Skip to content
LQM Registry / ensemble

Adaptive Exponential-Weights Ensemble

Recency-aware exponential weighting over candidate loss histories with deterministic champion selection and effective-model-count diagnostics.

Model keylq.ensemble.exponential-weights
Version1.0.0
Runtimepython-process
Model contract

Inputs are explicit.

{
    "type": "object",
    "required": [
        "candidates"
    ],
    "properties": {
        "candidates": {
            "type": "object"
        }
    }
}
Output contract

Outputs are explicit.

{
    "type": "object",
    "required": [
        "weights",
        "champion",
        "discounted_losses",
        "effective_model_count"
    ]
}
Default parameters
{
    "eta": 1,
    "decay": 0.9699999999999999733546474089962430298328399658203125,
    "weight_floor": 0
}
Example input
{
    "candidates": {
        "ar1": [
            0.8000000000000000444089209850062616169452667236328125,
            0.6999999999999999555910790149937383830547332763671875,
            0.59999999999999997779553950749686919152736663818359375,
            0.5500000000000000444089209850062616169452667236328125,
            0.5
        ],
        "trend": [
            0.6999999999999999555910790149937383830547332763671875,
            0.61999999999999999555910790149937383830547332763671875,
            0.57999999999999996003197111349436454474925994873046875,
            0.5,
            0.419999999999999984456877655247808434069156646728515625
        ],
        "naive": [
            0.90000000000000002220446049250313080847263336181640625,
            0.88000000000000000444089209850062616169452667236328125,
            0.84999999999999997779553950749686919152736663818359375,
            0.82999999999999996003197111349436454474925994873046875,
            0.8000000000000000444089209850062616169452667236328125
        ]
    }
}
Capabilities
cpuensembleadaptivemodel-selection