Analysis of CV methods under time-series dependence.
You will need an environment that can run python 3.10 and JAX. The script setup.sh might help with dependencies.
The script setup.py provides a shell command arx which is the entry point
for all experiments.
Once you've created a virtual environment and run setup.py, use arx --help
for usage information.
Usage: arx [OPTIONS] COMMAND [ARGS]...
Run CV experiments
This script is the entry point for all experiments in the paper. You'll
probably need to run setup.sh (or similar) to make it work.
Options:
--help Show this message and exit.
Commands:
by-alpha Simplified model selection by alpha.
by-dimension Simplified model selection by dimension and alpha.
by-excluded-effect Simplified model selection by excluded effect.
by-halo Simplified model selection by halo.
by-included-effect Simplified model selection by included effect.
by-length Simplified model selection by data length.
full-bayes Compute draws for full-Bayes experiment.
full-combine Combine full-Bayes result files.
full-kfold Compute draws for k-fold full-Bayes experiment.
generate-z Generate Z for use across experiments.
length-search Simplified model selection by data length.
loss Loss for simplified model selection by alpha.
pointwise-comparison Joint/pointwise loss comparison by alpha.
supplementary Joint/pointwise loss comparison by alpha.