def test_stanc_exception(self):
model_code = 'parameters {real z;} model {z ~ no_such_distribution();}'
assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
with assertRaisesRegex(ValueError, 'unknown distribution'):
stanc(model_code=model_code)
with assertRaisesRegex(ValueError, 'unknown distribution'):
StanModel(model_code=model_code)
def test_stanc_exception(self):
model_code = 'parameters {real z;} model {z ~ no_such_distribution();}'
assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
with assertRaisesRegex(ValueError, 'unknown distribution'):
stanc(model_code=model_code)
with assertRaisesRegex(ValueError, 'unknown distribution'):
StanModel(model_code=model_code)
def test_stanc_exception(self):
model_code = 'parameters {real z;} model {z ~ no_such_distribution();}'
assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
# distribution not found error
with assertRaisesRegex(ValueError, r'Probability function must end in _lpdf or _lpmf\. Found'):
stanc(model_code=model_code)
with assertRaisesRegex(ValueError, r'Probability function must end in _lpdf or _lpmf\. Found'):
StanModel(model_code=model_code)
def test_stanc_exception(self):
model_code = 'parameters {real z;} model {z ~ no_such_distribution();}'
assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
# distribution not found error
with assertRaisesRegex(
ValueError,
r'Probability function must end in _lpdf or _lpmf\. Found'):
stanc(model_code=model_code)
with assertRaisesRegex(
ValueError,
r'Probability function must end in _lpdf or _lpmf\. Found'):
StanModel(model_code=model_code)
def test_stanc_exception_semicolon(self):
model_code = """
parameters {
real z
real y
}
model {
z ~ normal(0, 1);
y ~ normal(0, 1);}
"""
assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
with assertRaisesRegex(ValueError, 'Failed to parse'):
stanc(model_code=model_code)
with assertRaisesRegex(ValueError, 'Failed to parse'):
StanModel(model_code=model_code)
def test_stanc_exception_semicolon(self):
model_code = """
parameters {
real z
real y
}
model {
z ~ normal(0, 1);
y ~ normal(0, 1);}
"""
assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
with assertRaisesRegex(ValueError, 'Parser expecting: ";"'):
stanc(model_code=model_code)
with assertRaisesRegex(ValueError, 'Parser expecting: ";"'):
StanModel(model_code=model_code)
def test_stanc():
model_code = 'parameters {real y;} model {y ~ normal(0,1);}'
result = stanc(model_code=model_code)
print(result.keys())
assert result.keys() == {"status", "model_cppname", "cppcode",
"model_name", "model_code"}
assert result['cppcode'].startswith("// Code generated by Stan ")
def build_model(stan_folder, model_name, model_code, verbose_compile):
code_hash = hashlib.sha1(model_code.encode('utf-8')).hexdigest()
model_filename = os.path.join(stan_folder, model_name + "_" + code_hash)
if os.path.exists(model_filename):
print('STAN: cached model found; loading')
f = open(model_filename, 'rb')
sm = pk.load(f)
f.close()
else:
if not os.path.exists(stan_folder):
os.mkdir(stan_folder)
print('STAN: no cached model found; building')
try:
stanc_ret = pystan.stanc(model_code=model_code)
stanc_ret['cppcode'] = load_modified_cpp_code(
stan_folder, model_name, model_code)
sm = pystan.StanModel(stanc_ret=stanc_ret, verbose=verbose_compile)
except:
print(
'Compiling failed. Did you make sure to modify the weighted_[model_name]_[code_hash].cpp file to handle weighted data?'
