teradataml 17.20.0.6__py3-none-any.whl → 20.0.0.0__py3-none-any.whl

teradataml/automl/model_evaluation.py

@@ -0,0 +1,163 @@
+# ################################################################## 
+# 
+# Copyright 2024 Teradata. All rights reserved.
+# TERADATA CONFIDENTIAL AND TRADE SECRET
+# 
+# Primary Owner: Sweta Shaw
+# Email Id: Sweta.Shaw@Teradata.com
+# 
+# Secondary Owner: Akhil Bisht
+# Email Id: AKHIL.BISHT@Teradata.com
+# 
+# Version: 1.1
+# Function Version: 1.0
+# ##################################################################
+
+# Python libraries
+import time
+
+# Teradata libraries
+from teradataml.dataframe.dataframe import DataFrame
+
+
+class _ModelEvaluator:
+    
+    def __init__(self, 
+                 df=None, 
+                 target_column=None, 
+                 task_type=None):
+        """
+        DESCRIPTION:
+            Function initializes the data, target column, features and models
+            for model evaluation.
+         
+        PARAMETERS:  
+            df:
+                Required Argument.
+                Specifies the model information.
+                Types: teradataml Dataframe
+            
+            target_column:
+                Required Arugment.
+                Specifies the target column present inside the dataset.
+                Types: str
+                
+            task_type:
+                Required Arugment.
+                Specifies the task type for AutoML, whether to apply regresion OR classification
+                on the provived dataset.
+                Default Value: "Regression"
+                Permitted Values: "Regression", "Classification"
+                Types: str
+
+        """
+        self.model_info = df
+        self.target_column = target_column
+        self.task_type = task_type
+        
+    def model_evaluation(self, rank, table_name_mapping, test_data_ind = False, target_column_ind = False):
+        """
+        DESCRIPTION:
+            Function performs the model evaluation on the specified rank in leaderborad.
+         
+        PARAMETERS:  
+            rank:
+                Required Argument.
+                Specifies the position of ML model for evaluation.
+                Types: int
+                        
+            table_name_mapping:
+                Required Argument.
+                Specifies the mapping of train,test table names.
+                Types: dict
+            
+            test_data_ind:
+                Optional Argument.
+                Specifies whether test data is present or not.
+                Default Value: False
+                Types: bool
+            
+            target_column_ind:
+                Optional Argument.
+                Specifies whether target column is present in the dataset or not.
+                Default Value: False
+                
+        RETURNS:
+            tuple containing, performance metrics and predicitions of specified rank ML model.
+             
+        """
+        # Setting test data indicator and target column indicator
+        self.test_data_ind = test_data_ind
+        self.target_column_ind = target_column_ind
+        self.table_name_mapping = table_name_mapping
+        
+        # Return predictions only if test data is present and target column is not present
+        return self._evaluator(rank)
+
+    def _evaluator(self,
+                   rank):
+        """
+        DESCRIPTION:
+            Internal Function runs evaluator function for specified rank ML model
+            based on regression/classification problem.
+         
+        PARAMETERS:  
+            rank:
+                Required Argument.
+                Specifies the position(rank) of ML model for evaluation.
+                Types: int
+                
+        RETURNS:
+            tuple containing, performance metrics and predictions of ML model.
+             
+        """
+        # Extracting model using rank
+        model = self.model_info.loc[rank]
+        
+        # Defining eval_params 
+        eval_params = {"id_column": "id",
+                        "accumulate": self.target_column}
+
+        # eval_params for Classification
+        if self.task_type != "Regression":
+            # XGboost
+            if model['Name'] == 'xgboost':
+                eval_params['model_type'] = 'Classification'
+                eval_params['object_order_column'] = ['task_index', 'tree_num', 'iter','class_num', 'tree_order']
+            else:
+            # DF,KNN,SVM,GLM
+                eval_params['output_prob'] = True
+        else:
+        # eval_params for Regression in XGboost
+            if model['Name'] == 'xgboost':
+                eval_params['model_type'] = 'Regression'
+                eval_params['object_order_column'] = ['task_index', 'tree_num', 'iter', 'tree_order']
+                    
+        
+        # Test Data
+        test = DataFrame(self.table_name_mapping['{}_test'.format(model['Feature selection'])])
+
+        # Getting test data from table
+        if not self.test_data_ind:
+            # Test Data
+            test = DataFrame(self.table_name_mapping['{}_test'.format(model['Feature selection'])])
+        else:
+            test = DataFrame(self.table_name_mapping['{}_new_test'.format(model['Feature selection'])])
+
+        print(model['Name'], model['Feature selection'])
+        
+        # Evaluation and predictions 
+        if model['Name'] == 'knn':
+            metrics = model['model-obj'].evaluate(test_data=test)
+            pred = model['model-obj'].predict(test_data=test)
+        else:
+            # Return predictions only if test data is present and target column is not present
+            if self.test_data_ind and not self.target_column_ind:
+                eval_params.pop("accumulate")
+                pred = model['model-obj'].predict(newdata=test, **eval_params)
+                return pred
+            # Return both metrics and predictions for all other cases
+            metrics = model['model-obj'].evaluate(newdata=test, **eval_params)
+            pred = model['model-obj'].predict(newdata=test, **eval_params)
+        
+        return (metrics, pred)