Predictive Test Selection: AI-Driven Test Optimization

class CodeTestMapper:
    def __init__(self):
        self.code_test_map = defaultdict(set)
        self.test_coverage = {}

    def analyze_coverage(self, test_run_data):
        """Build mapping from coverage data"""
        for test_name, coverage_data in test_run_data.items():
            covered_files = coverage_data['files']

            for file_path in covered_files:
                self.code_test_map[file_path].add(test_name)

            self.test_coverage[test_name] = {
                'files': covered_files,
                'lines': coverage_data['lines_covered']
            }

    def get_affected_tests(self, changed_files):
        """Get tests affected by code changes"""
        affected = set()

        for file_path in changed_files:
            affected.update(self.code_test_map.get(file_path, set()))

        return list(affected)

# Usage
mapper = CodeTestMapper()
mapper.analyze_coverage(coverage_report)

changed_files = git_diff.get_modified_files()
tests_to_run = mapper.get_affected_tests(changed_files)
print(f"Run {len(tests_to_run)} tests instead of {total_tests}")

from sklearn.ensemble import RandomForestClassifier
import pandas as pd

class TestFailurePredictor (as discussed in [AI Test Metrics Analytics: Intelligent Analysis of QA Metrics](/blog/ai-test-metrics)):
    def __init__(self):
        self.model = RandomForestClassifier(n_estimators=100)
        self.feature_extractor = FeatureExtractor()

    def extract_features(self, commit, test):
        """Extract features for prediction"""
        return {
            # Code change features
            'files_changed': len(commit['files']),
            'lines_added': commit['additions'],
            'lines_deleted': commit['deletions'],
            'complexity_change': self.calculate_complexity_delta(commit),

            # Test features
            'test_execution_time_ms': test['avg_duration'],
            'test_flakiness_score': test['flakiness'],
            'days_since_last_failure' (as discussed in [AI-Assisted Bug Triaging: Intelligent Defect Prioritization at Scale](/blog/ai-bug-triaging)): test['days_since_failure'],
            'failure_rate_30d': test['failures_last_30_days'] / test['runs_last_30_days'],

            # Developer features
            'author_test_failure_rate': commit['author_failure_rate'],
            'commit_hour': commit['timestamp'].hour,
            'is_friday_afternoon': commit['timestamp'].weekday() == 4 and commit['timestamp'].hour >= 14,

            # Change location features
            'changes_in_test_file': test['file_path'] in commit['files'],
            'changes_in_dependencies': self.has_dependency_changes(commit, test)
        }

    def train(self, historical_data):
        """Train on historical test outcomes"""
        features = []
        labels = []

        for commit, test, outcome in historical_data:
            feature_vector = self.extract_features(commit, test)
            features.append(list(feature_vector.values()))
            labels.append(1 if outcome == 'failed' else 0)

        self.model.fit(features, labels)

    def predict_failure_probability(self, commit, test):
        """Predict probability that test will fail"""
        features = self.extract_features(commit, test)
        feature_vector = [list(features.values())]

        probability = self.model.predict_proba(feature_vector)[0][1]

        return {
            'test': test['name'],
            'failure_probability': probability,
            'features': features
        }

# Usage
predictor = TestFailurePredictor()
predictor.train(load_test_history(days=90))

for test in all_tests:
    prediction = predictor.predict_failure_probability(current_commit, test)

    if prediction['failure_probability'] > 0.3:  # High risk
        priority_tests.append(test)

class TestPrioritizer:
    def __init__(self, predictor, mapper):
        self.predictor = predictor
        self.mapper = mapper

    def calculate_test_value(self, test, commit):
        """Calculate value score for test"""
        failure_prob = self.predictor.predict_failure_probability(commit, test)['failure_probability']
        code_coverage = test['line_coverage'] / total_lines
        bug_detection_history = test['bugs_caught_last_year']
        execution_cost = test['avg_duration_ms'] / 1000  # seconds

        # Value = (Failure Risk × Coverage × Bug History) / Cost
        value_score = (failure_prob * code_coverage * bug_detection_history) / max(execution_cost, 1)

        return value_score

    def prioritize(self, commit, time_budget_seconds):
        """Select tests to maximize value within time budget"""
        all_tests = self.mapper.get_test_catalog()

        # Calculate value for each test
        test_scores = [
            {
                'test': test,
                'value': self.calculate_test_value(test, commit),
                'duration': test['avg_duration_ms'] / 1000
            }
            for test in all_tests
        ]

        # Sort by value (descending)
        test_scores.sort(key=lambda x: x['value'], reverse=True)

        # Greedy selection within budget
        selected_tests = []
        total_time = 0

        for item in test_scores:
            if total_time + item['duration'] <= time_budget_seconds:
                selected_tests.append(item['test'])
                total_time += item['duration']

        return {
            'selected_tests': selected_tests,
            'estimated_duration': total_time,
            'coverage': len(selected_tests) / len(all_tests)
        }

