bugfix: correctly handle regression models with nominal inputs (#103)

* bugfix: correctly identify classification vs regression models

* code qual improvements

Author: jlwalke2

CHANGELOG.md

@@ -1,6 +1,8 @@
 Unreleased
 ----------
- -
+**Bugfixes**
+ - Fixed an issue with `register_model()` where random forest, gradient boosting, and SVM regression models with 
+ nominal inputs where incorrectly treated as classification models. 
 
 v1.6.1 (2021-09-01)
 -------------------

src/sasctl/utils/astore.py

@@ -350,12 +350,19 @@ def _get_model_properties(result):
     else:
         algorithm = None
 
+    def is_classification(r):
+        """Determine if the ASTORE model describes a classification model."""
+        return classification_target(r) is not None
+
     def classification_target(r):
-        target = r.OutputVariables.Name.str.startswith('I_')
-        target = r.OutputVariables.Name[target].iloc[0]
-        return target.replace('I_', '', 1)
+        """Get the name of the classification target variable."""
+        target = r.OutputVariables.Name[r.OutputVariables.Name.str.startswith('I_')]
+        if target.shape[0] > 0:
+            return target.iloc[0].replace('I_', '', 1)
+        return None
 
     def regression_target(r):
+        """Get the name of the regression target variable."""
         target = r.OutputVariables.Name.str.startswith('P_')
         target = r.OutputVariables.Name[target].iloc[0]
         return target.replace('P_', '', 1)
@@ -375,7 +382,7 @@ def regression_target(r):
     elif algorithm == 'forest':
         properties['algorithm'] = 'Random forest'
 
-        if 'Classification' in result.InputVariables.Type.values:
+        if is_classification(result):
             properties['function'] = 'classification'
             properties['targetVariable'] = classification_target(result)
         else:
@@ -385,7 +392,7 @@ def regression_target(r):
     elif algorithm == 'gradboost':
         properties['algorithm'] = 'Gradient boosting'
 
-        if 'Classification' in result.InputVariables.Type.values:
+        if is_classification(result):
             properties['function'] = 'classification'
             properties['targetVariable'] = classification_target(result)
 
@@ -398,7 +405,7 @@ def regression_target(r):
     elif algorithm == 'svmachine':
         properties['algorithm'] = 'Support vector machine'
 
-        if 'Classification' in result.InputVariables.Type.values:
+        if is_classification(result):
             properties['function'] = 'classification'
             properties['targetVariable'] = classification_target(result)
             properties['targetLevel'] = 'binary'

tests/cassettes/tests.integration.test_astore_models.test_forest_regression_with_nominals_swat.json

@@ -215,6 +215,37 @@ def test_forest_regression(cas_session, boston_dataset):
     check_input_variables(files, BOSTON_INPUT_VARS)
 
 
+def test_forest_regression_with_nominals(cas_session, boston_dataset):
+    target = {
+        'tool': 'SAS Visual Data Mining and Machine Learning',
+        'targetVariable': 'Price',
+        'scoreCodeType': 'ds2MultiType',
+        'function': 'prediction',
+        'algorithm': 'Random forest',
+    }
+
+    cas_session.loadactionset('decisiontree')
+    cas_session.loadactionset('astore')
+
+    tbl = cas_session.upload(boston_dataset).casTable
+
+    tbl.decisiontree.foresttrain(
+        target='Price',
+        inputs=list(boston_dataset.columns[:-1]),
+        nominals=['chas'],
+        saveState='astore',
+    )
+
+    desc = cas_session.astore.describe(rstore='astore', epcode=True)
+    props = _get_model_properties(desc)
+
+    for k, v in target.items():
+        assert props[k] == v
+
+    files = create_files_from_astore(cas_session.CASTable('astore'))
+    check_input_variables(files, BOSTON_INPUT_VARS)
+
+
 def test_gradboost_binary_classification(cas_session, cancer_dataset):
     target = {
         'tool': 'SAS Visual Data Mining and Machine Learning',
@@ -300,6 +331,37 @@ def test_gradboost_regression(cas_session, boston_dataset):
     check_input_variables(files, BOSTON_INPUT_VARS)
 
 
+def test_gradboost_regression_with_nominals(cas_session, boston_dataset):
+    target = {
+        'tool': 'SAS Visual Data Mining and Machine Learning',
+        'targetVariable': 'Price',
+        'scoreCodeType': 'ds2MultiType',
+        'function': 'prediction',
+        'algorithm': 'Gradient boosting',
+    }
+
+    cas_session.loadactionset('decisiontree')
+    cas_session.loadactionset('astore')
+
+    tbl = cas_session.upload(boston_dataset).casTable
+
+    tbl.decisiontree.gbtreetrain(
+        target='Price',
+        inputs=list(boston_dataset.columns[:-1]),
+        nominals=['chas'],
+        savestate='astore',
+    )
+
+    desc = cas_session.astore.describe(rstore='astore', epcode=True)
+    props = _get_model_properties(desc)
+
+    for k, v in target.items():
+        assert props[k] == v
+
+    files = create_files_from_astore(cas_session.CASTable('astore'))
+    check_input_variables(files, BOSTON_INPUT_VARS)
+
+
 def test_neuralnet_regression(cas_session, boston_dataset):
     target = {
         'tool': 'SAS Visual Data Mining and Machine Learning',
@@ -393,6 +455,34 @@ def test_svm_regression(cas_session, boston_dataset):
     check_input_variables(files, BOSTON_INPUT_VARS)
 
 
+def test_svm_regression_with_nominals(cas_session, boston_dataset):
+    target = {
+        'tool': 'SAS Visual Data Mining and Machine Learning',
+        'targetVariable': 'Price',
+        'scoreCodeType': 'ds2MultiType',
+        'function': 'prediction',
+        'algorithm': 'Support vector machine',
+    }
+
+    cas_session.loadactionset('svm')
+    cas_session.loadactionset('astore')
+
+    tbl = cas_session.upload(boston_dataset).casTable
+
+    tbl.svm.svmTrain(
+        target='Price', inputs=list(boston_dataset.columns[:-1]), nominals=['chas'], saveState='astore'
+    )
+
+    desc = cas_session.astore.describe(rstore='astore', epcode=True)
+    props = _get_model_properties(desc)
+
+    for k, v in target.items():
+        assert props[k] == v
+
+    files = create_files_from_astore(cas_session.CASTable('astore'))
+    check_input_variables(files, BOSTON_INPUT_VARS)
+
+
 def test_bayesnet_binary_classification(cas_session, cancer_dataset):
     target = {
         'tool': 'SAS Visual Data Mining and Machine Learning',