Implementation:Online ml River Stats Shift

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

Overview

Shift returns past values from a data stream by shifting observations backward in time.

Description

This statistic maintains a buffer of recent values and returns a value from a specified number of steps in the past. It is primarily used as a building block for other statistics to avoid data leakage when computing statistics over target values. The shift amount determines how many steps back to retrieve, and a fill_value is returned when not enough observations have been seen yet.

Usage

Use Shift when you need to access past values in streaming computations, particularly to prevent leakage when calculating statistics on target variables. It is commonly composed with other statistics using the pipe operator (|) to create lagged versions of statistics, such as computing yesterday's average or last week's statistics without including today's data.

Code Reference

Source Location

• Repository: Online_ml_River
• File: river/stats/shift.py

Signature

class Shift(stats.base.Univariate):
    def __init__(self, amount=1, fill_value=None):
        self.amount = amount
        self.fill_value = fill_value
        self.buffer = collections.deque(maxlen=self.amount + 1)

Import

from river import stats

I/O Contract

Inputs

Name	Type	Required	Description
x	Any	Yes	Value to add to the buffer
amount	int	Yes (init)	Number of steps to shift back (default: 1)
fill_value	Any	No (init)	Value to return before buffer is filled (default: None)

Outputs

Name	Type	Description
get()	Any	Value from 'amount' steps ago (or fill_value if not enough data)

Usage Examples

from river import stats

# Basic shift by 1 step
shift = stats.Shift(1) | stats.Mean()

for i in range(5):
    shift.update(i)
    print(f"Current: {i}, Shifted Mean: {shift.get():.1f}")

# Output:
# Current: 0, Shifted Mean: 0.0
# Current: 1, Shifted Mean: 0.0
# Current: 2, Shifted Mean: 0.5
# Current: 3, Shifted Mean: 1.0
# Current: 4, Shifted Mean: 1.5

# Using with feature aggregation
from river import feature_extraction

# Average sales per shop, excluding today's sales
agg = feature_extraction.Agg(
    on='sales',
    how=stats.Shift(1) | stats.Mean(),
    by='shop'
)

X = [
    {'shop': 'Ikea', 'sales': 10},
    {'shop': 'Ikea', 'sales': 15},
    {'shop': 'Ikea', 'sales': 20}
]

# First observation
agg.learn_one(X[0])
print(agg.transform_one(X[0]))
# Output: {'sales_mean_of_shift_1_by_shop': 0.0}

# Second observation
agg.learn_one(X[1])
print(agg.transform_one(X[0]))
# Output: {'sales_mean_of_shift_1_by_shop': 10.0}

# Third observation
agg.learn_one(X[2])
print(agg.transform_one(X[0]))
# Output: {'sales_mean_of_shift_1_by_shop': 12.5}

# Shift by multiple steps
shift_3 = stats.Shift(amount=3, fill_value=-1)

for i in range(7):
    shift_3.update(i * 10)
    print(f"Current: {i*10}, 3 steps ago: {shift_3.get()}")

# Output:
# Current: 0, 3 steps ago: -1
# Current: 10, 3 steps ago: -1
# Current: 20, 3 steps ago: -1
# Current: 30, 3 steps ago: -1
# Current: 40, 3 steps ago: 0
# Current: 50, 3 steps ago: 10
# Current: 60, 3 steps ago: 20

# Computing yesterday's average (shift by 1 day)
daily_avg = stats.Shift(1) | stats.Mean()

for day in range(1, 8):
    value = day * 100
    daily_avg.update(value)
    print(f"Day {day}: Yesterday's avg = {daily_avg.get():.1f}")

Related Pages

• Environment:Online_ml_River_Python_Runtime_Environment