Implementation:Online ml River TimeSeries Forecaster Base

Knowledge Sources	Online_ml_River
Domains	Online_Learning, Time_Series, Forecasting
Last Updated	2026-02-08 16:00 GMT

Overview

Abstract base class defining the interface for time series forecasting models in River.

Description

Defines the Forecaster interface with learn_one and forecast methods. learn_one accepts an endogenous variable (y) and optional exogenous variables (x). forecast generates predictions for a specified horizon, optionally using future exogenous variables. All River forecasters inherit from this base class.

Usage

Use as the parent class when implementing custom time series forecasters. Ensures compatibility with River's ecosystem and provides a consistent interface for all forecasting models.

Code Reference

Source Location

Repository: Online_ml_River
File: river/time_series/base.py

Signature

class Forecaster(base.Estimator):
    @property
    def _supervised(self):
        return True

    @abc.abstractmethod
    def learn_one(self, y: float, x: dict | None = None) -> None:
        """Updates the model.

        Parameters
        ----------
        y
            The endogenous variable
        x
            Optional exogenous variables
        """
        ...

    @abc.abstractmethod
    def forecast(self, horizon: int, xs: list[dict] | None = None) -> list:
        """Makes forecast at each step of the given horizon.

        Parameters
        ----------
        horizon
            Number of steps ahead to forecast
        xs
            Optional future exogenous variables
        """
        ...

Import

from river import time_series

I/O Contract

Method	Parameters	Returns	Description
learn_one	y: float, x: dict or None	None	Updates model with new observation
forecast	horizon: int, xs: list[dict] or None	list	Generates predictions for horizon steps

Usage Examples

from river import time_series
import abc

# Example: Implementing a simple forecaster
class NaiveForecaster(time_series.base.Forecaster):
    """Forecasts the last seen value"""

    def __init__(self):
        self.last_value = 0

    def learn_one(self, y, x=None):
        self.last_value = y

    def forecast(self, horizon, xs=None):
        return [self.last_value] * horizon

# Using the custom forecaster
model = NaiveForecaster()

# Learn from time series
for y in [10, 12, 11, 13, 15]:
    model.learn_one(y)

# Forecast next 3 steps
predictions = model.forecast(horizon=3)
print(f"Forecasts: {predictions}")  # [15, 15, 15]

# Using built-in forecasters
from river import time_series

# SNARIMAX example
snarimax = time_series.SNARIMAX(
    p=1,
    d=0,
    q=0,
    m=12,  # Monthly seasonality
    sp=1
)

# Simulate time series
import math
for t in range(50):
    y = 10 + 2 * math.sin(2 * math.pi * t / 12) + (t % 7) * 0.5
    snarimax.learn_one(y)

# Forecast
future = snarimax.forecast(horizon=5)
print(f"SNARIMAX forecast: {[f'{v:.2f}' for v in future]}")

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment