Platform For AI:PS-SMART Multiclass Classification

Limits

Input data for the PS-SMART multiclass classification must adhere to the following criteria:

• The target column in the PS-SMART multiclass classification only accepts numeric types. If the MaxCompute table data is a STRING type, it requires conversion. For instance, a classification target labeled as Good/Medium/Bad string should be transformed to 0/1/2.

• For KV format data, the feature ID must be a positive integer, and the feature value must be a real number. Serialization is necessary if the feature ID is a STRING type, using the serialization component. Categorical strings as feature values require feature discretization and other feature engineering processes.

• While the PS-SMART multiclass classification supports tasks with hundreds of thousands of features, it is resource-intensive and slow. GBDT-like algorithms, suitable for training with continuous features, can be used instead. It is advisable to perform One-Hot encoding on categorical features, excluding low-frequency ones, and to avoid discretizing other continuous numerical features.

• The PS-SMART algorithm incorporates elements of randomness, such as data_sample_ratio and fea_sample_ratio, which are responsible for data and feature sampling respectively. It also utilizes histogram approximation optimization and introduces randomness in the sequence of merging local Sketches into the global Sketch. Despite the potential for varying tree structures when executed by multiple workers in a distributed environment, the algorithm can theoretically ensure comparable model outcomes. Experiencing inconsistent results during training with identical data and parameters is expected and should not be cause for concern.

• To expedite training, consider increasing the Number Of Computing Cores. However, since PS-SMART requires all servers to secure resources before commencing training, requesting additional resources during peak cluster times may lead to longer wait times.

Precautions

Keep the following in mind when utilizing the PS-SMART multiclass classification component:

• Although the PS-SMART multiclass classification component can manage tasks with numerous features, it is resource-intensive and operates slowly. GBDT-like algorithms, which are well-suited for direct training with continuous features, are recommended. One-Hot encoding should be applied to categorical features, except for those that are low-frequency, and discretization of other continuous numerical features is generally discouraged.

• The PS-SMART algorithm incorporates elements of randomness, such as data_sample_ratio and fea_sample_ratio, which are indicative of data and feature sampling. It also utilizes histogram approximation optimization and introduces variability in the sequence of merging local Sketches into the global Sketch. Despite the potential for varying tree structures when executed by multiple workers in a distributed environment, the algorithm can theoretically maintain consistent model performance. Experiencing different outcomes from repeated training sessions with identical data and parameters is expected behavior.

• To speed up training, you can increase the Number Of Computing Cores. Note that PS-SMART requires all servers to acquire resources before training starts, so requesting more resources during busy times may result in increased waiting periods.

Configure the component

Method one: Visualization

Add the PS-SMART Multiclass Classification component to the Designer workflow page and set the parameters on the right side of the interface:

Method two: PAI command

Configure the PS-SMART Multiclass Classification component parameters using PAI commands. Invoke PAI commands using the SQL script component. For more information, see Scenario 4: Execute PAI commands in the SQL script component.

--Training
PAI -name ps_smart
    -project algo_public
    -DinputTableName="smart_multiclass_input"
    -DmodelName="xlab_m_pai_ps_smart_bi_545859_v0"
    -DoutputTableName="pai_temp_24515_545859_2"
    -DoutputImportanceTableName="pai_temp_24515_545859_3"
    -DlabelColName="label"
    -DfeatureColNames="features"
    -DenableSparse="true"
    -Dobjective="multi:softprob"
    -Dmetric="mlogloss"
    -DfeatureImportanceType="gain"
    -DtreeCount="5";
    -DmaxDepth="5"
    -Dshrinkage="0.3"
    -Dl2="1.0"
    -Dl1="0"
    -Dlifecycle="3"
    -DsketchEps="0.03"
    -DsampleRatio="1.0"
    -DfeatureRatio="1.0"
    -DbaseScore="0.5"
    -DminSplitLoss="0"
--Prediction
PAI -name prediction
    -project algo_public
    -DinputTableName="smart_multiclass_input";
    -DmodelName="xlab_m_pai_ps_smart_bi_545859_v0"
    -DoutputTableName="pai_temp_24515_545860_1"
    -DfeatureColNames="features"
    -DappendColNames="label,features"
    -DenableSparse="true"
    -DkvDelimiter=":"
    -Dlifecycle="28"

