API Reference¶

This section provides comprehensive documentation for all ONAD classes, functions, and modules. The API is organized into logical groups based on functionality.

Package Structure¶

ONAD follows a modular architecture with clear separation of concerns:

onad/
├── base/           # Abstract base classes and interfaces
├── model/          # Anomaly detection algorithms
│   ├── unsupervised/   # Unsupervised learning models
│   ├── supervised/     # Supervised learning models
│   └── statistics/     # Statistical methods
├── transform/      # Data preprocessing and transformation
├── stream/         # Data streaming and I/O utilities
└── utils/          # Helper functions and utilities

Core Concepts¶

Base Classes¶

All ONAD components inherit from well-defined base classes that establish consistent interfaces:

BaseModel: Foundation for all anomaly detection models
BaseTransformer: Base for data preprocessing components
BaseSimilaritySearchEngine: Interface for similarity search implementations

Model Categories¶

Unsupervised Models¶

Models that detect anomalies without labeled training data:

Forest-based: Tree ensemble methods for anomaly detection
- OnlineIsolationForest
- MondrianIsolationForest
SVM-based: Support Vector Machine approaches
- IncrementalOneClassSVMAdaptiveKernel
- GADGETSVM
Distance-based: Proximity-based detection methods
- IncrementalKNN

Statistical Models¶

Classical statistical approaches for anomaly detection:

Univariate: Single-variable statistical methods
Multivariate: Multi-variable statistical methods

Data Processing¶

Transformers¶

Data preprocessing and feature engineering components:

StandardScaler: Z-score normalization
MinMaxScaler: Min-max scaling
IncrementalPCA: Online Principal Component Analysis

Streaming¶

Efficient data streaming and I/O utilities:

ParquetStreamer: Stream data from Parquet files
Dataset: Built-in datasets for testing

Utilities¶

Helper functions and supporting components:

Similarity Search: Fast similarity search engines
Architecture Utils: Deep learning utilities

Usage Patterns¶

Basic Model Usage¶

All models follow a consistent interface:

from onad.model.iforest import OnlineIsolationForest

# Initialize model
model = OnlineIsolationForest(window_size=1000)

# Process streaming data
for data_point in data_stream:
    # Learn from the data point
    model.learn_one(data_point)

    # Get anomaly score
    score = model.score_one(data_point)

    # Check if anomalous
    if score > threshold:
        handle_anomaly(data_point, score)

Pipeline Construction¶

Combine transformers and models:

from onad.transform.preprocessing.scaler import StandardScaler
from onad.model.svm import IncrementalOneClassSVMAdaptiveKernel

# Create pipeline components
scaler = StandardScaler()
detector = IncrementalOneClassSVMAdaptiveKernel()


# Process data through pipeline
def detect_anomaly(raw_data):
    # Transform data
    scaler.learn_one(raw_data)
    normalized_data = scaler.transform_one(raw_data)

    # Detect anomaly
    detector.learn_one(normalized_data)
    return detector.score_one(normalized_data)

Memory Management¶

ONAD models are designed for streaming scenarios with memory constraints:

# Configure memory limits
model = OnlineIsolationForest(
    window_size=1000,      # Limit memory window
    max_leaf_samples=32    # Control tree complexity
)

# Monitor memory usage
info = model.get_model_info() if hasattr(model, 'get_model_info') else {}
print(f"Model state: {info}")

Type Safety¶

All ONAD components include comprehensive type hints:

from typing import Dict
from onad.model.iforest import OnlineIsolationForest

model: OnlineIsolationForest = OnlineIsolationForest()
features: Dict[str, float] = {"temperature": 23.5, "humidity": 0.65}
score: float = model.score_one(features)

Error Handling¶

ONAD provides robust error handling with informative messages:

try:
    model.learn_one(invalid_data)
except ValueError as e:
    print(f"Data validation error: {e}")
except RuntimeError as e:
    print(f"Model state error: {e}")

Quick Reference¶

Most Common Classes¶

Class	Purpose	Location
`OnlineIsolationForest`	Tree-based anomaly detection	`onad.model.unsupervised.forest`
`IncrementalOneClassSVMAdaptiveKernel`	SVM-based detection	`onad.model.unsupervised.svm`
`StandardScaler`	Data normalization	`onad.transform.scale`
`ParquetStreamer`	Data streaming	`onad.stream.streamer`

Key Methods¶

Method	Purpose	Available On
`learn_one(x)`	Learn from one data point	All models and transformers
`score_one(x)`	Get anomaly score	All models
`transform_one(x)`	Transform one data point	All transformers
`get_model_info()`	Get model statistics	Selected models

API Stability

ONAD follows semantic versioning. The public API documented here is considered stable within major versions. Internal APIs (prefixed with _) may change without notice.