13639495
Most college courses in applied math and statistics are designed to give you the tools necessary to get a job (data analyst, engineering etc) and LRs are probably one of the most used technique in those fields.

Anonymous

View Same Google iqdb SauceNAO but my manager.png, 395KiB, 1276x1017

Anonymous Tue 14 Sep 2021 12:11:30 No.13639549 Report

Quoted By: >>13639561 >>13639642 >>13639678

Most people are retarded.

Anonymous

Anonymous Tue 14 Sep 2021 12:20:19 No.13639561 Report

Quoted By:

>>13639549
actually laughed

Anonymous

Anonymous Tue 14 Sep 2021 12:24:20 No.13639565 Report

Quoted By:

>>13639495
It's simple. The idea is that linear regression is always applicable if you just transform your data first. Most machine learning / statistics methods do this: They transform the data to a space where it's linearly separable (or suitable for linear regression), and then they just use a linear model on the transformed data.

Anonymous

Anonymous Tue 14 Sep 2021 12:24:21 No.13639566 Report

Quoted By: >>13639588 >>13639607 >>13639641

>>13639495
a lot of the time its the most powerful predictive model (no joke).

also, modelling non linear relationships is very difficult because non linear can mean like 1000 different things. there are non linear regressions, but most of the time theyre not as useful.

Anonymous

Anonymous Tue 14 Sep 2021 12:30:38 No.13639580 Report

Quoted By:

>>13639495
y = y = a*x + b
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22,1,1)}
linest( dataY, dataX, noForcingThruOrigo, showStatistics ) gives values for a and b
-
--
y = a*x^($D$3) + b
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22^$D$3,1,1)}
linest( dataY, dataX^D3, noForcingThruOrigo, showStatistics ) gives values for a and b
--
--
y = a*x^2 + b*x + c
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22^{1,2},1,1)}
linest( dataY, dataX^{1,2}, noForcingThruOrigo, showStatistics ) gives values for a, b and c
--
-
y = a*x^3 + b*x^2 + c*x + d
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22^{1,2,3},1,1)}
linest( dataY, dataX^{1,2,3}, noForcingThruOrigo, showStatistics ) gives values for a, b, c and d
-

y = a*ln(x) + b
{=LINEST(rawDataY.B1:B22,LN(rawDataX.A1:A22),1,1)}
linest( dataY, LN(dataX), noForcingThruOrigo, showStatistics ) gives values for a and b

y = b * x^a
{=LINEST(LN(rawDataY.B1:B22),LN(rawDataX.A1:A22),1,1)}
linest( LN(dataY), LN(dataX), noForcingThruOrigo, showStatistics ) gives values for a and ln(b)
--
--
y = b * a^x
{=LINEST(LN(rawDataY.B1:B22),rawDataX.A1:A22,1,1)}
linest( LN(dataY), dataX, noForcingThruOrigo, showStatistics ) gives values for ln(a) and ln(b)
--
-
y = b * e^(a*x)
{=LINEST(LN(rawDataY.B1:B22),rawDataX.A1:A22,1,1)}
linest( LN(dataY), dataX, noForcingThruOrigo, showStatistics ) gives values for a and ln(b)

Anonymous

Anonymous Tue 14 Sep 2021 12:32:39 No.13639588 Report

Quoted By: >>13639602

>>13639566
>there are non linear regressions, but most of the time theyre not as useful.
What?
Please show me how to use linear regression for image recognition, say.

Anonymous

Anonymous Tue 14 Sep 2021 12:38:58 No.13639602 Report

Quoted By: >>13639605

>>13639588
>take image
>down scale data
>regression model
> reconstruct image
>train NN to minimize reconstruction error
>train NN2 for image recognition
done

Anonymous

Anonymous Tue 14 Sep 2021 12:39:14 No.13639603 Report

Quoted By:

