// Licensed under the Apache License, Version 2.0
// © mihakralj
//@version=6
indicator("Weighted Root Mean Squared Error", "WRMSE", overlay=false)

//@function Calculates Weighted Root Mean Squared Error
//@param   actual Series of actual values
//@param   predicted Series of predicted/forecast values
//@param   weight Series of weights for each observation
//@param   length Rolling window for calculation
//@returns WRMSE value in same units as original data
wrmse(series float actual, series float predicted, series float weight, simple int length) =>
    float epsilon = 1e-10

    // Compute weighted squared error for current bar
    float diff = nz(actual, 0.0) - nz(predicted, 0.0)
    float w = math.max(nz(weight, 1.0), 0.0)  // Ensure non-negative weight
    float weightedSqError = w * diff * diff

    // Rolling sums
    float sumWeightedError = ta.sum(weightedSqError, length)
    float sumWeight = ta.sum(w, length)

    // WRMSE = √(Σ(w*e²) / Σ(w))
    float result = sumWeight > epsilon ? math.sqrt(sumWeightedError / sumWeight) : 0.0
    result

//@function Calculates WRMSE with uniform weights (equivalent to RMSE)
//@param   actual Series of actual values
//@param   predicted Series of predicted/forecast values
//@param   length Rolling window for calculation
//@returns WRMSE value (same as RMSE when weights are uniform)
wrmse_uniform(series float actual, series float predicted, simple int length) =>
    wrmse(actual, predicted, 1.0, length)

// ---------- Main loop ----------

// Inputs
i_length = input.int(14, "Length", minval=1)
i_use_volume_weights = input.bool(false, "Use Volume as Weights", tooltip="When enabled, errors are weighted by volume")
i_actual = input.source(close, "Actual")
i_predicted = input.source(open, "Predicted")

// Calculation
weight = i_use_volume_weights ? volume : 1.0
wrmse_value = wrmse(i_actual, i_predicted, weight, i_length)

// Plot
plot(wrmse_value, "WRMSE", color=color.yellow, linewidth=2)
hline(0, "Perfect", color=color.green, linestyle=hline.style_dotted)