)
raise
f = open(model_filename, 'wb')
pk.dump(sm, f)
f.close()
return sm
def test_stanc(self):
model_code = 'parameters {real y;} model {y ~ normal(0,1);}'
result = stanc(model_code=model_code)
assert sorted(result.keys()) == sorted(
{"status", "model_cppname", "cppcode", "model_name", "model_code"})
assert result['cppcode'].startswith("// Code generated by Stan ")
self.assertEqual(result['status'], 0)
def test_stanc_2_includes(self):
model_code = 'parameters {\n#include external1.stan\n#include external2.stan\n} model {\ny ~ normal(0,1);\nz ~ normal(0,1);}'
testdata_path = os.path.join(os.path.dirname(__file__), 'data', "")
result = stanc(model_code=model_code, include_paths=[testdata_path])
desired = sorted({"status", "model_cppname", "cppcode", "model_name", "model_code", "include_paths"})
self.assertEqual(sorted(result.keys()), desired)
self.assertEqual(result['status'], 0)
def test_stanc(self):
model_code = 'parameters {real y;} model {y ~ normal(0,1);}'
result = stanc(model_code=model_code)
desired = sorted({"status", "model_cppname", "cppcode", "model_name", "model_code"})
self.assertEqual(sorted(result.keys()), desired)
self.assertTrue(result['cppcode'].startswith("// Code generated by Stan "))
self.assertEqual(result['status'], 0)
def test_stanc_2_includes(self):
model_code = 'parameters {\n#include external1.stan\n#include external2.stan\n} model {\ny ~ normal(0,1);\nz ~ normal(0,1);}'
testdata_path = os.path.join(os.path.dirname(__file__), 'data', "")
result = stanc(model_code=model_code, include_paths=[testdata_path])
desired = sorted({
"status", "model_cppname", "cppcode", "model_name", "model_code",
"include_paths"
})
self.assertEqual(sorted(result.keys()), desired)
self.assertEqual(result['status'], 0)
def test_stanc(self):
model_code = 'parameters {real y;} model {y ~ normal(0,1);}'
result = stanc(model_code=model_code)
desired = sorted({
"status", "model_cppname", "cppcode", "model_name", "model_code",
"include_paths"
})
self.assertEqual(sorted(result.keys()), desired)
self.assertTrue(
result['cppcode'].startswith("// Code generated by Stan "))
self.assertEqual(result['status'], 0)
def get_model(model_name):
"""Get a StanModel object, checking first in local cache and recompiling
if necessary."""
cachefile = os.path.join(config.modelcache, '{}.pkl'.format(model_name))
# Check if package exists, get model code.
with _include_path() as incpth, _model_file(model_name) as modelfile:
model_stanc = pystan.stanc(file=modelfile,
include_paths=[incpth],
model_name=model_name)
# Check if cached file exists and load it
model = None
if os.path.isfile(cachefile):
try:
with open(cachefile, 'rb') as f:
model = pickle.load(f)
except pickle.UnpicklingError:
os.remove(cachefile) # Bad file, remove
# If cache file did not exist or is from an old model, recompile and cache
if (model is None) or (model.model_cppcode != model_stanc['cppcode']):
model = pystan.StanModel(stanc_ret=model_stanc)
os.makedirs(config.modelcache, exist_ok=True)
with open(cachefile, 'wb') as f:
pickle.dump(model, f, protocol=pickle.HIGHEST_PROTOCOL)
return model
def test_stan_model_from_file(self):
bernoulli_model_code = """
data {
int<lower=0> N;
int<lower=0,upper=1> y[N];
}
parameters {
real<lower=0,upper=1> theta;
}
model {
for (n in 1:N)
y[n] ~ bernoulli(theta);
}
"""
bernoulli_data = {'N': 10, 'y': [0, 1, 0, 0, 0, 0, 0, 0, 0, 1]}
temp_dir = tempfile.mkdtemp()
temp_fn = os.path.join(temp_dir, 'modelcode.stan')
with io.open(temp_fn, 'wt') as outfile:
outfile.write(bernoulli_model_code)
code = stanc(file=temp_fn)['model_code']
fit = stan(model_code=code, data=bernoulli_data)
extr = fit.extract(permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
# permuted=False
extr = fit.extract(permuted=False)
assert extr.shape == (1000, 4, 2)
assert 0.1 < np.mean(extr[:, 0, 0]) < 0.4
# permuted=True
extr = fit.extract('lp__', permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
extr = fit.extract('theta', permuted=True)
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
extr = fit.extract('theta', permuted=False)
assert extr['theta'].shape == (1000, 4)
assert 0.1 < np.mean(extr['theta'][:, 0]) < 0.4
fit = stan(file=temp_fn, data=bernoulli_data)
extr = fit.extract(permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
# permuted=False
extr = fit.extract(permuted=False)
assert extr.shape == (1000, 4, 2)
assert 0.1 < np.mean(extr[:, 0, 0]) < 0.4
# permuted=True
extr = fit.extract('lp__', permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
extr = fit.extract('theta', permuted=True)
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
extr = fit.extract('theta', permuted=False)
