官术网_书友最值得收藏!

Ensemble learning random forest

One-sentence explanation for LOTR fans: if decision trees were Ents, the random forest would be an Entmoot. For everyone else, random forest algorithm works like this:

  • Split data into random subsets of equal size, maybe with replacement
  • On each of those subsets, build a decision tree, choosing for every split a random feature subset of fixed size
  • To perform inference, perform a voting among the trees (classification), or average their predictions (regression)

Such tree ensembles are very popular in certain domains, because their prediction quality beats most other models.

Most likely, this is not the model you want to train on a mobile device, due to the memory and time limitations, but you can still use it for inference thanks to Core ML. The workflow looks like this:

  • Pre-train random forest in scikit-learn
  • Export the model in the scikit-learn format
  • Convert it to the Apple mlmodel format with the help of the coremltool Python package
  • Import it in your iOS project using Core ML framework

By the way, if you look at the inner structure of the GameplayKit's tree learner in a debugger or playground, you'll see that it also uses random forest under the hood.

主站蜘蛛池模板: 偏关县| 南岸区| 五寨县| 新平| 大同市| 云梦县| 黔西县| 泰安市| 秦皇岛市| 巴中市| 德清县| 陵水| 哈巴河县| 准格尔旗| 土默特右旗| 铁岭县| 吴堡县| 合山市| 东乌| 新余市| 巴马| 无锡市| 特克斯县| 仁怀市| 洮南市| 巨野县| 乡城县| 富川| 扬中市| 南丰县| 民县| 玉环县| 浮梁县| 上林县| 昌吉市| 新宾| 吕梁市| 遵义市| 鄱阳县| 大石桥市| 康马县|