Feature

This issue proposes adding a new metric 'HarmonicMean' to ignite.metrics.

The harmonic mean is defined as:

H = n / sum(1 / x_i)

where x_i are the individual values.

This metric can be useful in evaluation scenarios where the harmonic mean of values is required. While Ignite provides several mean-based metrics, there is currently no general-purpose harmonic mean metric.

Proposed implementation

• A new metric class HarmonicMean in ignite.metrics
• Supports device placement
• Compatible with distributed training via sync_all_reduce
• Includes tests and documentation

Example usage:

metric = HarmonicMean()
metric.attach(evaluator, "harmonic_mean")