test_end_to_end.py

import tisane as ts
from tisane.effect_set import EffectSet, MainEffect, InteractionEffect, MixedEffect
from tisane.statistical_model import StatisticalModel

import unittest

class EndToEndTests(unittest.TestCase): 
    
    # def test_generate_effects_sets(self):
    #     analysis = ts.Tisane(task="explanation") # analysis has one task

    #     test_score = ts.Concept("Test Score")
    #     intelligence = ts.Concept("Intelligence")
    #     tutoring = ts.Concept("Tutoring")
    #     concepts = [test_score, intelligence, tutoring]

    #     analysis.addRelationship(intelligence, test_score, "cause")
    #     analysis.addRelationship(tutoring, test_score, "cause")
    #     analysis.addRelationship(intelligence, tutoring, "correlate")

    #     effects = analysis.generate_effects_sets(ivs=[intelligence, tutoring], dv=test_score)
        
    #     # check total number of effect sets 
    #     self.assertEqual(len(effects), 7)
    #     # check each effect set is valid
    #     for es in effects: 
    #         es_dict = es.to_dict()
    #         self.assertTrue(es_dict in DataForTests.expected_effects_set)
    

    # def test_effects_ASP(self):
    #     analysis = ts.Tisane(task="explanation")

    #     # Add concepts
    #     test_score = ts.Concept("Test Score")
    #     intelligence = ts.Concept("Intelligence")
    #     tutoring = ts.Concept("Tutoring")
    #     concepts = [test_score, intelligence, tutoring]

    #     # Add relationships
    #     analysis.addRelationship(intelligence, test_score, "cause")
    #     analysis.addRelationship(tutoring, test_score, "cause")
    #     analysis.addRelationship(intelligence, tutoring, "correlate")

    #     # Specify data (schema)
    #     test_score.specifyData(dtype="numeric")
    #     intelligence.specifyData(dtype="numeric")
    #     tutoring.specifyData(dtype="nominal", categories=["After school", "Before school"])
    #     # tutoring.specifyData(dtype="nominal", categories=["After school", "Before school", "None"])

    #     # Get valid statistical models
    #     # Assert statistical properties needed for linear regression
    #     # Generate effects sets
    #     effects = analysis.generate_effects_sets(ivs=[intelligence, tutoring], dv=test_score)

    #     # Add individual variable assertions
    #     # Some assertions are inferred from data schema, see above
    #     self.assertTrue(test_score.getVariable().has_property_value(prop="dtype", val="numeric"))
    #     self.assertTrue(intelligence.getVariable().has_property_value(prop="dtype", val="numeric"))
    #     self.assertTrue(tutoring.getVariable().has_property_value(prop="dtype", val="nominal"))
        
    #     self.assertFalse(test_score.getVariable().has_property(prop="cardinality"))
    #     self.assertFalse(intelligence.getVariable().has_property(prop="cardinality"))
    #     self.assertTrue(tutoring.getVariable().has_property_value(prop="cardinality", val="binary"))

    #     # Add assertions that pertain to effects sets
    #     linear_reg_es = None
    #     for es in effects: 
    #         if es.to_dict() == DataForTests.expected_effects_set[2]: 
    #             linear_reg_es = es
    #             break
    #     linear_reg_es.assert_property(prop="tolerate_correlation", val=True)
        
    #     # Convert linear regression model EffectSet to statistical model
    #     linear_reg_sm = StatisticalModel.create(model_type='linear_regression', effect_set=linear_reg_es)
        
    #     # Add assertions that pertain to the models
    #     # Note: Make assertions that will involve interaction (interactions compile to these statements)
    #     # Note: These assertions should be made on the *MODEL* not the *Effect Set*
    #     linear_reg_sm.get_residuals().assert_property(prop="distribution", val="normal")
    #     linear_reg_sm.get_residuals().assert_property(prop="homoscedastic", val=True)
        
    #     # TODO: Compile all these assertions into constraints
    #     # Should be in KnowledgeBase class or main Tisane class?
        
    #     # Query KB for statistical models
    #     analysis.start_model(linear_reg_es) 

    def test_sample_tisane_program(self): 
        analysis = ts.Tisane(task="explanation")

        ### PHASE 0: CONCEPTUAL RELATIONSHIPS
        # Add concepts
        test_score = ts.Concept("Test Score")
        intelligence = ts.Concept("Intelligence")
        tutoring = ts.Concept("Tutoring")
    
        # Add relationships
        analysis.addRelationship(intelligence, test_score, "cause")
        analysis.addRelationship(tutoring, test_score, "cause")
        analysis.addRelationship(intelligence, tutoring, "correlate")
        

        ### PHASE 1: IV, DV SPECIFICATION
        ### PHASE 2: EFFECTS SETS GENERATION 
        # Generate effects sets
        effects = analysis.generate_effects_sets(ivs=[intelligence, tutoring], dv=test_score)

        ### PHASE 3A: ASSERTIONS ABOUT VARIABLES AND SETS OF VARIABLES
        # Specify data (schema) --> generates assertions about variables/data
        # If have data: automatically detect and verify these based on uploaded data
        test_score.specifyData(dtype="numeric")
        intelligence.specifyData(dtype="numeric")
        tutoring.specifyData(dtype="nominal", categories=["After school", "Before school"])

        # THIS MIMICS END-USERS SELECTING A SET OF EFFECTS TO MODEL 
        linear_reg_es = None
        for es in effects: 
            if es.to_dict() == DataForTests.expected_effects_set[2]: 
                linear_reg_es = es
                break
        # Add assertions that pertain to effects sets
        linear_reg_es.assert_property(prop="tolerate_correlation", val=True)
        linear_reg_es.assert_property(prop="normal_residuals", val=True)
        linear_reg_es.assert_property(prop="homoscedastic_residuals", val=True)

        # PHASE 3B: QUERYING KNOWLEDGE BASE
        # TODO: TEST INCREMENTAL SOLVING ASPECT
        valid_models = analysis.start_model(effect_set=linear_reg_es) 
        print(valid_models) # "Linear Regression"
        
class DataForTests: 
    test_score = ts.Concept("Test Score")
    intelligence = ts.Concept("Intelligence")
    tutoring = ts.Concept("Tutoring")

    expected_effects_set = [
        {'dv': test_score.name, 'main': set({intelligence.name}), 'interaction': None, 'mixed': None},
        {'dv': test_score.name, 'main': set({tutoring.name}), 'interaction': None, 'mixed': None},
        {'dv': test_score.name, 'main': set({tutoring.name, intelligence.name}), 'interaction': None, 'mixed': None},
        {'dv': test_score.name, 'main': set({tutoring.name, intelligence.name}), 'interaction': set({(intelligence.name, tutoring.name)}), 'mixed': None},
        {'dv': test_score.name, 'main': set({intelligence.name}), 'interaction': set({(intelligence.name, tutoring.name)}), 'mixed': None},
        {'dv': test_score.name, 'main': set({tutoring.name}), 'interaction': set({(intelligence.name, tutoring.name)}), 'mixed': None},
        {'dv': test_score.name, 'main': None, 'interaction': set({(intelligence.name, tutoring.name)}), 'mixed': None},
    ]