api-mocker 0.4.0__py3-none-any.whl → 0.5.1__py3-none-any.whl

api_mocker/ml_integration.py

@@ -0,0 +1,716 @@
+"""
+Machine Learning Integration System
+
+This module provides comprehensive ML capabilities for API mocking including:
+- Intelligent response generation using ML models
+- Request pattern analysis and prediction
+- Anomaly detection for API behavior
+- Smart caching based on usage patterns
+- Automated test case generation
+- Performance optimization recommendations
+"""
+
+import json
+import numpy as np
+import pandas as pd
+from typing import Any, Dict, List, Optional, Union, Tuple, Callable
+from dataclasses import dataclass, field
+from enum import Enum
+from datetime import datetime, timedelta
+import pickle
+import joblib
+from sklearn.ensemble import RandomForestClassifier, IsolationForest
+from sklearn.cluster import KMeans, DBSCAN
+from sklearn.preprocessing import StandardScaler, LabelEncoder
+from sklearn.model_selection import train_test_split
+from sklearn.metrics import accuracy_score, classification_report
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.linear_model import LinearRegression
+from sklearn.neural_network import MLPClassifier
+import asyncio
+import threading
+from collections import defaultdict, deque
+import hashlib
+
+
+class MLModelType(Enum):
+    """ML model types"""
+    CLASSIFICATION = "classification"
+    REGRESSION = "regression"
+    CLUSTERING = "clustering"
+    ANOMALY_DETECTION = "anomaly_detection"
+    TEXT_ANALYSIS = "text_analysis"
+    RECOMMENDATION = "recommendation"
+
+
+class PredictionType(Enum):
+    """Prediction types"""
+    RESPONSE_TIME = "response_time"
+    ERROR_PROBABILITY = "error_probability"
+    USER_BEHAVIOR = "user_behavior"
+    CACHE_HIT = "cache_hit"
+    RESOURCE_USAGE = "resource_usage"
+    ANOMALY_SCORE = "anomaly_score"
+
+
+@dataclass
+class MLModel:
+    """ML model representation"""
+    name: str
+    model_type: MLModelType
+    model: Any
+    features: List[str]
+    target: str
+    accuracy: float = 0.0
+    created_at: datetime = field(default_factory=datetime.now)
+    last_trained: datetime = field(default_factory=datetime.now)
+    training_samples: int = 0
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class PredictionRequest:
+    """Prediction request"""
+    features: Dict[str, Any]
+    model_name: str
+    prediction_type: PredictionType
+    confidence_threshold: float = 0.5
+
+
+@dataclass
+class PredictionResult:
+    """Prediction result"""
+    prediction: Any
+    confidence: float
+    model_name: str
+    features_used: List[str]
+    timestamp: datetime = field(default_factory=datetime.now)
+    metadata: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class TrainingData:
+    """Training data for ML models"""
+    features: List[Dict[str, Any]]
+    targets: List[Any]
+    feature_names: List[str]
+    target_name: str
+    created_at: datetime = field(default_factory=datetime.now)
+
+
+class FeatureExtractor:
+    """Feature extraction for ML models"""
+    
+    def __init__(self):
+        self.text_vectorizer = TfidfVectorizer(max_features=1000, stop_words='english')
+        self.label_encoders: Dict[str, LabelEncoder] = {}
+        self.scaler = StandardScaler()
+    
+    def extract_request_features(self, request_data: Dict[str, Any]) -> Dict[str, Any]:
+        """Extract features from API request data"""
+        features = {}
+        
+        # Basic features
+        features['method_encoded'] = self._encode_categorical(request_data.get('method', ''), 'method')
+        features['path_length'] = len(request_data.get('path', ''))
+        features['has_query_params'] = 1 if '?' in request_data.get('path', '') else 0
+        features['has_path_params'] = 1 if '{' in request_data.get('path', '') else 0
+        
+        # Header features
+        headers = {k.lower(): v for k, v in request_data.get('headers', {}).items()}
+        features['header_count'] = len(headers)
+        features['has_auth_header'] = 1 if 'authorization' in headers else 0
+        features['has_content_type'] = 1 if 'content-type' in headers else 0
+        
+        # Body features
+        body = request_data.get('body', '')
