tests/system/test_training_inference.py

"""
System tests for end-to-end workflows.

Tests the complete system including training and inference pipelines.
"""
import pytest
import numpy as np
import pandas as pd
import tempfile
from pathlib import Path
from unittest.mock import patch, MagicMock
import joblib

from sklearn.ensemble import RandomForestClassifier
from sklearn.multioutput import MultiOutputClassifier


@pytest.mark.system
@pytest.mark.slow
class TestTrainingPipeline:
    """System tests for model training pipeline."""
    
    def test_complete_training_workflow(self, sample_dataframe):
        """Test complete training workflow from data to model."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        from sklearn.model_selection import train_test_split
        
        # Extract features
        features, vectorizer = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        # Split data
        X_train, X_test, y_train, y_test = train_test_split(
            features, labels.values, test_size=0.2, random_state=42
        )
        
        # Train model
        rf = RandomForestClassifier(n_estimators=10, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(X_train, y_train)
        
        # Predict
        predictions = model.predict(X_test)
        
        # Verify
        assert predictions.shape[0] == X_test.shape[0]
        assert predictions.shape[1] == y_test.shape[1]
        assert np.all((predictions == 0) | (predictions == 1))
    
    def test_training_with_oversampling(self, sample_dataframe):
        """Test training pipeline with oversampling."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        from imblearn.over_sampling import RandomOverSampler
        from sklearn.model_selection import train_test_split
        
        # Prepare data
        features, _ = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        # Use only one label column for oversampling
        y_single = labels.iloc[:, 0].values
        
        # Split
        X_train, X_test, y_train, y_test = train_test_split(
            features, y_single, test_size=0.2, random_state=42
        )
        
        # Oversample
        ros = RandomOverSampler(random_state=42)
        X_resampled, y_resampled = ros.fit_resample(X_train, y_train)
        
        # Train
        rf = RandomForestClassifier(n_estimators=10, random_state=42)
        rf.fit(X_resampled, y_resampled)
        
        # Predict
        predictions = rf.predict(X_test)
        
        # Verify
        assert len(predictions) == len(X_test)
        assert np.all((predictions == 0) | (predictions == 1))
    
    def test_model_serialization(self, sample_dataframe):
        """Test model can be saved and loaded."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        
        # Train simple model
        features, _ = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(features, labels.values)
        
        # Save and load
        with tempfile.NamedTemporaryFile(suffix='.pkl', delete=False) as f:
            model_path = f.name
        
        try:
            joblib.dump(model, model_path)
            loaded_model = joblib.load(model_path)
            
            # Verify predictions match
            pred_original = model.predict(features)
            pred_loaded = loaded_model.predict(features)
            
            np.testing.assert_array_equal(pred_original, pred_loaded)
        finally:
            Path(model_path).unlink()


@pytest.mark.system
class TestInferencePipeline:
    """System tests for inference pipeline."""
    
    def test_inference_on_new_text(self, sample_dataframe):
        """Test inference pipeline on new unseen text."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
            clean_github_text,
        )
        
        # Train model
        features, vectorizer = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(features, labels.values)
        
        # New text
        new_texts = [
            "Fixed critical bug in authentication module",
            "Added new REST API endpoint for users",
        ]
        
        # Process new text
        cleaned_texts = [clean_github_text(text) for text in new_texts]
        new_features = vectorizer.transform(cleaned_texts).toarray()
        
        # Predict
        predictions = model.predict(new_features)
        
        # Verify
        assert predictions.shape[0] == len(new_texts)
        assert predictions.shape[1] == labels.shape[1]
        assert np.all((predictions == 0) | (predictions == 1))
    
    def test_inference_with_empty_input(self, sample_dataframe):
        """Test inference handles empty input gracefully."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
            clean_github_text,
        )
        
        # Train model
        features, vectorizer = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(features, labels.values)
        
        # Empty text
        empty_text = ""
        cleaned = clean_github_text(empty_text)
        new_features = vectorizer.transform([cleaned]).toarray()
        
        # Should not crash
        predictions = model.predict(new_features)
        
        assert predictions.shape[0] == 1
        assert predictions.shape[1] == labels.shape[1]
    
    def test_batch_inference(self, sample_dataframe):
        """Test inference on batch of samples."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        
        # Train model
        features, vectorizer = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(features, labels.values)
        
