MultiLabel Classification: objectives and metrics

Objectives and metricsObjectives and metrics

MultiLoglossMultiLogloss

$\displaystyle\frac{-\sum\limits_{j=0}^{M-1} \sum\limits_{i=1}^{N} w_{i} (c_{ij} \log p_{ij} + (1-c_{ij}) \log (1 - p_{ij}) )}{M\sum\limits_{i=1}^{N}w_{i}} { ,}$

where $p_{ij} = \sigma(a_{ij}) = \frac{e^{a_{ij}}}{1 + e^{a_{ij}}}$ and $c_{ij} \in {0, 1}$

Usage information See more.

User-defined parameters

use_weights

Use object/group weights to calculate metrics if the specified value is true and set all weights to 1 regardless of the input data if the specified value is false.

Default: true

MultiCrossEntropyMultiCrossEntropy

$\displaystyle\frac{-\sum\limits_{j=0}^{M-1} \sum\limits_{i=1}^{N} w_{i} (t_{ij} \log p_{ij} + (1-t_{ij}) \log (1 - p_{ij}) )}{M\sum\limits_{i=1}^{N}w_{i}} { ,}$

where $p_{ij} = \sigma(a_{ij}) = \frac{e^{a_{ij}}}{1 + e^{a_{ij}}}$ and $t_{ij} \in [0, 1]$

Usage information See more.

User-defined parameters

use_weights

Use object/group weights to calculate metrics if the specified value is true and set all weights to 1 regardless of the input data if the specified value is false.

Default: true

PrecisionPrecision

This function is calculated separately for each class k numbered from 0 to M – 1.

$\frac{TP}{TP + FP}$

Can't be used for optimization. See more.

User-defined parameters

use_weights

Use object/group weights to calculate metrics if the specified value is true and set all weights to 1 regardless of the input data if the specified value is false.

Default: true

RecallRecall

This function is calculated separately for each class k numbered from 0 to M – 1.

$\frac{TP}{TP+FN}$

Can't be used for optimization. See more.

User-defined parameters

use_weights

Use object/group weights to calculate metrics if the specified value is true and set all weights to 1 regardless of the input data if the specified value is false.

Default: true

FF

This function is calculated separately for each class k numbered from 0 to M – 1.

$(1 + \beta^2) \cdot \frac{Precision * Recall}{(\beta^2 \cdot Precision) + Recall}$

Can't be used for optimization. See more.

User-defined parameters

beta

The $\beta$ parameter of the F metric.

Valid values are real numbers in the following range: $(0; +\infty)$.

Default: This parameter is obligatory (the default value is not defined)

use_weights

Use object/group weights to calculate metrics if the specified value is true and set all weights to 1 regardless of the input data if the specified value is false.

Default: true

F1F1

This function is calculated separately for each class k numbered from 0 to M – 1.

$2 \frac{Precision * Recall}{Precision + Recall}$

Can't be used for optimization. See more.

User-defined parameters

use_weights

Use object/group weights to calculate metrics if the specified value is true and set all weights to 1 regardless of the input data if the specified value is false.

Default: true

AccuracyAccuracy

The formula depends on the value of the $type$ parameter:

ClassicClassic

$\displaystyle\frac{\sum\limits_{i=1}^{N}w_{i} \prod\limits_{j=0}^{M-1} [[p_{ij} > 0.5]==t_{ij}]}{\sum\limits_{i=1}^{N}w_{i}} { , }$

where $p_{ij} = \sigma(a_{ij}) = \frac{e^{a_{ij}}}{1 + e^{a_{ij}}}$

PerClassPerClass

This function is calculated separately for each class k numbered from 0 to M – 1.

$\frac{TP + TN}{\sum\limits_{i=1}^{N} w_{i}}$

Can't be used for optimization. See more.

User-defined parameters

type

The type of calculated accuracy.

Default: Classic.
Possible values: Classic, PerClass.

HammingLossHammingLoss

$\displaystyle\frac{\sum\limits_{j=0}^{M-1} \sum\limits_{i = 1}^{N} w_{i} [[p_{ij} > 0.5] == t_{ij}]]}{M \sum\limits_{i=1}^{N} w_{i}}$

Can't be used for optimization. See more.

User-defined parameters

use_weights

Use object/group weights to calculate metrics if the specified value is true and set all weights to 1 regardless of the input data if the specified value is false.

Default: true

Used for optimizationUsed for optimization

Name Optimization GPU Support
MultiLogloss + +
MultiCrossEntropy + +
Precision - -
Recall - -
F - -
F1 - -
Accuracy - -
HammingLoss - -