RollingMedian / Computation Layer

Calculate the rolling (moving) median over a sliding window. Similar to pandas.rolling().median() or R's rollmedian(). More robust to outliers than mean.

Mathematical form: For each window, find value where: $P (X \leq x_{m e d ian}) = P (X \geq x_{m e d ian}) = 0.5$

Common applications:

Robust trend detection
Outlier-resistant smoothing
Image processing (median filtering)
Economic indicators (house prices)
Noise reduction in sensors
Clinical data analysis
Financial risk assessment

Note: More computationally intensive than mean but better handles outliers.

Table

Select

column

The numeric column to compute rolling medians for. Ideal for:

Price data with outliers
Sensor data with noise
Economic indicators
Population statistics
Quality measurements

Especially useful when mean is skewed by extreme values.

WindowSize

u32

Number of observations in each sliding window. Common choices:

5: Weekly median (business days)
7: Weekly median (calendar days)
30: Monthly median
90: Quarterly median Odd window sizes provide single middle value; even sizes average two middle values.

Weights

[f64, ...]

Optional weights for window values. Must match window size if provided. Note:

Weights affect value selection process
Changes interpretation of median
Less common than weighted means

Consider carefully whether weighted median meets analytical needs.

Require complete windows for median calculation. Ensures statistical validity by only computing when all values in window are available. Critical for accurate central tendency measurement in presence of outliers.

Period

Value

u32

Minimum number of valid observations (Value) required for median calculation. Examples:

1: Calculate with any valid data point
window_size/2: Require majority of window
window_size: Require complete window

More observations generally provide more reliable central tendency.

Center

bool

false

Window label position. When true, aligns to window center; when false, to end. Consider:

true: Better for historical analysis and pattern identification
false: Better for real-time monitoring and immediate decision making

AsColumn

name

Name for the new column. If not provided, the system generates a unique name. If AsColumn matches an existing column, the existing column is replaced. The name should follow valid column naming conventions.