        # Batch prediction
        predictions = model.predict(features)
        
        assert predictions.shape == labels.shape
        assert np.all((predictions == 0) | (predictions == 1))


@pytest.mark.system
@pytest.mark.requires_data
class TestEndToEndDataFlow:
    """System tests for complete data flow from raw to predictions."""
    
    def test_full_pipeline_database_to_predictions(self, temp_db):
        """Test complete pipeline from database to predictions."""
        from hopcroft_skill_classification_tool_competition.features import (
            load_data_from_db,
            extract_tfidf_features,
            prepare_labels,
        )
        from sklearn.model_selection import train_test_split
        
        # Load data
        df = load_data_from_db(temp_db)
        
        # Extract features
        features, vectorizer = extract_tfidf_features(df, max_features=50)
        labels = prepare_labels(df)
        
        # Split
        X_train, X_test, y_train, y_test = train_test_split(
            features, labels.values, test_size=0.4, random_state=42
        )
        
        # Train
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(X_train, y_train)
        
        # Predict
        predictions = model.predict(X_test)
        
        # Evaluate (simple check)
        from sklearn.metrics import accuracy_score
        
        # Per-label accuracy
        accuracies = []
        for i in range(y_test.shape[1]):
            acc = accuracy_score(y_test[:, i], predictions[:, i])
            accuracies.append(acc)
        
        # Should have some predictive power (better than random for at least one label)
        assert np.mean(accuracies) > 0.4  # Very lenient threshold for small test data


@pytest.mark.system
class TestModelValidation:
    """System tests for model validation workflows."""
    
    def test_cross_validation_workflow(self, sample_dataframe):
        """Test cross-validation workflow."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        from sklearn.model_selection import cross_val_score
        
        # Prepare data
        features, _ = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        # Use single label for CV
        y_single = labels.iloc[:, 0].values
        
        # Cross-validation
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        
        # Should not crash (though scores may be poor with small data)
        scores = cross_val_score(rf, features, y_single, cv=2, scoring='accuracy')
        
        assert len(scores) == 2
        assert all(0 <= score <= 1 for score in scores)
    
    def test_grid_search_workflow(self, sample_dataframe):
        """Test grid search workflow."""
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        from sklearn.model_selection import GridSearchCV
        
        # Prepare data
        features, _ = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        # Use single label
        y_single = labels.iloc[:, 0].values
        
        # Small grid search
        param_grid = {
            'n_estimators': [5, 10],
            'max_depth': [5, 10],
        }
        
        rf = RandomForestClassifier(random_state=42)
        grid_search = GridSearchCV(rf, param_grid, cv=2, scoring='accuracy')
        grid_search.fit(features, y_single)
        
        # Verify
        assert hasattr(grid_search, 'best_params_')
        assert hasattr(grid_search, 'best_score_')
        assert grid_search.best_score_ >= 0


@pytest.mark.system
@pytest.mark.regression
class TestRegressionScenarios:
    """Regression tests for known issues and edge cases."""
    
    def test_empty_feature_vectors_handling(self):
        """
        Regression test: Ensure empty feature vectors don't crash training.
        
        This was identified in Great Expectations TEST 2 - 25 samples with
        zero features after TF-IDF extraction.
        """
        from sklearn.ensemble import RandomForestClassifier
        from sklearn.multioutput import MultiOutputClassifier
        
        # Create data with some zero vectors
        X = np.array([
            [0.1, 0.2, 0.3],
            [0.0, 0.0, 0.0],  # Empty vector
            [0.4, 0.5, 0.6],
            [0.0, 0.0, 0.0],  # Another empty vector
        ])
        
        y = np.array([
            [1, 0],
            [0, 1],
            [1, 1],
            [0, 0],
        ])
        
        # Should not crash
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(X, y)
        
        predictions = model.predict(X)
        assert predictions.shape == y.shape
    
    def test_zero_occurrence_labels_handling(self):
        """
        Regression test: Handle labels with zero occurrences.
        
