Decision Tree Regression Clearly Explained! 

u kinda bad tho

8:24 He explains that the algorithm compares every possible split and finds the one with the best variance reduction. So the y in X0

​@@onetapmanbbrv

The weights, I think, it's just a fraction of items for each side. For example, 2 items on left and 6 items on right gives .25 and 0.75 weights respectively.

This condition is, indeed, true. You may have confused x0 and x1: x0 = 16 but x1 = -2. Here, we are talking about x1

Thanks!

Actually, we try every possible value of the threshold and find which one produces the best split. If can go through the code for a better understanding. github.com/Suji04/ML_from_Scratch/blob/master/decision%20tree%20regression.ipynb

@@NormalizedNerd Thanks buddy :)

@@NormalizedNerd you have done well but plz first explain taking an example by showing every steps from first to the last with every maths used and the computation on how we get the results and why this value and not the other etc. plz teach like this so that every learner even the one who don't have basic can understand it is a request

Thanks! Well, I use Manim (a python library)

MSE is used if both input and output is continuous ,variance is for discrete input n continuous output

Do not confuse the variance reduction with the variance of each split. VarR1 < VarR2, hence the VarR1 split reduces more variance from the parent node (1 - VarR1 > 1 - Var2)

i think that's the depth of the tree you want, that you need to find by trying out yourself. you want to stop before variance becomes 0, as then the prediction really goes wild

Suggestion noted...

😁❤

By checking every possible value of a feature as the threshold and splitting the dataset based on that. Then taking that particular feature and the corresponding threshold that gives the maximum information gain. Please see the code provided in the next video for more clarity.

You're most welcome...Keep supporting!

Ha bhai :))

@@NormalizedNerd bengoli accent ftw!

I am not sure if u should use MSE for every split

Yeah!

@@NormalizedNerd nice

Thanks!! More to come :)

Most welcome

It depends on the problem you are trying to solve. As a thumb rule: if the target variable takes continuous values then go for regression and if it takes discrete (and few) values then go for classification.

@@NormalizedNerd I understand, thanks a lot :)

It's not O(m*k) but exactly m*k computations at every split.

@@p337maB sure

"but what is determining the other filters from the initial filter?" The initial filter (at root) divides the data into two sets. The left one is then again divided so is the right one. We do this process recursively. While splitting a set we choose the condition that maximizes variance reduction. Please see the implementation to get more clarity: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-P2ZB8c5Ha1Q.html

@@NormalizedNerd So are you saying that x sub 0 and x sub 1 are two sets of decision sets? Rather, two collections of boundaries? Something still looks off. If it's an array of decision boundaries, how do you jump from 1 (of x sub 0) to -7 and -12 (of x sub 1)? I've even tried to figure out the symmetry in the tree to find logic. 4 elements of x sub 1 3 elements of x sub 0. 20 partitions in the grid for 20 elements in the set. I've watched several of your videos trying to understand your message. Explain this one where it makes sense and I'll definitely continue to watch your other content. I try to give everybody that says they're providing "knowledge" a chance. This is outstandingly bonkers to me. I'm also a programmer and MBA