Standard Scaler Train-Platform For AI(PAI)-阿里云帮助中心

When features in your training data have different scales and units, algorithms sensitive to feature magnitude — such as linear models, support vector machines (SVMs), and neural networks — can produce biased results. The Standard Scaler Train component solves this by applying standardization to selected numeric columns, centering them around zero and scaling to unit variance. The output is a scaler model that the Standard Scaler Batch Predict component uses at inference time.

StandardScaler assumes the input data follows a normal distribution.

Standardization is most effective for algorithms sensitive to feature scale, such as linear models, SVMs, and neural networks. Tree-based models (such as gradient boosting and random forests) generally do not require standardization.

Limitations

Supported compute engines: MaxCompute and Realtime Compute for Apache Flink.

Configure the component in Machine Learning Designer

Input ports

Input port (left to right)	Data type	Recommended upstream component	Required
data	Integer	Read Table, Read CSV File	Yes

Component parameters

Tab	Parameter	Description
Field Setting	selectedCols	Columns to standardize. Select one or more columns. All selected columns must be numeric.
Parameter Setting	withMean	Subtract the column mean before scaling. Enabled by default (`true`).
Parameter Setting	withStd	Divide by the column standard deviation. Enabled by default (`true`).
Execution Tuning	Number of Workers	Number of parallel workers. Must be a positive integer. Valid values: 1–9,999. Configure together with Memory per worker, unit MB.
Execution Tuning	Memory per worker, unit MB	Memory allocated to each worker. Valid values: 1,024–65,536. Unit: MB.

Output ports

Output port (left to right)	Storage location	Recommended downstream component	Model type
model	N/A	Standard Scaler Batch Predict	None

Example

The following PyAlink code replicates the Standard Scaler Train component. Copy it into the code editor of the PyAlink Script component.

from pyalink.alink import *

def main(sources, sinks, parameter):
    data = sources[0]
    selectedColNames = ["col2", "col3"]
    # Fit the scaler on training data and output a scaler model
    modelop = StandardScalerTrainBatchOp()\
        .setSelectedCols(selectedColNames)
    result = modelop.linkFrom(data)
    # The output model is passed to Standard Scaler Batch Predict for inference
    result.link(sinks[0])
    BatchOperator.execute()

What's next

Standard Scaler Batch Predict — apply the trained scaler model to new data
PyAlink Script — run custom Alink code in Machine Learning Designer