        This was identified in Great Expectations TEST 5 - 75 labels with
        zero occurrences in the dataset.
        """
        from hopcroft_skill_classification_tool_competition.features import get_label_columns
        
        # Create dataframe with some zero-occurrence labels
        df = pd.DataFrame({
            'issue text': ['text1', 'text2', 'text3'],
            'Label1': [1, 1, 0],  # Has occurrences
            'Label2': [0, 0, 0],  # Zero occurrences
            'Label3': [1, 0, 1],  # Has occurrences
        })
        
        label_cols = get_label_columns(df)
        
        # Should include all labels
        assert 'Label1' in label_cols
        assert 'Label2' in label_cols
        assert 'Label3' in label_cols
        
        # Training code should filter these out before stratification
        # This test just verifies detection works
    
    def test_high_sparsity_features(self):
        """
        Regression test: Handle very sparse features (>99% zeros).
        
        This was identified in Great Expectations TEST 6 - 99.88% sparsity.
        """
        from sklearn.ensemble import RandomForestClassifier
        
        # Create highly sparse feature matrix
        X = np.zeros((100, 1000))
        
        # Only 0.12% non-zero values (very sparse)
        for i in range(100):
            indices = np.random.choice(1000, size=1, replace=False)
            X[i, indices] = np.random.rand(1)
        
        y = np.random.randint(0, 2, size=100)
        
        # Should handle high sparsity without crashing
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        rf.fit(X, y)
        
        predictions = rf.predict(X)
        assert len(predictions) == len(y)
    
    def test_duplicate_samples_detection(self):
        """
        Regression test: Detect duplicate samples.
        
        This was identified in Deepchecks validation - 481 duplicates (6.72%).
        """
        df = pd.DataFrame({
            'issue text': ['duplicate', 'duplicate', 'unique'],
            'issue description': ['desc', 'desc', 'different'],
            'Label1': [1, 1, 0],
        })
        
        # Check for duplicates
        duplicates = df[['issue text', 'issue description']].duplicated()
        
        assert duplicates.sum() == 1  # One duplicate found
        
        # Removal should be done in data cleaning pipeline
        df_cleaned = df.drop_duplicates(subset=['issue text', 'issue description'])
        assert len(df_cleaned) == 2


@pytest.mark.system
@pytest.mark.acceptance
class TestAcceptanceCriteria:
    """Acceptance tests verifying requirements are met."""
    
    def test_multi_label_classification_support(self, sample_dataframe):
        """
        Acceptance test: System supports multi-label classification.
        
        Requirement: Each issue can have multiple skill labels.
        """
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        
        features, _ = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        # Train multi-output model
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(features, labels.values)
        
        # Predict multiple labels
        predictions = model.predict(features)
        
        # Verify multiple labels can be predicted
        labels_per_sample = predictions.sum(axis=1)
        assert np.any(labels_per_sample > 1), "System should support multiple labels per sample"
    
    def test_handles_github_text_format(self):
        """
        Acceptance test: System handles GitHub issue text format.
        
        Requirement: Process text from GitHub issues with URLs, code, etc.
        """
        from hopcroft_skill_classification_tool_competition.features import clean_github_text
        
        github_text = """
        Fixed bug in authentication #123
        
        See: https://github.com/repo/issues/123
        
        ```python
        def login(user):
            return authenticate(user)
        ```
        
        Related to <b>security</b> improvements 🔒
        """
        
        cleaned = clean_github_text(github_text)
        
        # Should remove noise but keep meaningful content
        assert "https://" not in cleaned
        assert "```" not in cleaned
        assert "<b>" not in cleaned
        assert len(cleaned) > 0
    
    def test_produces_binary_predictions(self, sample_dataframe):
        """
        Acceptance test: System produces binary predictions (0 or 1).
        
        Requirement: Clear yes/no predictions for each skill.
        """
        from hopcroft_skill_classification_tool_competition.features import (
            extract_tfidf_features,
            prepare_labels,
        )
        
        features, _ = extract_tfidf_features(sample_dataframe, max_features=50)
        labels = prepare_labels(sample_dataframe)
        
        rf = RandomForestClassifier(n_estimators=5, random_state=42)
        model = MultiOutputClassifier(rf)
        model.fit(features, labels.values)
        
        predictions = model.predict(features)
        
        # All predictions should be 0 or 1
        assert np.all((predictions == 0) | (predictions == 1))