>>13639495
y = a*x + b
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22,1,1)}
linest( dataY, dataX, noForcingThruOrigo, showStatistics ) gives values for a and b
-
--
y = a*x^($D$3) + b
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22^$D$3,1,1)}
linest( dataY, dataX^D3, noForcingThruOrigo, showStatistics ) gives values for a and b
--
--
y = a*x^2 + b*x + c
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22^{1,2},1,1)}
linest( dataY, dataX^{1,2}, noForcingThruOrigo, showStatistics ) gives values for a, b and c
--
-
y = a*x^3 + b*x^2 + c*x + d
{=LINEST(rawDataY.B1:B22,rawDataX.A1:A22^{1,2,3},1,1)}
linest( dataY, dataX^{1,2,3}, noForcingThruOrigo, showStatistics ) gives values for a, b, c and d
-

y = a*ln(x) + b
{=LINEST(rawDataY.B1:B22,LN(rawDataX.A1:A22),1,1)}
linest( dataY, LN(dataX), noForcingThruOrigo, showStatistics ) gives values for a and b

y = b * x^a
{=LINEST(LN(rawDataY.B1:B22),LN(rawDataX.A1:A22),1,1)}
linest( LN(dataY), LN(dataX), noForcingThruOrigo, showStatistics ) gives values for a and ln(b)
--
--
y = b * a^x
{=LINEST(LN(rawDataY.B1:B22),rawDataX.A1:A22,1,1)}
linest( LN(dataY), dataX, noForcingThruOrigo, showStatistics ) gives values for ln(a) and ln(b)
--
-
y = b * e^(a*x)
{=LINEST(LN(rawDataY.B1:B22),rawDataX.A1:A22,1,1)}
linest( LN(dataY), dataX, noForcingThruOrigo, showStatistics ) gives values for a and ln(b)

Anonymous

Anonymous Tue 14 Sep 2021 12:40:30 No.13639605 Report

Quoted By: >>13639607 >>13639615

>>13639602
>>train NN2 for image recognition
Ah, so you're using a non linear regression.
Don't know why you had to do all the random shit before it though.

Anonymous

Anonymous Tue 14 Sep 2021 12:42:35 No.13639607 Report

Quoted By: >>13639631

>>13639605
Youre using the second NN on the linearly regressed data, my fren
Otherwise I misunderstood your question
Also, I'm not >>13639566

Anonymous

Anonymous Tue 14 Sep 2021 12:42:51 No.13639608 Report

Quoted By:

>>13639495
if you have exponential data just logarithm it so you can you linear regression brainlet

Anonymous

Anonymous Tue 14 Sep 2021 12:45:35 No.13639615 Report

Quoted By: >>13639631

>>13639605
Also
I just googled "linear regression image classification" then described this paper https://ieeexplore.ieee.org/document/8014822

Anonymous

Anonymous Tue 14 Sep 2021 12:49:21 No.13639631 Report

Quoted By: >>13639641

>>13639607
>>13639615
Yes, but I don't understand your point. Sure, you're using linear regression in the beginning, but in the end you use a neural network which is non linear regression. So the entire process is nonlinear, and you could've just used a neural network from the start.

Anonymous

Anonymous Tue 14 Sep 2021 12:54:57 No.13639641 Report

Quoted By: >>13639704

>>13639631
Then I dont understand why you are bringing up image classification when anon said (>>13639566)
>a lot of the time its the most powerful predictive model (no joke).

If the best way to do image recognition is a NN, and NNs can't be done with linear regression, then your original question was... pointless

Anonymous

Anonymous Tue 14 Sep 2021 12:54:58 No.13639642 Report

Quoted By:

>>13639549
Retards are fine, but the fact it's one of the most popular regression questions. Kek

Anonymous

Anonymous Tue 14 Sep 2021 13:04:15 No.13639678 Report

Quoted By:

>>13639549
The manager can't be this, can he?

Anonymous

Anonymous Tue 14 Sep 2021 13:14:53 No.13639704 Report

Quoted By: >>13639805

>>13639641
No, it's not pointless. My point is that most problems are nonlinear. Image recognition is just a widely popular example. Basically all of computer vision and natural language processing is nonlinear.

Anonymous

Anonymous Tue 14 Sep 2021 13:52:27 No.13639805 Report

Quoted By:

>>13639704
>My point is that most problems are nonlinear.
Which is what he said
He said "a lot of the time". Not most of the time
Sounds like you are sperging out
Regardless, your question and this discussion are pointless
Not worth my time buddyboy

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches