K-nearest neighbors-Platform For AI(PAI)-阿里云帮助中心

The k-nearest neighbors algorithm classifies data by finding the K nearest records in the training table for each row in the prediction table. The class that appears most frequently among these K records is assigned as the predicted class for that row.

Component configuration

You can configure the k-nearest neighbors component in one of the following ways.

Method 1: Visual configuration

You can configure the component parameters on the workflow page in Designer.

Tab	Parameter	Description
Fields setting	Training table feature columns	The feature columns to use for training.
	Training table label column	The target column for training.
	Prediction table feature columns	If you do not configure this parameter, the feature columns of the prediction table are the same as those of the training table.
	Append ID columns to output table	Specifies the ID columns to obtain the predicted value for a specific column. By default, the feature columns of the prediction table are used as the ID columns.
	Input data is in sparse format	Use the key-value pair format for sparse data.
	Separator for key-value pairs	The default separator is a comma (,).
	Separator for key and value	The default separator is a colon (:).
Parameters setting	Number of neighbors	The default value is 100.
Execution tuning	Number of cores	By default, the system assigns it automatically.
Execution tuning	Memory size	The system assigns this by default.

Method 2: PAI command

You can configure the component parameters by running a PAI command in the SQL script component. For more information, see SQL script.

PAI -name knn
    -DtrainTableName=pai_knn_test_input
    -DtrainFeatureColNames=f0,f1
    -DtrainLabelColName=class
    -DpredictTableName=pai_knn_test_input
    -DpredictFeatureColNames=f0,f1
    -DoutputTableName=pai_knn_test_output
    -Dk=2;

Parameter	Required	Description	Default value
trainTableName	Yes	The name of the training table.	None
trainFeatureColNames	Yes	The names of the feature columns in the training table.	None
trainLabelColName	Yes	The name of the label column in the training table.	None
trainTablePartitions	No	The partitions in the training table to use for training.	All partitions
predictTableName	Yes	The name of the prediction table.	None
outputTableName	Yes	The name of the output table.	None
predictFeatureColNames	No	The names of the feature columns in the prediction table.	Same as trainFeatureColNames
predictTablePartitions	No	The partitions in the prediction table to use for prediction.	All partitions
appendColNames	No	The names of the columns from the prediction table to append to the output table.	Same as predictFeatureColNames
outputTablePartition	No	The partition of the output table.	The entire table
k	No	The number of nearest neighbors. Valid values: 1 to 1000.	100
enableSparse	No	Specifies whether the input table data is in sparse format. Valid values: {true,false}.	false
itemDelimiter	No	If the input data is in sparse format, this is the separator between key-value pairs.	Comma (,)
kvDelimiter	No	If the input data is in sparse format, this is the separator between a key and its value.	Colon (:)
coreNum	No	The number of nodes. Use this parameter with memSizePerCore. Valid values: 1 to 20,000.	Calculated by the system
memSizePerCore	No	The memory of a single node. Valid values: 1024 MB to 64 × 1024 MB.	Calculated by the system
lifecycle	No	The lifecycle of the output table.	None

Example

Generate training data.

create table pai_knn_test_input as
select * from
(
  select 1 as f0,2 as f1, 'good' as class
  union all
  select 1 as f0,3 as f1, 'good' as class
  union all
  select 1 as f0,4 as f1, 'bad' as class
  union all
  select 0 as f0,3 as f1, 'good' as class
  union all
  select 0 as f0,4 as f1, 'bad' as class
)tmp;

Use a PAI command to submit the parameters for the k-nearest neighbors component.

pai -name knn
    -DtrainTableName=pai_knn_test_input
    -DtrainFeatureColNames=f0,f1
    -DtrainLabelColName=class
    -DpredictTableName=pai_knn_test_input
    -DpredictFeatureColNames=f0,f1
    -DoutputTableName=pai_knn_test_output
    -Dk=2;

View the training results.The results include the following information:
- f0 and f1 are the appended columns.
- prediction_result is the classification result.
- prediction_score is the probability of the classification result.
- prediction_detail shows the K nearest classes and their probabilities.