About ranking metric evaluation

[vincenttsai2015]

Hi,

I'm wondering if it is possible to evaluate the ranking metrics of MAP(mean average precision) @ K and HR(hit ratio) @ K of GBMF/GBGCN under librecframework. If yes, how can I modify the code? Thanks.

[Sweetnow]

You can modify the framework https://github.com/Sweetnow/librecframework/blob/master/librecframework/metric.py.

Just follow the current code,

• create a new class derived from Metric
• implement __call__ (score is the model output, ground_truth is the corresponding data from Dataset.__getitem__
• plus the metric into self._sum
• plus the count into self._cnt
• register the metric class into _ALL_METRICS

And then, you can choose the new metric in config.json by its class name.

[Sweetnow]

BTW, I think HR is the same as Precision.