Implementation:Online ml River Optim Initializers

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

Overview

A collection of weight initialization strategies for online machine learning models including constant, zeros, and normal distribution initializers.

Description

The initializers module provides strategies for initializing model weights before training begins. Proper initialization is crucial for effective training and convergence. The module includes three main initializers: Constant (returns a fixed value for all weights), Zeros (specialized constant initializer that returns zero), and Normal (samples from a normal distribution with specified mean and standard deviation). All initializers support both scalar initialization (shape=1) and vector initialization (shape>1), returning either a single float or a numpy array depending on the requested shape. The Normal initializer supports setting a random seed for reproducibility.

Usage

Import from river.optim.initializers to initialize weights in models or custom implementations. Commonly used in neural networks and linear models.

Code Reference

Source Location

• Repository: Online_ml_River
• File: river/optim/initializers.py

Signature

class Constant(Initializer):
    def __init__(self, value: float):
        ...
    def __call__(self, shape=1):
        ...

class Zeros(Constant):
    def __init__(self):
        ...

class Normal(Initializer):
    def __init__(self, mu=0.0, sigma=1.0, seed: int | None = None):
        ...
    def __call__(self, shape=1):
        ...

Import

from river import optim

I/O Contract

Inputs

Name	Type	Required	Description
value	float	Yes (Constant)	Constant value to initialize all weights
mu	float	No (default=0.0)	Mean of normal distribution
sigma	float	No (default=1.0)	Standard deviation of normal distribution
seed	int or None	No (default=None)	Random seed for reproducibility
shape	int	No (default=1)	Number of weights to initialize (passed to __call__)

Outputs

Name	Type	Description
weights	float or ndarray	Single value if shape=1, otherwise numpy array of specified shape

Usage Examples

from river import optim

# Constant initializer
init = optim.initializers.Constant(value=3.14)
print(init(shape=1))  # 3.14
print(init(shape=3))  # array([3.14, 3.14, 3.14])

# Zeros initializer (special case of Constant)
init = optim.initializers.Zeros()
print(init(shape=1))  # 0.0
print(init(shape=4))  # array([0., 0., 0., 0.])

# Normal distribution initializer
init = optim.initializers.Normal(mu=0, sigma=1, seed=42)
print(init(shape=1))  # Single random value
print(init(shape=5))  # Array of 5 random values

# Xavier/He-style initialization
init = optim.initializers.Normal(mu=0, sigma=0.01, seed=123)
weights = init(shape=100)

# Use in custom models
class SimpleLinear:
    def __init__(self, n_features):
        init = optim.initializers.Normal(mu=0, sigma=0.1, seed=42)
        self.weights = init(shape=n_features)

# Common initialization strategies
zero_init = optim.initializers.Zeros()
small_random = optim.initializers.Normal(mu=0, sigma=0.01)
large_constant = optim.initializers.Constant(value=1.0)

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment