Logistic regression for multiclass classification

Method 1: Configure in Machine Learning Designer

Open the Logistic Regression for Multiclass Classification component in Machine Learning Designer (formerly Machine Learning Studio) and set the following parameters.

Fields Setting tab

Parameters Setting tab

Method 2: Use PAI commands

Pass parameters directly to the logisticregression_multi algorithm using PAI commands. Run PAI commands through the SQL Script component. For more information, see SQL Script.

The following example shows the command syntax:

PAI -name logisticregression_multi
    -project algo_public
    -DmodelName="xlab_m_logistic_regression_6096"
    -DregularizedLevel="1"
    -DmaxIter="100"
    -DregularizedType="l1"
    -Depsilon="0.000001"
    -DlabelColName="y"
    -DfeatureColNames="pdays,emp_var_rate"
    -DgoodValue="1"
    -DinputTableName="bank_data"

Parameters

Step 1: Create training data

Run the following SQL statements to create the multi_lr_test_input table:

drop table if exists multi_lr_test_input;
create table multi_lr_test_input
as
select
    *
from
(
    select
        cast(1 as double) as f0,
        cast(0 as double) as f1,
        cast(0 as double) as f2,
        cast(0 as double) as f3,
        cast(0 as bigint) as label
    union all
        select
            cast(0 as double) as f0,
            cast(1 as double) as f1,
            cast(0 as double) as f2,
            cast(0 as double) as f3,
            cast(0 as bigint) as label
    union all
        select
            cast(0 as double) as f0,
            cast(0 as double) as f1,
            cast(1 as double) as f2,
            cast(0 as double) as f3,
            cast(2 as bigint) as label
    union all
        select
            cast(0 as double) as f0,
            cast(0 as double) as f1,
            cast(0 as double) as f2,
            cast(1 as double) as f3,
            cast(1 as bigint) as label
) a;

The table contains four DOUBLE feature columns (f0–f3) and one BIGINT label column:

Step 2: Train the model

Run the following PAI command to train the model and save it as multi_lr_test_model:

drop offlinemodel if exists multi_lr_test_model;
PAI -name logisticregression_multi
    -project algo_public
    -DmodelName="multi_lr_test_model"
    -DitemDelimiter=","
    -DregularizedLevel="1"
    -DmaxIter="100"
    -DregularizedType="None"
    -Depsilon="0.000001"
    -DkvDelimiter=":"
    -DlabelColName="label"
    -DfeatureColNames="f0,f1,f2,f3"
    -DenableSparse="false"
    -DinputTableName="multi_lr_test_input";

Step 3: Run predictions

Run the following PAI command to generate predictions and write results to multi_lr_test_prediction_result:

drop table if exists multi_lr_test_prediction_result;
PAI -name prediction
    -project algo_public
    -DdetailColName="prediction_detail"
    -DmodelName="multi_lr_test_model"
    -DitemDelimiter=","
    -DresultColName="prediction_result"
    -Dlifecycle="28"
    -DoutputTableName="multi_lr_test_prediction_result"
    -DscoreColName="prediction_score"
    -DkvDelimiter=":"
    -DinputTableName="multi_lr_test_input"
    -DenableSparse="false"
    -DappendColNames="label";

Step 4: View results

Query the multi_lr_test_prediction_result table to review the prediction output:

The output columns contain the following information:

• prediction_result: The predicted class label.

• prediction_score: The probability assigned to the predicted class.

• prediction_detail: A JSON object mapping each class label to its predicted probability. Each key is a class label and each value is the model's confidence for that class. For example, {"0": 0.999..., "1": 2.32e-07, "2": 2.32e-07} indicates that the model assigns near-certainty to class 0 and near-zero probability to classes 1 and 2.