One-Class SVM Outlier - Platform For AI - Alibaba Cloud Documentation Center

One-Class Support Vector Machine (SVM) is an unsupervised machine learning algorithm that is different from traditional SVM algorithms. You can use the One-Class SVM Outlier component to detect outliers by learning a decision boundary. This topic describes how to configure the One-Class SVM Outlier component in Platform for AI (PAI).

Limits

You can use the One-Class SVM Outlier component based only on the computing resources of MaxCompute.

Configure the component

You can use one of the following methods to configure the parameters of the One-Class SVM Outlier component:

Method 1: Configure the component in the PAI console

Configure the component on the pipeline page of Machine Learning Designer. The following table describes the parameters.

Tab	Parameter	Description
Field Setting	featureCols	An array of the names of feature columns.
	groupCols	An array of the names of group columns.
	tensorCol	The tensor column.
	vectorCol	The name of the vector column.
Parameter Setting	Prediction Result Column	The name of the prediction result column.
	coef0	The coef0 parameter of the kernel function. Default value: 0.0. Note This parameter takes effect only if the type of the kernel function is polynomial or sigmoid.
	degree	The degree of the polynomial.
	epsilon	The value that you want to obtain from the training results before the iteration stops. Default value: 1.0E-6.
	gamma	The gamma parameter of the kernel function. Default value: -1.0. Note This parameter takes effect only if the type of the kernel function is RBF, polynomial, or sigmoid. If you do not configure this parameter, the default value 1/data dimension is used.
	kernelType	The type of the kernel function. Valid values: RBF POLY SIGMOID LINEAR
	maxOutlierNumPerGroup	The maximum number of outliers per group.
	maxOutlierRatio	The maximum ratio of outliers that are detected by the algorithm.
	maxSampleNumPerGroup	The maximum number of samples per group.
	nu	The nu parameter of the kernel function. This parameter is positively correlated with the number of support vectors. Valid values: (0,1). Default value: 0.01.
	outlierThreshold	If the score exceeds the specified threshold, the data point is considered an anomalous point.
	Column name of detail prediction information	The name of the prediction details column.
	numThreads	The number of threads of the component.
Execute Tuning	Number of Workers	The number of worker nodes. The value must be a positive integer. This parameter must be used with the Memory per worker parameter. Valid values: 1 to 9999.
Execute Tuning	Memory per worker	The memory size of each worker node. Unit: MB. The value must be a positive integer. You must specify a value from 1024 to 65536.

Method 2: Configure the component by using Python code

You can configure the One-Class SVM Outlier component by using the PyAlink Script component to call Python code. For more information, see PyAlink script.

Parameter	Required	Description	Default value
predictionCol	Yes	The name of the prediction results column.	N/A
degree	No	The degree of the polynomial.	2
epsilon	No	The value that you want to obtain from the training results before the iteration stops.	1.0E-6
featureCols	No	An array of the names of feature columns.	Select All
groupCols	No	The array of the names of group columns.	N/A
maxOutlierNumPerGroup	No	The maximum number of outliers per group.	N/A
maxOutlierRatio	No	The maximum ratio of outliers that are detected by the algorithm.	N/A
maxSampleNumPerGroup	No	The maximum number of samples per group.	N/A
outlierThreshold	No	If the score exceeds the specified threshold, the data point is considered an anomalous point.	N/A
predictionDetailCol	No	The name of the prediction details column.	N/A
tensorCol	No	The name of the tensor column.	N/A
vectorCol	No	The name of the vector column.	N/A
kernelType	No	The type of the kernel function. Valid values: RBF POLY SIGMOID LINEAR	RBF
coef0	No	The coef0 parameter of the kernel function. Note This parameter takes effect only if the type of the kernel function is polynomial or sigmoid.	0.0
gamma	No	The gamma parameter of the kernel function. Note This parameter takes effect only if the type of the kernel function is RBF, polynomial, or sigmoid. If you do not configure this parameter, the default value 1/data dimension is used.	-1.0
nu	No	The nu parameter of the kernel function. This parameter is positively correlated with the number of support vectors. Valid values: (0,1).	0.01
numThreads	No	The number of threads of the component.	1

Sample Python code:

df = pd.DataFrame([
[0.730967787376657,0.24053641567148587,0.6374174253501083,0.5504370051176339],
[0.7308781907032909,0.41008081149220166,0.20771484130971707,0.3327170559595112],
[0.7311469360199058,0.9014476240300544,0.49682259343089075,0.9858769332362016],
[0.731057369148862,0.07099203475193139,0.06712000939049956,0.768156984078079],
[0.7306094602878371,0.9187140138555101,0.9186071189908658,0.6795571637816596],
[0.730519863614471,0.08825840967622589,0.4889045498516358,0.461837214623537],
[0.7307886238322471,0.5796252073129174,0.7780122870716483,0.11499709190022733],
[0.7306990420600421,0.7491696031336331,0.34830970303125697,0.8972771427421047]])

# load data
data = BatchOperator.fromDataframe(df, schemaStr="x1 double, x2 double, x3 double, x4 double")

OcsvmOutlierBatchOp() \
            .setFeatureCols(["x1", "x2", "x3", "x4"]) \
            .setGamma(0.5) \
            .setNu(0.1) \
            .setKernelType("RBF") \
            .setPredictionCol("pred").linkFrom(data).print()