stan_data, t0 = ncs.get_stan_data(csv, args)
if stan_data is None:
print("No data for %s; skipping fit." % args.roi)
sys.exit(0)
if args.n_data_only:
print(ncs.get_n_data(stan_data))
sys.exit(0)
init_fun = ncs.get_init_fun(args, stan_data)
model_path = Path(args.models_path) / ('%s.stan' % args.model_name)
model_path = model_path.resolve()
assert model_path.is_file(), "No such .stan file: %s" % model_path
control = {'adapt_delta': args.adapt_delta}
stanrunmodel = ncs.load_or_compile_stan_model(
args.model_name, args.models_path, force_recompile=args.force_recompile)
# Fit Stan
# fit = stanrunmodel.sampling(data=stan_data, init=init_fun, control=control,
# chains=args.n_chains,
# chain_id=np.arange(args.n_chains),
# warmup=args.n_warmups, iter=args.n_iter,
# thin=args.n_thin)
fit = stanrunmodel.vb(data=stan_data, init=init_fun)
# Uncomment to print fit summary
print(fit)
# Save fit
save_dir = Path(args.fits_path)
save_dir.mkdir(parents=True, exist_ok=True)
if not args.model_names:
args.model_names = ncs.list_models(args.models_path)
assert len(args.model_names),\
("No such model files matching: *.stan' at %s" % (args.models_path))
model_paths = [
Path(args.models_path) / ('%s.stan' % model_name)
for model_name in args.model_names
]
for model_path in model_paths:
assert model_path.is_file(), "No such .stan file: %s" % model_path
run_script_path = Path(__file__).parent / 'run.py'
run_flags = [('--%s' % key.replace('_', '-'), value)
for key, value in args.__dict__.items()
if key not in ['model_names', 'rois']]
# This next section will be run in serial since this is only a reference
# implementation. Parallel implementations are possible with multiprocessing
# or the library of your choice. Best performance comes from parallelizing
# run.py on a cluster.
iterator = tqdm(args.model_names, desc='Model')
for model_name in iterator:
iterator.set_description("Compiling %s" % model_name)
ncs.load_or_compile_stan_model(model_name,
models_path=args.models_path,
force_recompile=True)
print("Finished compiling all models")