Installation instructions from fork of CmdStan at syclik/cmdstan:
git clone https://github.com/syclik/cmdstan.git cmdstan-colors
cd cmdstan-colors
git checkout v2.29.1-colors
git submodule update --init --recursive --depth=1
make build -j4 # or num of cores
Build a CmdStan executable and run as usual. Hereβs how to do it with the included bernoulli example:
make examples/bernoulli/bernoulli
./examples/bernoulli/bernoulli sample data file=examples/bernoulli/bernoulli.data.json
From start to finish, it took ~8 hours of elapsed time; under 2 hours of actual coding time. Most of that time was figuring out how to print utf-32 emojis in C++.
More generally, what would a Stan-native version of this kind of feature look like?
An auto-generated, tweetable, colourful summary that captures information about some characteristics of your model, how long the model took to run, how well it sampled, if it was truly compiled with AVX optimisations enabled?
In time, Stan user support veterans could learn to diagnose common user or modelling problems from the summary.
another interpretation could be 1 block per parameter. that might get a bit wild and challenge tweetability for larger models.