+        if isinstance(body, str):
+            features['body_length'] = len(body)
+            features['is_json'] = 1 if body.startswith('{') or body.startswith('[') else 0
+        else:
+            features['body_length'] = 0
+            features['is_json'] = 0
+        
+        # Time-based features
+        now = datetime.now()
+        features['hour_of_day'] = now.hour
+        features['day_of_week'] = now.weekday()
+        features['is_weekend'] = 1 if now.weekday() >= 5 else 0
+        
+        return features
+    
+    def extract_response_features(self, response_data: Dict[str, Any]) -> Dict[str, Any]:
+        """Extract features from API response data"""
+        features = {}
+        
+        features['status_code'] = response_data.get('status_code', 200)
+        features['is_success'] = 1 if 200 <= features['status_code'] < 300 else 0
+        features['is_client_error'] = 1 if 400 <= features['status_code'] < 500 else 0
+        features['is_server_error'] = 1 if 500 <= features['status_code'] < 600 else 0
+        
+        # Response body features
+        body = response_data.get('body', '')
+        if isinstance(body, str):
+            features['response_length'] = len(body)
+        else:
+            features['response_length'] = 0
+        
+        # Header features
+        headers = response_data.get('headers', {})
+        features['response_header_count'] = len(headers)
+        features['has_cache_header'] = 1 if 'cache-control' in headers else 0
+        
+        return features
+    
+    def _encode_categorical(self, value: str, field_name: str) -> int:
+        """Encode categorical values"""
+        if field_name not in self.label_encoders:
+            self.label_encoders[field_name] = LabelEncoder()
+            # Fit with known values
+            known_values = ['GET', 'POST', 'PUT', 'DELETE', 'PATCH', 'HEAD', 'OPTIONS']
+            self.label_encoders[field_name].fit(known_values)
+        
+        try:
+            return self.label_encoders[field_name].transform([value])[0]
+        except ValueError:
+            return 0  # Unknown value
+
+
+class MLModelManager:
+    """ML model management system"""
+    
+    def __init__(self):
+        self.models: Dict[str, MLModel] = {}
+        self.feature_extractor = FeatureExtractor()
+        self.training_data: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
+        self.model_lock = threading.Lock()
+    
+    def add_model(self, model: MLModel) -> None:
+        """Add a model to the manager"""
+        with self.model_lock:
+            self.models[model.name] = model
+    
+    def get_model(self, name: str) -> Optional[MLModel]:
+        """Get a model by name"""
+        return self.models.get(name)
+    
+    def train_model(self, model_name: str, training_data: TrainingData) -> Dict[str, Any]:
+        """Train a model with provided data"""
+        if model_name not in self.models:
+            return {"success": False, "error": "Model not found"}
+        
+        model = self.models[model_name]
+        
+        try:
+            # Prepare features and targets
+            X = []
+            y = []
+            
+            for i, features in enumerate(training_data.features):
+                X.append(list(features.values()))
+                y.append(training_data.targets[i])
+            
+            X = np.array(X)
+            y = np.array(y)
+            
+            # Split data
+            X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
+            
+            # Scale features
+            X_train_scaled = self.feature_extractor.scaler.fit_transform(X_train)
+            X_test_scaled = self.feature_extractor.scaler.transform(X_test)
+            
+            # Train model
+            model.model.fit(X_train_scaled, y_train)
+            
+            # Evaluate
+            y_pred = model.model.predict(X_test_scaled)
+            
+            if model.model_type == MLModelType.REGRESSION:
+                 from sklearn.metrics import r2_score
+                 # R2 score can be negative, but it runs.
+                 # For very simple/random data, it might be poor.
