Optimally Quantized and Smoothed Histograms.

摘要: We propose an approach using optimal quantization and smoothing to generate adaptive histograms for multi-class one dimensional data. The discretization of data is in maximizing a quantizer performance measure, defined by combination average log likelihood, entropy correct classification probability. partition found dynamic programming. density each bin obtained technique that can be considered generalized k nearest neighbor estimation algorithm. However, our much more efficient. Experimental results demonstrated the effectiveness optimally quantized smoothed histograms. Even though obtaining takes about quadratic time sample size, histogram efficient use than typical kernel methods. Therefore, are suitable applications with massive set.