# Usage
prioritizer = TestPrioritizer(predictor, mapper)

selection = prioritizer.prioritize(
    commit=current_commit,
    time_budget_seconds=600  # 10 minutes
)

print(f"Running {len(selection['selected_tests'])} highest-value tests")
print(f"Estimated time: {selection['estimated_duration']:.0f}s")
print(f"Coverage: {selection['coverage']:.1%} of test suite")

name: Intelligent Test Selection

on: [push, pull_request]

jobs:
  test:
    runs-on: ubuntu-latest

    steps:

    - uses: actions/checkout@v3
      with:
        fetch-depth: 0  # Full history for analysis

    - name: Analyze Code Changes
      id: changes
      run: |
        CHANGED_FILES=$(git diff --name-only ${{ github.event.before }} ${{ github.sha }})
        echo "files=$CHANGED_FILES" >> $GITHUB_OUTPUT

    - name: Predict Test Selection
      id: selection
      env:
        CHANGED_FILES: ${{ steps.changes.outputs.files }}
      run: |
        python predict_tests.py \
          --changed-files "$CHANGED_FILES" \
          --time-budget 600 \
          --output selected_tests.json

    - name: Run Selected Tests
      run: |
        pytest $(cat selected_tests.json | jq -r '.tests[]')

    - name: Record Outcomes
      if: always()
      run: |
        python record_results.py \
          --commit ${{ github.sha }} \
          --results test-results.xml

class TestImpactAnalyzer:
    def __init__(self):
        self.impact_graph = nx.DiGraph()

    def build_impact_graph(self, codebase):
        """Build dependency graph"""
        # Add nodes
        for file in codebase.files:
            self.impact_graph.add_node(file.path, type='code')

        for test in codebase.tests:
            self.impact_graph.add_node(test.name, type='test')

        # Add edges (dependencies)
        for test in codebase.tests:
            for covered_file in test.coverage:
                self.impact_graph.add_edge(covered_file, test.name)

        # Add code-to-code dependencies
        for file in codebase.files:
            for imported_file in file.imports:
                self.impact_graph.add_edge(imported_file, file.path)

    def get_impacted_tests(self, changed_files):
        """Find all transitively impacted tests"""
        impacted = set()

        for changed_file in changed_files:
            # Find all reachable tests (transitive dependencies)
            reachable = nx.descendants(self.impact_graph, changed_file)

            for node in reachable:
                if self.impact_graph.nodes[node]['type'] == 'test':
                    impacted.add(node)

        return list(impacted)

class FlakinessFilter:
    def __init__(self, flakiness_threshold=0.1):
        self.threshold = flakiness_threshold

    def calculate_flakiness(self, test_history):
        """Calculate test flakiness score"""
        if len(test_history) < 10:
            return 0  # Not enough data

        # Count inconsistent results on same code
        flaky_instances = 0

        for commit_sha in set(test_history['commit']):
            commit_runs = test_history[test_history['commit'] == commit_sha]

            if len(commit_runs) > 1:
                outcomes = commit_runs['outcome'].unique()
                if len(outcomes) > 1:  # Different outcomes on same code
                    flaky_instances += 1

        flakiness = flaky_instances / len(set(test_history['commit']))

        return flakiness

    def should_always_run(self, test):
        """Decide if test is too flaky for intelligent selection"""
        if self.calculate_flakiness(test['history']) > self.threshold:
            return True  # Always run flaky tests to gather data

        return False

class SelectionMetrics:
    def __init__(self):
        self.metrics = []

    def record_selection(self, commit, selected, skipped, outcomes):
        """Record selection effectiveness"""
        selected_failures = [t for t in selected if outcomes[t] == 'failed']
        skipped_failures = [t for t in skipped if outcomes[t] == 'failed']

        self.metrics.append({
            'commit': commit,
            'tests_selected': len(selected),
            'tests_skipped': len(skipped),
            'time_saved_percent': len(skipped) / (len(selected) + len(skipped)),
            'caught_failures': len(selected_failures),
            'missed_failures': len(skipped_failures),  # False negatives
            'precision': len(selected_failures) / len(selected) if selected else 0,
            'recall': len(selected_failures) / (len(selected_failures) + len(skipped_failures)) if (selected_failures or skipped_failures) else 1.0
        })

    def get_dashboard(self):
        """Generate metrics dashboard"""
        df = pd.DataFrame(self.metrics)

        return {
            'avg_time_saved': df['time_saved_percent'].mean(),
            'avg_recall': df['recall'].mean(),  # What % of failures we catch
            'total_missed_failures': df['missed_failures'].sum(),
            'tests_per_commit': df['tests_selected'].mean()
        }