+                 accuracy = r2_score(y_test, y_pred)
+            else:
+                 accuracy = accuracy_score(y_test, y_pred)
+            
+            # Update model
+            model.accuracy = accuracy
+            model.last_trained = datetime.now()
+            model.training_samples = len(training_data.features)
+            
+            return {
+                "success": True,
+                "accuracy": accuracy,
+                "training_samples": len(training_data.features),
+                "test_samples": len(X_test)
+            }
+            
+        except Exception as e:
+            return {"success": False, "error": str(e)}
+    
+    def predict(self, request: PredictionRequest) -> PredictionResult:
+        """Make a prediction using a model"""
+        if request.model_name not in self.models:
+            return PredictionResult(
+                prediction=None,
+                confidence=0.0,
+                model_name=request.model_name,
+                features_used=[],
+                metadata={"error": "Model not found"}
+            )
+        
+        model = self.models[request.model_name]
+        
+        try:
+            # Extract features
+            features = self.feature_extractor.extract_request_features(request.features)
+            feature_values = [features.get(f, 0) for f in model.features]
+            
+            # Scale features
+            feature_array = np.array(feature_values).reshape(1, -1)
+            scaled_features = self.feature_extractor.scaler.transform(feature_array)
+            
+            # Make prediction
+            if model.model_type == MLModelType.CLASSIFICATION:
+                prediction = model.model.predict(scaled_features)[0]
+                confidence = model.model.predict_proba(scaled_features).max()
+            else:
+                prediction = model.model.predict(scaled_features)[0]
+                confidence = 1.0  # For regression, we don't have confidence scores
+            
+            return PredictionResult(
+                prediction=prediction,
+                confidence=confidence,
+                model_name=request.model_name,
+                features_used=model.features,
+                metadata={"model_accuracy": model.accuracy}
+            )
+            
+        except Exception as e:
+            return PredictionResult(
+                prediction=None,
+                confidence=0.0,
+                model_name=request.model_name,
+                features_used=[],
+                metadata={"error": str(e)}
+            )
+    
+    def save_model(self, model_name: str, filepath: str) -> bool:
+        """Save a model to disk"""
+        if model_name not in self.models:
+            return False
+        
+        try:
+            model = self.models[model_name]
+            model_data = {
+                "name": model.name,
+                "model_type": model.model_type.value,
+                "features": model.features,
+                "target": model.target,
+                "accuracy": model.accuracy,
+                "created_at": model.created_at.isoformat(),
+                "last_trained": model.last_trained.isoformat(),
+                "training_samples": model.training_samples,
+                "metadata": model.metadata
+            }
+            
+            # Save model and metadata
+            joblib.dump(model.model, f"{filepath}_model.pkl")
+            with open(f"{filepath}_metadata.json", 'w') as f:
+                json.dump(model_data, f, indent=2)
+            
+            return True
+        except Exception as e:
+            print(f"Error saving model: {e}")
+            return False
+    
+    def load_model(self, model_name: str, filepath: str) -> bool:
+        """Load a model from disk"""
+        try:
+            # Load metadata
+            with open(f"{filepath}_metadata.json", 'r') as f:
+                model_data = json.load(f)
+            
+            # Load model
+            model_obj = joblib.load(f"{filepath}_model.pkl")
+            
+            # Create model
+            model = MLModel(
+                name=model_data["name"],
+                model_type=MLModelType(model_data["model_type"]),
+                model=model_obj,
+                features=model_data["features"],
+                target=model_data["target"],
+                accuracy=model_data["accuracy"],
+                created_at=datetime.fromisoformat(model_data["created_at"]),
+                last_trained=datetime.fromisoformat(model_data["last_trained"]),
+                training_samples=model_data["training_samples"],
+                metadata=model_data["metadata"]
+            )
+            
