RSSLoss — Relevance Squared Sum Loss¶
Relevance-based objective measuring how much of the centered residual is relevant within the gnostic concept space.
Overview¶
Uses bounded surrogate mean(hi²) where hi = tanh(2θ) with θ = (y_pred - y_true) / S. Behaves like a conventional regression objective (RSS) while remaining aligned with MAGNET's centered-coordinate framework. The centered coordinate is θ = (y_pred - y_true) / S.
Value range: [0, ∞) (bounded by tanh saturation; zero at perfect alignment).
Key Features¶
- Bounded RSS surrogate (
mean(tanh²(2θ))) - Custom autograd kernel (
RSSLossFunction) - Compatible with standard regression workflows
Parameters¶
| Parameter | Type | Default | Description |
|---|---|---|---|
S |
float | str |
1 |
Scale parameter; "auto" for batch-local scale |
name |
str | None |
None |
Optional display name |
verbose |
bool |
False |
Verbose logging |
Attributes¶
S: Scale parameterS_local: Resolved local scale
Methods¶
forward(y_pred, y_true)¶
Return batch residual-squared surrogate (mean(tanh²(2θ))).
Example Usage¶
from machinegnostics.magnet import RSSLoss, Tensor
import numpy as np
y_pred = Tensor(np.zeros((2, 1)), requires_grad=True)
y_true = Tensor(np.zeros((2, 1)))
print(float(RSSLoss()(y_pred, y_true))) # 0.0
Author: Nirmal Parmar