Flink SVM算法参数

向右看齐 2021-11-05 01:20 221阅读 0赞

在机器学习中，支持向量机（SVM）是在分类与回归分析中分析数据的监督式学习模型。SVM模型是将实例表示为空间中的点，通过映射使得单独类别的实例被尽可能宽的明显间隔分开，然后将新的实例映射到同一空间中，并基于它们落在间隔的哪一侧来预测所属类别。除了进行线性分类外，SVM还可以使用所谓的核技巧有效地进行非线性分类，并将其输入隐式映射到高维特征空间中。

在 Flink中SVM是一个预测函数，含有ft和 predict两个方法，其中伍fit方法是基于训练数据集进行转化和训练，生成SVM模型；而 Predict方法使用已经训练好的模型进行预测，并生成 Labelvector类型预测结果。

针对SVM模型中参数配置如表所示，用户可以根据情况自行调整和优化。

**SVM参数汇总**

<table style="width:654px;"> 
 <tbody> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-color:rgb(0,0,0);">参数名称</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-top-color:rgb(0,0,0);">参数说明</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-top-color:rgb(0,0,0);">默认值</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">Blocks</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">设定介入的数据集将要被切分的Block数量，每块数据集上本地执行随机双坐标上升方法，且Blocks数量至少要和并行度一样</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">None</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">Iterations</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">最大迭代数量， 超过该值则终止迭代</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">10</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">LocalIterations</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">定义随机双坐标上升（SDCA）最大的迭代次数</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">10</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">Regurlarization</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">定义SVM算法正则化参数</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">1.0</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">Stepsize</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">定义更新权重向量的初始化步长，步长越长，则本次权重向量每次更新到下一个权重向量的贡献度将更多</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">1.0</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">ThresholdValue</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">定义正负标签确定函数的阈值，大于该值则认为是正标签，小于该值则认为是负标签</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">0.0</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">OutputDecisionFunction</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">确定预测和评估函数中应该返回的实例类型，设定为True返回超平面的原生实例类型，设定为False则返回Lable标签信息</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">False</td> 
 </tr> 
 <tr> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);border-left-color:rgb(0,0,0);">Seed</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">初始化随机数的种子</td> 
 <td style="font-family:'SF Pro', Roboto, 'Droid Sans', Noto, Helvetica, Tahoma, Arial, 'Microsoft YaHei', '微软雅黑', 'PingFang SC', 'Hiragino Sans GB', 'Heiti SC', sans-serif;border-bottom-color:rgb(0,0,0);border-right-color:rgb(0,0,0);">随机值</td> 
 </tr> 
 </tbody> 
</table>

![640?wx\_fmt=png][640_wx_fmt_png]

[640_wx_fmt_png]: /images/20211104/f6f21cd6d17742a9af7a214be2687dee.png