+            self.add_model(model)
+            return True
+            
+        except Exception as e:
+            print(f"Error loading model: {e}")
+            return False
+
+
+class AnomalyDetector:
+    """Anomaly detection system"""
+    
+    def __init__(self):
+        self.isolation_forest = IsolationForest(contamination=0.1, random_state=42)
+        self.is_fitted = False
+        self.feature_extractor = FeatureExtractor()
+    
+    def fit(self, normal_data: List[Dict[str, Any]]) -> None:
+        """Fit the anomaly detector with normal data"""
+        features = []
+        for data in normal_data:
+            features.append(list(self.feature_extractor.extract_request_features(data).values()))
+        
+        X = np.array(features)
+        self.isolation_forest.fit(X)
+        self.is_fitted = True
+    
+    def detect_anomaly(self, request_data: Dict[str, Any]) -> Dict[str, Any]:
+        """Detect if request is anomalous"""
+        if not self.is_fitted:
+            return {"is_anomaly": False, "score": 0.0, "message": "Model not fitted"}
+        
+        features = self.feature_extractor.extract_request_features(request_data)
+        feature_values = list(features.values())
+        X = np.array(feature_values).reshape(1, -1)
+        
+        anomaly_score = self.isolation_forest.decision_function(X)[0]
+        is_anomaly = self.isolation_forest.predict(X)[0] == -1
+        
+        return {
+            "is_anomaly": bool(is_anomaly),
+            "score": float(anomaly_score),
+            "confidence": abs(anomaly_score),
+            "features": features
+        }
+
+
+class SmartCache:
+    """ML-powered smart caching system"""
+    
+    def __init__(self, max_size: int = 1000):
+        self.cache: Dict[str, Any] = {}
+        self.access_times: Dict[str, datetime] = {}
+        self.access_counts: Dict[str, int] = defaultdict(int)
+        self.max_size = max_size
+        self.ml_manager = MLModelManager()
+        self._setup_cache_model()
+    
+    def _setup_cache_model(self) -> None:
+        """Setup ML model for cache prediction"""
+        # Create a simple model to predict cache hit probability
+        model = MLModel(
+            name="cache_predictor",
+            model_type=MLModelType.CLASSIFICATION,
+            model=RandomForestClassifier(n_estimators=100, random_state=42),
+            features=['path_length', 'has_query_params', 'hour_of_day', 'day_of_week'],
+            target="cache_hit"
+        )
+        self.ml_manager.add_model(model)
+    
+    def get(self, key: str) -> Optional[Any]:
+        """Get value from cache"""
+        if key in self.cache:
+            self.access_times[key] = datetime.now()
+            self.access_counts[key] += 1
+            return self.cache[key]
+        return None
+    
+    def set(self, key: str, value: Any, ttl: int = None) -> None:
+        """Set value in cache"""
+        if len(self.cache) >= self.max_size:
+            self._evict_least_used()
+        
+        self.cache[key] = value
+        self.access_times[key] = datetime.now()
+        self.access_counts[key] = 0
+    
+    def predict_cache_hit(self, request_data: Dict[str, Any]) -> float:
+        """Predict cache hit probability for a request"""
+        features = self.ml_manager.feature_extractor.extract_request_features(request_data)
+        
+        # Use only features available in the model
+        model_features = {k: features.get(k, 0) for k in self.ml_manager.get_model("cache_predictor").features}
+        
+        request = PredictionRequest(
+            features=model_features,
+            model_name="cache_predictor",
+            prediction_type=PredictionType.CACHE_HIT
+        )
+        
+        result = self.ml_manager.predict(request)
+        return result.confidence if result.prediction else 0.0
+    
+    def _evict_least_used(self) -> None:
+        """Evict least used items from cache"""
+        if not self.cache:
+            return
+        
+        # Find item with lowest access count and oldest access time
+        least_used_key = min(
+            self.cache.keys(),
+            key=lambda k: (self.access_counts[k], self.access_times[k])
+        )
+        
+        del self.cache[least_used_key]
+        del self.access_times[least_used_key]
+        del self.access_counts[least_used_key]
+
+
+class MLIntegration:
+    """Main ML integration system"""
+    
+    def __init__(self):
+        self.model_manager = MLModelManager()
+        self.anomaly_detector = AnomalyDetector()
+        self.smart_cache = SmartCache()
+        self.request_history: deque = deque(maxlen=10000)
+        self.response_patterns: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
