We currently have a list of miners. The idea is to evaluate their performance and compare them to the analogues from the PyTorchMetricLearning library. Since miners only require sample IDs and their labels, the performance test will be simple and compact, and can be evaluated on synthetic data. If some of the miners appear to be slow, we should optimize them.
We currently have a list of miners. The idea is to evaluate their performance and compare them to the analogues from the PyTorchMetricLearning library. Since miners only require sample IDs and their labels, the performance test will be simple and compact, and can be evaluated on synthetic data. If some of the miners appear to be slow, we should optimize them.