assert extr['theta'].shape == (1000, 4)
assert 0.1 < np.mean(extr['theta'][:, 0]) < 0.4
os.remove(temp_fn)
def test_utf8(self):
model_code = 'parameters {real y;} model {y ~ normal(0,1);}'
result = stanc(model_code=model_code)
self.assertEqual(sorted(result.keys()), self.desired)
self.assertTrue(result['cppcode'].startswith("// Code generated by Stan "))
self.assertEqual(result['status'], 0)
def test_utf8_inprogramcode(self):
model_code = u'parameters {real ö;\n} model {ö ~ normal(0,1);}'
assertRaisesRegex = self.assertRaisesRegexp if PY2 else self.assertRaisesRegex
with assertRaisesRegex(ValueError, 'Failed to parse Stan model .*'):
stanc(model_code=model_code)
def test_stanc_exception():
model_code = 'parameters {real z;} model {y ~ no_such_distribution();}'
result = stanc(model_code=model_code)
assert result['status'] == -1
def test_stan_model_from_file(self):
bernoulli_model_code = """
data {
int<lower=0> N;
int<lower=0,upper=1> y[N];
}
parameters {
real<lower=0,upper=1> theta;
}
model {
for (n in 1:N)
y[n] ~ bernoulli(theta);
}
"""
bernoulli_data = {'N': 10, 'y': [0, 1, 0, 0, 0, 0, 0, 0, 0, 1]}
temp_dir = tempfile.mkdtemp()
temp_fn = os.path.join(temp_dir, 'modelcode.stan')
with io.open(temp_fn, 'wt') as outfile:
outfile.write(bernoulli_model_code)
code = stanc(file=temp_fn)['model_code']
fit = stan(model_code=code, data=bernoulli_data)
extr = fit.extract(permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
# permuted=False
extr = fit.extract(permuted=False)
assert extr.shape == (1000, 4, 2)
assert 0.1 < np.mean(extr[:, 0, 0]) < 0.4
# permuted=True
extr = fit.extract('lp__', permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
extr = fit.extract('theta', permuted=True)
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
extr = fit.extract('theta', permuted=False)
assert extr.shape == (1000, 4, 2)
assert 0.1 < np.mean(extr[:, 0, 0]) < 0.4
fit = stan(file=temp_fn, data=bernoulli_data)
extr = fit.extract(permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
# permuted=False
extr = fit.extract(permuted=False)
assert extr.shape == (1000, 4, 2)
assert 0.1 < np.mean(extr[:, 0, 0]) < 0.4
# permuted=True
extr = fit.extract('lp__', permuted=True)
assert -7.9 < np.mean(extr['lp__']) < -7.0
extr = fit.extract('theta', permuted=True)
assert 0.1 < np.mean(extr['theta']) < 0.4
assert 0.01 < np.var(extr['theta']) < 0.02
extr = fit.extract('theta', permuted=False)
assert extr.shape == (1000, 4, 2)
assert 0.1 < np.mean(extr[:, 0, 0]) < 0.4
os.remove(temp_fn)
def test_stanc_exception():
model_code = 'parameters {real z;} model {y ~ no_such_distribution();}'
result = stanc(model_code=model_code)
assert result['status'] == -1
def stan(self, line, cell):
'''
Allow jupyter notebook cells to output stan code to file and compile it using
stanc compiler in your path or pystan.stanc if no compiler is found in path.
Output of compile is stored in "_stan_model object" or in an <object name>
specified in -v <object_name> (e.g %%stan -f test.stan -v test)
%%stan
- Saves the cell code to a string. The code string can be accessed via
_stan_model.model_code or <object_name>.model_code
(if you specify -v <object_name> option)
%%stan -f <stan_file_name>
- Saves the cell code to a file specified in <stan_file_name>. The file name
can also be accessed in _stan_model.model_file or in
<object_name>.model_file (if you specify -v <object_name> option)
%%stan -v <object_name> [default: _stan_model]
stan magic currently outputs a StanMagicOutput object in default _stan_model object.
-v allows you to specify an alternate compile output object name, so that
you can use specified object name instead of "_stan_model".
This is useful if you have multiple %%stan model cells. Cuurently the output object
exposes 3 attributes (model_name, model_code, model_file)
[_stan_model | <object_name>].model_file -> Name of stan_file
[_stan_model | <object_name>].model_name -> Name of stan model [None]
[_stan_model | <object_name>].model_code -> Model code
%%stan -f <stan_file_name> --save_only
- Saves the cell code to a file specified in <stan_file_name>. Skips
compile step
%%stan -f <stan_file_name> -o <cpp_file_name>
- Saves the cell code to a file specified in <stan_file_name> and outputs the
compiled cpp file to the file name specified by <cpp_file_name>
%% stan -f <stan_file_name> --allow_undefined
- passes the --allow_undefined argument to stanc compiler
%%stan --stanc <stanc_compiler> [default: stanc in $PATH or pystan.stanc]
- Saves the cell code to a file specified in <stan_file_name> and compiles
using the stan compiler specified in <stanc_compiler>.
Optionally, if you specify --stanc pystan, it uses pystan.stanc compiler.