+    
+    def record_request(self, request_data: Dict[str, Any], response_data: Dict[str, Any]) -> None:
+        """Record request-response pair for ML training"""
+        self.request_history.append({
+            "request": request_data,
+            "response": response_data,
+            "timestamp": datetime.now()
+        })
+        
+        # Update response patterns
+        path = request_data.get('path', '')
+        self.response_patterns[path].append(response_data)
+    
+    def create_response_time_model(self) -> MLModel:
+        """Create a model to predict response times"""
+        model = MLModel(
+            name="response_time_predictor",
+            model_type=MLModelType.REGRESSION,
+            model=LinearRegression(),
+            features=['method_encoded', 'path_length', 'body_length', 'header_count', 'hour_of_day'],
+            target="response_time"
+        )
+        self.model_manager.add_model(model)
+        return model
+    
+    def create_error_probability_model(self) -> MLModel:
+        """Create a model to predict error probability"""
+        model = MLModel(
+            name="error_probability_predictor",
+            model_type=MLModelType.CLASSIFICATION,
+            model=MLPClassifier(hidden_layer_sizes=(100, 50), random_state=42),
+            features=['method_encoded', 'path_length', 'body_length', 'hour_of_day', 'day_of_week'],
+            target="error_probability"
+        )
+        self.model_manager.add_model(model)
+        return model
+    
+    def train_models(self) -> Dict[str, Any]:
+        """Train all models with collected data"""
+        results = {}
+        
+        # Prepare training data
+        if len(self.request_history) < 100:
+            return {"error": "Insufficient training data"}
+        
+        # Response time model
+        response_time_model = self.create_response_time_model()
+        response_time_data = self._prepare_response_time_data()
+        if response_time_data:
+            results["response_time"] = self.model_manager.train_model(
+                "response_time_predictor", response_time_data
+            )
+        
+        # Error probability model
+        error_model = self.create_error_probability_model()
+        error_data = self._prepare_error_data()
+        if error_data:
+            results["error_probability"] = self.model_manager.train_model(
+                "error_probability_predictor", error_data
+            )
+        
+        # Anomaly detector
+        normal_data = [item["request"] for item in self.request_history]
+        self.anomaly_detector.fit(normal_data)
+        results["anomaly_detector"] = {"fitted": True}
+        
+        return results
+    
+    def _prepare_response_time_data(self) -> Optional[TrainingData]:
+        """Prepare training data for response time prediction"""
+        features = []
+        targets = []
+        
+        for item in self.request_history:
+            request = item["request"]
+            response = item["response"]
+            
+            # Extract features
+            request_features = self.model_manager.feature_extractor.extract_request_features(request)
+            response_features = self.model_manager.feature_extractor.extract_response_features(response)
+            
+            # Combine features
+            combined_features = {**request_features, **response_features}
+            features.append(combined_features)
+            
+            # Target is response time (simulated)
+            targets.append(np.random.uniform(0.1, 2.0))  # Simulated response time
+        
+        if not features:
+            return None
+        
+        return TrainingData(
+            features=features,
+            targets=targets,
+            feature_names=list(features[0].keys()),
+            target_name="response_time"
+        )
+    
+    def _prepare_error_data(self) -> Optional[TrainingData]:
+        """Prepare training data for error probability prediction"""
+        features = []
+        targets = []
+        
+        for item in self.request_history:
+            request = item["request"]
+            response = item["response"]
+            
+            # Extract features
+            request_features = self.model_manager.feature_extractor.extract_request_features(request)
+            features.append(request_features)
+            
+            # Target is error probability (1 if error, 0 if success)
+            is_error = response.get("status_code", 200) >= 400
+            targets.append(1 if is_error else 0)
+        
+        if not features:
+            return None
+        
+        return TrainingData(
+            features=features,
