Skip to main content

← HomOps Reference

HomMinMax

Element-wise min or max core. Prefer HomMin / HomMax in graphs.

NmComparison
RunsServer
Changes shapeYes - broadcast
Changes scaleYes - via nested ops
Levels spentTunable - poly depth + mul
RotatesNo
KeysSquare key

What it does​

HomMinMax is the shared comparison core behind HomMin and HomMax. It is a composite: HomCompare plus arithmetic, not a registered leaf.
Prefer HomMin or HomMax in user-facing graphs. Use this class directly only when you need to select min vs max at runtime via is_min.
Current accuracy calibration requires inputs in [0, 1]. Comparison is element-wise and needs identical input shapes.

Signature​

HomMinMax(
    is_min,             # required: True = min, False = max
    x_accuracy=10,
    y_accuracy=10,
    left=0.0,
    right=1.0,
    tol=1e-8,
    rows_budget=None,
)
Two ciphertext inputs. No set_data.

Parameters​

ParameterTypeDefaultDescription
is_minboolrequiredTrue for min, False for max.
x_accuracyint10Horizontal accuracy for the min/max decision. Must be in 4..14.
y_accuracyfloat10Vertical accuracy for approximation error, forwarded to Compare / Sign.
leftfloat0.0Domain left; currently must be 0.0.
rightfloat1.0Domain right; currently must be 1.0.
tolfloat1e-8Tolerance forwarded to comparison internals.
rows_budgetSequence[int] | NoneNoneAbsolute rows allowed for internal mod-switch.
SettingRuleIf you break it
domainInputs must lie in [0, 1]. left must be 0.0 and right must be 1.0.Construction raises: current calibration requires inputs in [0, 1].

Example​

pipeline.pyPYTHON
from lattica_build.operators import HomMin, HomMax, HomMinMax

# Prefer HomMin / HomMax in graphs
h_min = HomMin(x_accuracy=7, y_accuracy=7)
h_max = HomMax(x_accuracy=7, y_accuracy=7)

# HomMinMax only when selecting min vs max at runtime
h_minmax = HomMinMax(is_min=True, x_accuracy=7, y_accuracy=7)

See also​