If you want to use specific stanc compiler
use this option (e.g %%stan -f binom.stan --stanc "~/cmdstan-2.16.0/bin/stanc")
'''
_stan_vars = {}
__stan_code_str__ = None
__stan_file__ = None
__model_name__ = None
ip = get_ipython()
args = shlex.split(line.strip())
options = {
k: True if v.startswith('-') else v
for k, v in zip(args, args[1:] + ["--"]) if k.startswith('-')
}
source_filename = options.get('-f', None)
output_filename = options.get('-o', None)
variable_name = options.get('-v', '_stan_model')
allow_undefined = options.get('--allow_undefined', False)
save_only = options.get('--save_only', False)
if save_only is True:
if source_filename is None:
print("Need to specify stan file name (-f) for --save_only")
return
else:
with open(source_filename, 'w') as f:
f.write(cell)
_stan_vars['stan_file'] = source_filename
print(
"File {0} saved..Skipping Compile".format(source_filename))
return
compiler_location = options.get('--stanc', check_program('stanc'))
if not (isinstance(compiler_location, str)):
print("Pls specify a location for stanccompiler or \
specify '--stanc pystan' to use pystan.stanc compiler")
return
temp_dir = tempfile.gettempdir()
if compiler_location is None or compiler_location == 'pystan':
print("Using pystan.stanc compiler..")
else:
print("Using stanc compiler: ", compiler_location)
# We define the source and executable filenames.
if source_filename is None:
temp_name = str(uuid.uuid4())
source_filename = os.path.join(temp_dir,'anon_' + \
temp_name + '.stan')
output_filename = os.path.join(temp_dir,'anon_' + temp_name + \
'_model.cpp')
model_name = 'anon_' + temp_name + '_model'
__stan_code_str__ = cell
__stan_file__ = None
__model_name__ = None
else:
__stan_file = source_filename
model_name = source_filename.split('.')
if len(model_name) > 1:
model_name = '_'.join(model_name[:-1])
else:
model_name = ''.join(model_name)
model_name = model_name + '_model'
__stan_code_str__ = cell
__stan_file__ = source_filename
__model_name__ = model_name
with open(source_filename, 'w') as f:
f.write(cell)
if compiler_location is None or compiler_location == 'pystan':
try:
import pystan as _pystan
try:
out = _pystan.stanc(source_filename, verbose=True)
except Exception as e:
print(e)
return
except ImportError:
print(
"Ensure stan command line (stanc) is installed and in your \
PATH or pystan is installed or specify compiler path using --stanc \
(e.g. %% stan -f model.stan --stanc ./bin/stanc")
return
else:
compile_str = compiler_location
if output_filename is not None:
compile_str = compile_str + ' --o=' + output_filename
else:
compile_str = compile_str + ' --o=' + \
os.path.join(temp_dir,str(uuid.uuid4()) +'.cpp')
if allow_undefined is True:
compile_str = compile_str + ' --allow_undefined '
compile_str = compile_str + ' ' + source_filename
print(compile_str)
compile_output = ip.getoutput(compile_str)
print('\n'.join(compile_output))
_stan_vars[variable_name] = StanMagicOutput(__model_name__,
__stan_code_str__,
__stan_file__)
ip.user_global_ns[variable_name] = _stan_vars[variable_name]
print_compile_success(variable_name, __stan_file__, __model_name__,
__stan_code_str__)
print(" ")
print("Init:")
print(fit.get_inits()[0])
print(fit.get_inits()[1])
print(" ")
print("Adaptation:")
print(fit.get_adaptation_info()[0])
print(fit.get_adaptation_info()[1])
print(" ")
print("Chain 1, Draw 1-3")
print(list(res[:3, 0, :]))
print(" ")
print("Chain 2, Draw 1-3")
print(list(res[:3, 1, :]))
print("\n AS NUMPY")
print("Chain 1, Draw 1-3")
print(res[:3, 0, :])
print(" ")
print("Chain 2, Draw 1-3")
print(res[:3, 1, :])
print(" ")
print(" ### C++ code ###")
print(pystan.stanc("Stan-models/8schools.stan")["cppcode"])
pooled_parameters = """
parameters {
vector[2] beta;
real<lower=0> sigma;
}
"""
pooled_model = """
model {
y ~ normal(beta[1] + beta[2] * x, sigma);
}
"""
pooled_data_dict = {'N': len(log_cbo),
'x': final.controller,
'y': log_cbo}
compiled_model = pystan.stanc(model_code=pooled_data + pooled_parameters + pooled_model)
model = pystan.StanModel(stanc_ret=compiled_model)
pooled_fit = model.sampling(data=pooled_data_dict, iter=1000, chains=2)
pooled_sample = pooled_fit.extract(permuted=True)
b0, m0 = pooled_sample['beta'].T.mean(1)
# plt.scatter(final.controller, log_cbo)
import seaborn as sns
ax = sns.violinplot(final.controller, log_cbo)
xvals = np.linspace(-0.2, 1.2)
# here "0" means "not controller" which is a bit different from the original paper
ax.plot(xvals, m0*xvals+b0, 'r--')
# unpooled *change 10 to 120 in production*
unpooled_model = """data {
int<lower=0> N;
