Nixtla/statsforecast

Features

• Time Series Forecasting - Implements high-performance statistical forecasting for millions of individual time series using distributed computing.
• Statistical Forecasting - Provides a high-performance library for statistical point forecasting and prediction intervals using univariate models.
• C-Based Engines - Uses compiled C kernels to accelerate ARIMA and exponential smoothing computations for millions of series.
• Model Selectors - Identifies and fits the optimal statistical model for every unique series in a dataset.
• Time Series Anomaly Detection - Identifies outliers in time series data by comparing observed values against probabilistic forecast intervals.
• Automated Model Selection - Automatically identifies and fits the optimal statistical model for every unique series in a large dataset.
• Probabilistic Forecasting - Generates point forecasts and probabilistic prediction intervals to quantify uncertainty and risk for planning.
• Parallel Execution Engines - Enables high-performance fitting of statistical models to millions of individual time series simultaneously.
• Univariate Forecasting - Predicts values for single time series using complex exponential smoothing.
• Distributed Cluster Execution - Scales forecasting workloads by partitioning millions of time series across server clusters.
• Distributed Computing Frameworks - Scales forecasting workloads across server clusters using distributed computing and parallel execution.
• Exponential Smoothing Models - Implements a comprehensive suite of exponential smoothing methods for various time series types.
• Parallel Task Execution - Distributes model fitting and forecasting tasks across multiple processor cores to handle millions of series.
• Multi-Core Parallelization - Distributes model fitting and prediction tasks across available processor cores to process multiple series simultaneously.
• Gradient Boosted Tree Forecasting - Implements time series forecasting using gradient boosting decision trees.
• Model Performance Selection - Provides automated selection of the best performing statistical forecasting models for every unique time series in a dataset.
• Forecasting - Tunes forecasting parameters using time series cross validation and grid search to minimize prediction error.
• Forecast Evaluation - Implements evaluation of time series prediction accuracy using historical cross-validation and error metrics.
• Intermittent Demand Forecasting - Predicts values for irregular series with long periods of zero using specialized intermittent demand models.
• Linear Trend Forecasting - Provides linear trend forecasting using separate smoothing parameters for the level and the trend.
• Multivariate Forecasting - Provides multivariate forecasting by combining past observations with external exogenous variables.
• Non-Stationary Forecasting - Predicts values for series with evolving trends by optimizing parameters and updating model components.
• Exogenous Variable Integration - Integrates external regressors and static covariates into statistical models to refine point forecasts.
• Cross Validation Evaluation - Performs model evaluation using sliding-window historical data to measure forecasting accuracy.
• Multiple Seasonality Modeling - Decomposes time series into several seasonal patterns and a trend using local polynomial regression.
• Seasonal ARIMA Models - Implements seasonal ARIMA to handle time series with recurring patterns by adding seasonal components.
• Parallel Dataframe Operations - Automatically distributes tabular data computations across all available CPU cores for high-speed processing.
• Dataframe Structures - Employs high-performance dataframe structures to organize multiple time series for efficient memory management.
• Lazy Query Pipelines - Implements data processing pipelines using lazy evaluation to reduce memory overhead for massive datasets.
• Parameter Optimization - Uses statistical optimization to identify the optimal autoregression, integration, and moving average parameters for ARIMA models.
• Time Series Decomposition - Decomposes time series into multiple lines using theta coefficients to isolate specific behaviors.
• Forecasting Validations - Provides accuracy evaluation by sliding a window across historical data to predict subsequent periods.
• Lazy Evaluation - Utilizes lazy evaluation to defer data manipulation tasks and optimize memory usage.
• Statistical Kernels - Provides high-performance compiled implementations of statistical algorithms like ARIMA and smoothing.
• Simple Exponential Smoothing - Implements simple exponential smoothing for time series without clear trends.
• Triple Exponential Smoothing - Implements triple exponential smoothing to estimate and forecast seasonal time series components.
• GARCH Volatility Forecasting - Provides GARCH and ARCH models to predict conditional variance and volatility clusters in financial return data.
• Volatility Clustering Prediction - Predicts series where large changes cluster together, typical of financial return data.
• Residual-Based Interval Estimation - Computes distribution-free prediction intervals by analyzing cross-validation residuals.
• Long-Format Data Layouts - Organizes multiple time series into a single dataframe to optimize memory locality and batch processing throughput.
• Custom C++ Kernels - Uses compiled C source code to accelerate core ARIMA and exponential smoothing operations.
• State-Space Models - Implements exponential smoothing by representing time series as internal states updated through innovations.
• Time Series - Fast forecasting using statistical and econometric models.
• Time Series Analysis - High-performance statistical and econometric forecasting models.
• Time Series Analysis - Statistical and econometric time series forecasting models.
• Analysis Toolkits - Efficient statistical models for rapid time series forecasting.

Star history