Merge pull request #84 from VivekSinghDS/hotfix/qa-test-methods

Hotfix/qa test methods

toolkit/src/qa/qa.py

@@ -1,7 +1,8 @@
 from abc import ABC, abstractmethod
 from typing import Union, List, Tuple, Dict
-
-
+import pandas as pd
+from toolkit.src.ui.rich_ui import RichUI
+import statistics
 
 class LLMQaTest(ABC):
     @property
@@ -37,7 +38,6 @@ def run_tests(self) -> Dict[str, List[Union[float, int, bool]]]:
             test_results[test.test_name] = metrics
 
         self.test_results = test_results
-
         return test_results
 
     @property
@@ -46,9 +46,16 @@ def test_results(self):
 
 
     def print_test_results(self):
-        # TODO: format these!
-        print(self.test_results)
-
+        result_dictionary = self.test_results()
+        column_data = {key: [value for value in result_dictionary[key]] for key in result_dictionary}
+        mean_values = {key: statistics.mean(column_data[key]) for key in column_data}
+        median_values = {key: statistics.median(column_data[key]) for key in column_data}
+        stdev_values = {key: statistics.stdev(column_data[key]) for key in column_data}
+        # Use the RichUI class to display the table
+        RichUI.display_table(result_dictionary, mean_values, median_values, stdev_values)
+
     def save_test_results(self, path:str):
         # TODO: save these!
-        pass
+        resultant_dataframe = pd.DataFrame(self.test_results())
+        resultant_dataframe.to_csv(path, index = False)
+

toolkit/src/ui/rich_ui.py

@@ -183,3 +183,27 @@ def after_qa():
     @staticmethod
     def qa_found():
         pass
+
+    @staticmethod
+    def qa_display_table(self, result_dictionary, mean_values, median_values, stdev_values):
+
+        # Create a table
+        table = Table(show_header=True, header_style="bold", title="Test Results")
+
+        # Add columns to the table
+        table.add_column("Metric", style="cyan")
+        table.add_column("Mean", style="magenta")
+        table.add_column("Median", style="green")
+        table.add_column("Standard Deviation", style="yellow")
+
+        # Add data rows to the table
+        for key in result_dictionary:
+            table.add_row(
+                key,
+                f"{mean_values[key]:.4f}",
+                f"{median_values[key]:.4f}",
+                f"{stdev_values[key]:.4f}"
+            )
+
+        # Print the table
+        console.print(table)

toolkit/src/ui/ui.py

@@ -110,3 +110,7 @@ def after_qa(cls):
     @abstractstaticmethod
     def qa_found(cls):
         pass
+
+    @abstractstaticmethod
+    def qa_display_table(cls):
+        pass