Usage example

1. Generate input data using the following SQL statement (example given for KV format data).

   drop table if exists smart_multiclass_input;
   create table smart_multiclass_input lifecycle 3 as
   select
   *
   from
   (
   select 2 as label, '1:0.55 2:-0.15 3:0.82 4:-0.99 5:0.17' as features from dual
       union all
   select 1 as label, '1:-1.26 2:1.36 3:-0.13 4:-2.82 5:-0.41' as features from dual
       union all
   select 1 as label, '1:-0.77 2:0.91 3:-0.23 4:-4.46 5:0.91' as features from dual
       union all
   select 2 as label, '1:0.86 2:-0.22 3:-0.46 4:0.08 5:-0.60' as features from dual
       union all
   select 1 as label, '1:-0.76 2:0.89 3:1.02 4:-0.78 5:-0.86' as features from dual
       union all
   select 1 as label, '1:2.22 2:-0.46 3:0.49 4:0.31 5:-1.84' as features from dual
       union all
   select 0 as label, '1:-1.21 2:0.09 3:0.23 4:2.04 5:0.30' as features from dual
       union all
   select 1 as label, '1:2.17 2:-0.45 3:-1.22 4:-0.48 5:-1.41' as features from dual
       union all
   select 0 as label, '1:-0.40 2:0.63 3:0.56 4:0.74 5:-1.44' as features from dual
       union all
   select 1 as label, '1:0.17 2:0.49 3:-1.50 4:-2.20 5:-0.35' as features from dual
   ) tmp;

   The resulting data appears as follows.

2. Construct the experiment. For details, see Algorithm modeling.

3. Set the parameters for the PS-SMART multiclass classification component (refer to the table below for specific parameters, using default values for others).

   Parameter type	Parameter	Description
   Field settings	Feature column	Select the features column.
   	Label column	Select the label column.
   	Is sparse format	Select the Is Sparse Format check box.
   Parameter settings	Number of categories	Enter 3.
   	Evaluation metric type	Select multiclass negative log likelihood.
   	Number of trees	Enter 5.

4. Adjust the parameters for the unified prediction component (refer to the table below for specific parameters, using default values for others).

   Parameter type	Parameter	Description
   Field settings	Feature column	Default select all. Extra columns do not affect prediction results.
   	Original output column	Select the label column.
   	Sparse matrix	Select the Sparse Matrix check box.
   	Key and value separator	Enter a colon (:).
   	Separator between kv pairs	Enter \u0020.

5. Configure the prediction component parameters (refer to the table below for specific parameters, using default values for others).

   Parameter type	Parameter	Description
   Field settings	Feature column	Default select all. Extra columns do not affect prediction results.
   	Original output column	Select the label column.
   	Sparse matrix	Select the Sparse Matrix check box.
   	Key and value separator	Enter a colon (:).
   	Separator between kv pairs	Enter \u0020.

6. Execute the experiment and examine the unified prediction component's prediction results. Notably:

   • The numbers 0, 1, and 2 in the prediction_detail column correspond to the multiclass classification categories.

   • The predict_result column indicates the predicted category.

   • The predict_score column shows the probability of the predicted category.

7. Review the prediction results from the PS-Smart prediction component. Specifically:

   • The score_class_k column displays the probability of predicting the k-th category.

   • The leaf_index column denotes the predicted leaf node number. If the number of trees is N and the number of categories is M, each sample's leaf_index value is N×M numbers. For instance, in this example, the leaf_index value is 5×3=15, with each tree corresponding to a number indicating the leaf node where the sample falls.

8. Right-click the PS-SMART Multiclass Classification component and select View Data > View Output Pile 3 from the shortcut menu to observe feature importance.

   Specifically:

   • The id column lists the feature sequence number provided. In this example, since the input data is in KV format, the id column represents the key in the KV pair.

   • The value column indicates the feature importance type, with gain being the default, representing the sum of the information gain contributed by the feature to the model.

PS-SMART model deployment

If you want to deploy the model generated by the PS-SMART Binary Classification Training component to EAS as an online service, you must add the Model export component as a downstream node for the PS-SMART Binary Classification Training component and configure the Model export component. For more information, see Model export.

After the Model export component is successfully run, you can deploy the generated model to EAS as an online service on the EAS-Online Model Services page. For more information, see Deploy a model service in the PAI console.