Feature selection generally use three main methods, filter and wrapper and embedded method, it can combine now more fire the depth of the study for feature selection, to see some other people think, is mostly characteristic number and less samples to choose more, according to the stand or fall of the final model is derived to add or remove features to choose good characteristics, but how to less samples, and the characteristics of the thousands of tens of thousands of, how to select the characteristic, have Daniel have some ideas, or have the resources to reference, thank you thank you