Classification: objectives and metrics¶
Objectives and metrics¶
The formulas use the variables common to all metrics.
logistic¶
\[
\displaystyle\frac{ - \sum\limits_{i=1}^N 2 w_{i}\left(c_i \log(p_{i}) + (1-c_{i}) \log(1 - p_{i})\right)}{\sum\limits_{i = 1}^{N} w_{i}}
\]
where \(c_i\) is 1 if the i-th object belongs to the second class (classes_[1]) and 0
otherwise. This is the binomial deviance: twice the log loss.
TBoostClassifier uses this objective when the target has exactly two distinct values. The
class labels can be of any type; they are sorted, and classes_[1] is the positive class.
Used for optimization¶
| Name | Optimization | Link |
|---|---|---|
| logistic | + | logit |