8000 tweedie loss function · Issue #6867 · microsoft/LightGBM · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content
tweedie loss function #6867
Open
Open
@Alit10

Description

@Alit10

Hello,
I tried to define a custom loss using tweedie.
So I did a first test define grandient and hessian without modification and comparing to the run with objective = 'tweedie' and I have two different results

Any ideas why ?

My code:

grad_tweedie = (-y_true * np.exp((1 - p) * y_pred) + np.exp((2 - p) * y_pred)) * w
hess_tweedie = (-y_true * (1 - p) * np.exp((1 - p) * y_pred) + (2 - p) * np.exp((2 - p) * y_pred)) * w

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions

      0