AutoMl

Do we Need Hundreds of Classifiers to Solve Real World Classification Problems? http://www.jmlr.org/papers/volume15/delgado14a/delgado14a.pdf autoML( aotumated Machine Learning) https://github.com/automl https://github.com/rhiever/tpot Data Science Machine 参考论文： auto-sklearn:《Efficient and Robust Automated Machine Learning》，偏框架介绍。github: auto-weka: 《Auto-WEKA: Combined Selection and Hyperparameter Optimization of Classification Algorithms》，偏算法介绍SMAC,TPE,SMBO等。 BOA: 《The Bayesian Optimization Algorithm》 主要完成的功能有2个，也就是是一个CASH problem (Combined Algorithm Selection and Hyperparameter optimization)： Model/Algorithm Selection : It is well known that ensembles often outperform individual models ，模型/算法选择 Hyperparameter Optimization : 模型超参数优化 整体过程： meta-learning warmstart: Meta-learning for finding good instantiations of machine learning frameworks. 超参数优化（Algorithms for Hyper-Parameter Optimization）用到的算法： TPE: tree parzen estimator SMAC： random forest based 《Sequential model-based optimization for general algorithm configuration》 SMBO： bayesian optimization 主要用调优器来选择参数，输入参数为不同hyperparameter，Loss function为准确率等，调优器会在随机选择一些值的基础上，利用贪心算法去寻优。 Meta-Learning Basic Ideas: Domain experts derive knowledge from previous tasks: They learn about the performance of machine learning algorithms. The area of meta-learning mimics this strategy by reasoning about the performance of learning algorithms across datasets. we apply meta-learning to select instantiations of our given machine learning framework that are likely to perform well on a new dataset. instantiations of our given machine learning framework that are likely to perform well on a new dataset. More specifically, for a large number of datasets, we collect both performance data and a set of meta-features, i.e., characteristics of the dataset that can be computed efficiently and that help to determine which algorithm to use on a new dataset. Meta Feature (数据集的一些外在feature)，计算不同数据集之间的相似度，相似的数据可以采取类似的hyper parameter. Model Ensemble