+            targets=targets,
+            feature_names=list(features[0].keys()),
+            target_name="error_probability"
+        )
+    
+    def predict_response_time(self, request_data: Dict[str, Any]) -> float:
+        """Predict response time for a request"""
+        request = PredictionRequest(
+            features=request_data,
+            model_name="response_time_predictor",
+            prediction_type=PredictionType.RESPONSE_TIME
+        )
+        
+        result = self.model_manager.predict(request)
+        return result.prediction if result.prediction else 1.0
+    
+    def predict_error_probability(self, request_data: Dict[str, Any]) -> float:
+        """Predict error probability for a request"""
+        request = PredictionRequest(
+            features=request_data,
+            model_name="error_probability_predictor",
+            prediction_type=PredictionType.ERROR_PROBABILITY
+        )
+        
+        result = self.model_manager.predict(request)
+        return result.confidence if result.prediction else 0.0
+    
+    def detect_anomaly(self, request_data: Dict[str, Any]) -> Dict[str, Any]:
+        """Detect anomalies in request"""
+        return self.anomaly_detector.detect_anomaly(request_data)
+    
+    def get_cache_recommendation(self, request_data: Dict[str, Any]) -> Dict[str, Any]:
+        """Get cache recommendation for a request"""
+        cache_hit_probability = self.smart_cache.predict_cache_hit(request_data)
+        
+        return {
+            "should_cache": cache_hit_probability > 0.7,
+            "cache_hit_probability": cache_hit_probability,
+            "recommended_ttl": int(3600 * cache_hit_probability)  # TTL in seconds
+        }
+    
+    def generate_smart_response(self, request_data: Dict[str, Any]) -> Dict[str, Any]:
+        """Generate a smart response using ML predictions"""
+        # Predict response characteristics
+        response_time = self.predict_response_time(request_data)
+        error_probability = self.predict_error_probability(request_data)
+        anomaly_result = self.detect_anomaly(request_data)
+        cache_recommendation = self.get_cache_recommendation(request_data)
+        
+        # Generate response based on predictions
+        if error_probability > 0.5:
+            status_code = np.random.choice([400, 401, 403, 404, 500], p=[0.3, 0.2, 0.1, 0.3, 0.1])
+        else:
+            status_code = 200
+        
+        response = {
+            "status_code": status_code,
+            "headers": {
+                "Content-Type": "application/json",
+                "X-ML-Predicted": "true",
+                "X-Response-Time": str(response_time),
+                "X-Error-Probability": str(error_probability)
+            },
+            "body": {
+                "message": "ML-generated response",
+                "predicted_response_time": response_time,
+                "error_probability": error_probability,
+                "is_anomaly": anomaly_result["is_anomaly"],
+                "cache_recommendation": cache_recommendation
+            }
+        }
+        
+        # Add cache headers if recommended
+        if cache_recommendation["should_cache"]:
+            response["headers"]["Cache-Control"] = f"max-age={cache_recommendation['recommended_ttl']}"
+        
+        return response
+
+
+# Global ML integration instance
+ml_integration = MLIntegration()
+
+
+# Convenience functions
+def create_ml_model(name: str, model_type: MLModelType, features: List[str], target: str) -> MLModel:
+    """Create a new ML model"""
+    if model_type == MLModelType.CLASSIFICATION:
+        model = RandomForestClassifier(n_estimators=100, random_state=42)
+    elif model_type == MLModelType.REGRESSION:
+        model = LinearRegression()
+    else:
+        model = RandomForestClassifier(n_estimators=100, random_state=42)
+    
+    ml_model = MLModel(
+        name=name,
+        model_type=model_type,
+        model=model,
+        features=features,
+        target=target
+    )
+    
+    ml_integration.model_manager.add_model(ml_model)
+    return ml_model
+
+
+def train_ml_models() -> Dict[str, Any]:
+    """Train all ML models"""
+    return ml_integration.train_models()
+
+
+def predict_response_characteristics(request_data: Dict[str, Any]) -> Dict[str, Any]:
+    """Predict response characteristics for a request"""
+    return {
+        "response_time": ml_integration.predict_response_time(request_data),
+        "error_probability": ml_integration.predict_error_probability(request_data),
+        "anomaly_detection": ml_integration.detect_anomaly(request_data),
+        "cache_recommendation": ml_integration.get_cache_recommendation(request_data)
+    }