Feature Selection (Filter Method) - Platform For AI - Alibaba Cloud Documentation Center

The Feature Selection (Filter Method) component selects the top N features from all feature data in sparse or dense formats by using a filter based on the feature selection method that you specify. This component saves the selected features in a feature importance table. This reduces the difficulty of model training and improves the accuracy of the trained model. This topic describes how to configure parameters for the Feature Selection (Filter Method) component provided by Machine Learning Designer (formerly known as Machine Learning Studio). This topic also provides an example on how to use the Feature Selection (Filter Method) component.

Limits

The Feature Selection (Filter Method) component does not support filtering of data in the LIBSVM or key-value pair format.

Configure the component

You can use one of the following methods to configure the Feature Selection (Filter Method) component.

Method 1: Configure the component on the pipeline page

You can configure the parameters of the Feature Selection (Filter Method) component on the pipeline page of Machine Learning Designer of Machine Learning Platform for AI (PAI). Machine Learning Designer is formerly known as Machine Learning Studio. The following table describes the parameters.


Tab	Parameter	Description
Fields Setting	Feature Columns	The names of the feature columns that are selected from the input table for training.
	Target Column	The name of the label column that is selected from the input table.
	Enumeration Features	The columns of features to be processed as enumeration features. Only columns of the INT and DOUBLE data types are supported.
	Sparse Features (K:V,K:V)	Specifies whether the features are sparse features in the key-value pair format.
Parameters Setting	Feature Selection Method	The method that is used to select features. Valid values: IV: measures the prediction capability of features. Gini Gain: evaluates the importance of a single feature. Information Gain: measures the differences among multiple features. Lasso: filters ultra-large-scale features by using dimensionality reduction.
	Top N Features	The top N features to be selected. If the specified number is greater than the number of input features, all features are selected.
	Continuous Feature Partitioning Mode	The partitioning mode of continuous features. Valid values: Automated Partitioning Equal Width Partitioning
	Continuous Feature Discretization Intervals	You need to set this parameter only if you set the Continuous Feature Partitioning Mode parameter to Equal Width Partitioning.

Method 2: Use PAI commands

Configure the component parameters by using PAI commands. You can use the SQL Script component to call PAI commands. For more information, see SQL Script.

PAI -name fe_select_runner -project algo_public 
     -DfeatImportanceTable=pai_temp_2260_22603_2 
     -DselectMethod=iv 
     -DselectedCols=pdays,previous,emp_var_rate,cons_price_idx,cons_conf_idx,euribor3m,nr_employed,age,campaign 
     -DtopN=5 
     -DlabelCol=y 
     -DmaxBins=100 
     -DinputTable=pai_dense_10_9 
     -DoutputTable=pai_temp_2260_22603_1;


Parameter	Required	Description	Default value
inputTable	Yes	The name of the input table.	N/A
inputTablePartitions	No	The partitions that are selected from the input table for training. The following formats are supported: Partition_name=value name1=value1/name2=value2: multi-level partitions Note If you specify multiple partitions, separate them with commas (,).	All partitions
outputTable	Yes	The feature result table that is generated after filtering.	N/A
featImportanceTable	Yes	The table that stores the importance weight values of all input features.	N/A
selectedCols	Yes	The feature columns that are used for training.	N/A
labelCol	Yes	The label column that is selected from the input table.	N/A
categoryCols	No	The columns of enumeration features. Only columns of the INT or DOUBLE data types are supported.	N/A
maxBins	No	The maximum number of intervals for continuous feature partitioning.	100
selectMethod	No	The method that is used to select features. Valid values: iv, GiniGain, InfoGain, and Lasso.	iv
topN	No	The top N features to be selected. If the specified number is greater than the number of input features, all features are selected.	10
isSparse	No	Specifies whether the features are sparse features in the key-value pair format. A value of false indicates dense features.	false

Example

Input data

Execute the following SQL statement to generate test data:

create table if not exists pai_dense_10_9 as 
select  
    age,campaign,pdays,previous,emp_var_rate,cons_price_idx,cons_conf_idx,euribor3m,nr_employed,y
from bank_data limit 10;

Parameter settings
The input table is pai_dense_10_9. Select the y column for Target Column and other columns for Feature Columns. The following figures show the detailed parameter settings.
Figure 1. Fields Setting tab
Figure 2. Parameters Setting tab

Output

The left output is the filtered data that is stored in the following table.


pdays	nr_employed	emp_var_rate	cons_conf_idx	cons_price_idx	y
999.0	5228.1	1.4	-36.1	93.444	0.0
999.0	5195.8	-0.1	-42.0	93.2	0.0
6.0	4991.6	-1.7	-39.8	94.055	1.0
999.0	5099.1	-1.8	-47.1	93.075	0.0
3.0	5076.2	-2.9	-31.4	92.201	1.0
999.0	5228.1	1.4	-42.7	93.918	0.0
999.0	5099.1	-1.8	-46.2	92.893	0.0
999.0	5099.1	-1.8	-46.2	92.893	0.0
3.0	5076.2	-2.9	-40.8	92.963	1.0
999.0	5099.1	-1.8	-47.1	93.075	0.0

The right output is the feature importance table shown below. The featname column stores the feature names. The weight column stores the weight values that are calculated by the feature selection method.


featname	weight
pdays	30.675544191232486
nr_employed	29.08332850085075
emp_var_rate	29.08332850085075
cons_conf_idx	28.02710269740324
cons_price_idx	28.02710269740324
euribor3m	27.829058450563718
age	27.829058450563714
previous	14.319325030742775
campaign	10.658129656314467