Skip to content

Runtime & Support Scaling

The case study in cases/runtime_support_scaling/ investigates computational efficiency and memory scaling as the Cartesian support of baseline covariates \(X\) expands.


The Combinatorial Challenge

Evaluating calibration constraints requires computing:

\[ \sum_{x \in \mathcal{S}_k} P(Y = 1 \mid X = x; \theta) P(X = x) \]

As the number of discrete variables \(p\) or the number of levels per variable increases, the total Cartesian support \(\prod_{j=1}^p \lvert \mathcal{X}_j \rvert\) grows exponentially.

The scaling harness systematically benchmarks: - Total wall-clock time per iteration. - Peak memory consumption. - Vectorized integration throughput. - Robust rejection thresholds when cardinality exceeds safety bounds.


Execution

Smoke Scaling Benchmark

python -m cases.runtime_support_scaling.run_scaling --profile smoke

Full Protocol Profile

python -m cases.runtime_support_scaling.run_scaling --profile protocol

Outputs are saved in data/runtime_support_scaling/runtime_protocol.csv and support_scaling.csv.