7.1.5. Implementing Linear Regression from scratch in Python

[@brainsharpener8688]

U are doing great work for the students. I have checked out many playlist but none of them have built the model from scratch 🏅🏅🏅🏅🏅🏅

[@bhaminisekar6707]

Thank you so much. Got good understanding of the concept.

[@saisundarmasetty4283]

how to do linear regression using mini-batch gradient descent how it will be useful?

[@abundanceontheway2511]

Thank you very much for the detailed explanation!!!

[@0xrpj]

In your code, how does the model know that it has reached global minimum? It will just run continuously till the iteration count. What if you set iteration count to 100 and it found the best m and c values on 50th iteration, what then? also, where is the cost function calculated in the code?

[@Siddhardhan]

hi! that's why we use gradient Descent. beyond the global minimum, your parameters won't change much. you can also use a technique called as early stopping which will stop the training process if there's not a considerable change in the loss function, no matter what is the number of iterations we give. I planned to cover those topics in model optimization.

[@0xrpj]

@@Siddhardhan okay thank you :)

[@kasyapdharanikota8570]

@@Siddhardhan your explanations are very good

[@ravipanavi]

When we need to get the dataset

[@srikhanth.r4482]

Nice explanation Sir can u please put an video for MSE and R2

[@cozycorner988]

when we calculated the weight and bias what is the final cost?

[@Siddhardhan]

here, we are not calculating the cost directly. we are trying to minimise it. that's why we are calculating the gradient. if u still want to calculate the final cost, after training the model, calculate y_pred and plug in the values to cost function formula. you will get the value.

[@cozycorner988]

@@Siddhardhan thank you so much for responding but what is the cost function and how to code it

[@vamsibolla4517]

Actually in the previous videos you've used droping and putting in another variable for splitting the features and target values but here you used iloc for the same process ... Why not you use the same old process and you had used transpose in calculating dw...is this the reason for using iloc in this video

[@vamsibolla4517]

I have another doubt why didn't you use sum for dw you used sum for db though both formulas we have to do summation then why not in dw

[@peaceandlov]

same doubt

[@papireddysrmap8703]

@@vamsibolla4517 dot product actually caluclates the sum . for example : dot product of [1,2,3] and [4,5,6] gives 1.4 + 2.5 + 3.6 = ans. but in db there is no Xi . so dot product is not needed but summation is needed.

[@rogue.3715]

The bias is typically a scalar value (a single value) , not a vector or matrix like weights ​@@vamsibolla4517

[@vigneshskcet]

@@papireddysrmap8703 nice

[@rogue.3715]

I got error in model = linear_Regression(learning_rate =0.02, no_of_iterations = 1000) Errors is type error linear regression takes no arguments. Can anyone explain this

[@treesabinoy1043]

TypeError because you've defined the __init__ method incorrectly in your Linear_Regression class. In Python, the __init__ method should have double underscores before and after "init".

[@godisgreat3892]

Really helpful. Thank you.

[@p.v.sakhil9124]

why didnt we use standardization in this case?

[@Siddhardhan]

hi! it won't affect the accuracy in this case. u can definitely add it. same procedure.

[@p.v.sakhil9124]

@@Siddhardhan ok thank you sir

[@deepikat4520]

I'm getting attribute error while in model .fit What to do next

[@saqibaqeel9196]

Check building a model from scratch code were you get one error "selb" you have to correct it as self

[@payalkaur779]

Thanks a ton!!!❤❤❤❤

[@mirosaide8786]

How can we calculate accuracy?

[@prithvichavan1388]

use evaluation metrics such as F1 score ,confusion matrix ,and much more are available

[@gauravfamily2209]

Nice

[@Siddhardhan]

Thanks