Scoring type

Read(289) Label: batch scoring, scoring,

After importing to-be-predicted data, you get the following window where there are two scoring types – “Batch scoring” and “Scoring”.

 

Batch scoring

You can score all imported data with “Batch scoring”.

 

Execute scoring

Click scoring button  on the toolbar from the “Batch scoring” tab to score all valid data. The scoring is completed when “Prediction finished” message appears:

Click “Finish” to view the scoring result:

Scoring results of different types of target variables have their own formats. A binary target variable will have scoring results of percentages. A numerical target variable will get scoring results of numerical data.

 

Check model performance by scoring result

If to-be-scored data contains the target variable, you can check model performance according to the scoring result via “Model performance of scoring results” option on the lower part of the main screen. This lets you test model performance and evaluate model quality by comparing the actual performance with the model parameters configured in the modeling file.

Scoring options

Click “Scoring options” in “Batch scoring” page to configure the current scoring operation. Different types of target variables have different options.

One target variable

Below is the dialog when the targe variable is of numerical type:

Below is the dialog when the targe variable is of binary type:

“Import test data in batches”: Check the option to set the number of test data rows imported at each batch.

“Output scoring label”: Check the option to set positive rate threshold. Click “Use suggested threshold” to use the displayed suggested positive rate threshold. With this option checked, scoring label will be displayed on the “Batch scoring” page after data scoring is executed. The corresponding field name is usually “TargetVariable_label”.

For example, if the target variable under scoring is Survived and we set the positive rate threshold as 0.5, the scoring label displayed after scoring is finished is Survived_label. Values of “Survived_1_percentage” field are percentages. If the specified percentage is <50%, the label value is 0, which means dead; when the percentage is ≥50%, the label value is 1, which means survived.

Multiple target variables

“Import test data in batches”: Check the option to set the number of test data rows imported at each batch.

“Number of aggregate groups”: This option sets the number of groups into which the to-be-scored data is divided and on which an aggregate operation is performed when multiple target variables are scored.

 

Export

Click “Export” button to export to-be-scored data and scoring result to a TXT file, a CSV file or an Excel file, or you can copy it to the clipboard. By default an exported file or the copied data contains headers.

“Copy to system clipboard”: Copy all data on the current page.

“Copy selected data to system clipboard”: Copy selected data only; use Ctrl to select multiple records.

 

Scoring

This lets you score selected data or perform scoring according to a certain condition.

Types

Type 1:

Select a line in “Batch scoring” tab and click “Score the first selected line”:

The system will jump to “Scoring” tab and show the result on the top, as shown below:

 

Type 2:

Under the “Edit” in the “Scoring” tab, you can shift the pointer to reset variable values on which the prediction is made. The scoring result will change accordingly.

For example, if values of the other variables remain unchanged and you shift Age value from 21 to 31, the scoring result will become 9.561%, lower than the previous 25.707% when the variable value is 31:

Save scoring result

You can save the current scoring result by clicking “Save scoring result”. And you can export the scoring result, shift it up, shift it down or delete it via “Export”, “Shift up”, “Shift down” and “Delete” buttons.