Tutorial: Building with CleaveDB
Forecasting (FORECAST) — Overview
In conventional data stacks, computing statistical predictions and time-series projections requires exporting raw tables into external Python scripts or analytical warehouses. CleaveDB integrates Feature 29: Statistical Prediction (FORECAST) directly into its query engine, enabling in-database machine learning and trend extrapolation with zero data movement.
Supported Prediction Methods
1. METHOD LINEAR
Computes Ordinary Least Squares (OLS) linear regression via vector math. Fits a best-fit trendline, calculates the R² coefficient of determination, and extrapolates values across future periods.
2. METHOD MOVING_AVERAGE
Smooths short-term fluctuations and noisy time series by computing rolling mean projections across a customizable window frame (WINDOW <n>, default 3).
3. METHOD EXPONENTIAL
Applies single exponential smoothing with a decay coefficient (α = 0.5), placing greater weight on recent observations while dampening historical variance.
4. Flexible Time Horizons
Project forward across granular units including DAYS, HOURS, MINUTES, and SECONDS, with built-in security caps up to 10,000 steps.
General Syntax
Every forecast statement follows a structured declarative pattern:
Explore Forecasting Guides
Select a guide below to master time-series extrapolation in CleaveDB: