• Overview
  • Parameter tuning
  • Speeding up the training
  • FAQ
  • Development and contributions
  • Contacts

FilteredDCG

This function is usually used to assess the quality of ranking.

Calculation principles

The calculation of this function consists of the following steps:

  1. Filter out all objects with negative predicted relevancies (aia_i).

  2. The FilteredDCG metric is calculated for each group (groupgroupsgroup \in groups) with filtered objects.

    The calculation principle depends on the specified value of the type and denominator parameters:

    type/denominator LogPosition Position
    Base FilteredDCG(group)=itg(i,group)log2(i+1)FilteredDCG(group) = \sum\limits_{i}\displaystyle\frac{t_{g(i,group)}}{log_{2}(i+1)} FilteredDCG(group)=itg(i,group)iFilteredDCG(group) = \sum\limits_{i}\displaystyle\frac{t_{g(i,group)}}{i}
    Exp FilteredDCG(group)=i2tg(i,group)1log2(i+1)FilteredDCG(group) = \sum\limits_{i}\displaystyle\frac{2^{t_{g(i,group)}} - 1}{log_{2}(i+1)} FilteredDCG(group)=i2tg(i,group)1iFilteredDCG(group) = \sum\limits_{i}\displaystyle\frac{2^{t_{g(i,group)}} - 1}{i}

    tg(i,group)t_{g(i, group)} is the label value for the i-th object in the group after filtering objects with negative predicted relevancies.

  3. The aggregated value of the metric for all groups is calculated as follows:
    FilteredDCG=groupgroupsFilteredDCG(group)groupsFilteredDCG = \frac{\sum\limits_{group \in groups} FilteredDCG(group)}{|groups|}

User-defined parameters

type

Description

Metric calculation principles.

Possible values:

  • Base
  • Exp

Default: Base

denominator

Description

Metric denominator type.

Possible values:

  • LogPosition
  • Position

Default: Position