search

datatrigger / asymmetric_loss

Implement a custom asymmetric loss to train and drive a regression model towards underestimation or overestimation
https://blog.vlgdata.io/post/asymmetric_loss/
MIT License
4 stars 2 forks source link

do you have example code for linear regression with LINEX lost #1

Open Sandy4321 opened 2 years ago

Sandy4321 commented 2 years ago

great code but for one hot data xgboost is not perforaming

do you have example code for linear regression with LINEX lost

Sandy4321 commented 2 years ago

as stated in https://en.wikipedia.org/wiki/Conjugate_gradient_method The conjugate gradient method is often implemented as an iterative algorithm, applicable to sparse systems that are too large to be handled by a direct implementation

datatrigger commented 2 years ago

Hi @Sandy4321, I did not implemented linear regression with the linear-exponential loss at the moment. That's a great idea though, thanks for you suggestion!

Sandy4321 commented 2 years ago

thanks for soon answer meantime what first derivative maybe for linear regression coefficients ?

Sandy4321 commented 2 years ago

meaning linear-exponential loss for

y_pred = x1 w1 + x2w2 +x3*w3 ...etc

where x1, x2, x3 observations w1, w2, w2 - coefficients for liner regression ? so we want to find w1,w2,w3 for given data

x11 x12 x13 x21 x22 x23 x31 x32 x33 etc xn1 xn2 xn3

Sandy4321 commented 2 years ago

then image

is it correct ?