def test_6(self):
""" Test short estimation tasks.
"""
num_agents = np.random.randint(5, 100)
constr = {
"simulation": {
"agents": num_agents
},
"num_periods": np.random.randint(1, 4),
"estimation": {
"maxfun": np.random.randint(0, 5),
"agents": num_agents
},
}
# Simulate a dataset
params_spec, options_spec = generate_random_model(point_constr=constr)
respy_obj = RespyCls(params_spec, options_spec)
write_interpolation_grid(respy_obj)
# Run estimation task.
simulate_observed(respy_obj)
base_x, base_val = respy_obj.fit()
# We also check whether updating the class instance and a single evaluation of
# the criterion function give the same result.
respy_obj.update_optim_paras(base_x)
respy_obj.attr["maxfun"] = 0
alt_x, alt_val = respy_obj.fit()
for arg in [(alt_val, base_val), (alt_x, base_x)]:
np.testing.assert_almost_equal(arg[0], arg[1])
def test_4(self):
""" Test the evaluation of the criterion function for random requests, not just
at the true values.
"""
# Constraints that ensure that two alternative initialization files can be used
# for the same simulated data.
num_agents = np.random.randint(5, 100)
constr = {
"simulation": {
"agents": num_agents
},
"num_periods": np.random.randint(1, 4),
"edu_spec": {
"start": [7],
"max": 15,
"share": [1.0]
},
"estimation": {
"maxfun": 0,
"agents": num_agents
},
}
# Simulate a dataset
params_spec, options_spec = generate_random_model(point_constr=constr)
respy_obj = RespyCls(params_spec, options_spec)
simulate_observed(respy_obj)
# Evaluate at different points, ensuring that the simulated dataset still fits.
params_spec, options_spec = generate_random_model(point_constr=constr)
respy_obj = RespyCls(params_spec, options_spec)
respy_obj.fit()
def test_2(self):
""" This test ensures that the record files are identical.
"""
# Generate random initialization file. The number of periods is higher than
# usual as only FORTRAN implementations are used to solve the random request.
# This ensures that also some cases of interpolation are explored.
constr = {
"program": {
"version": "fortran"
},
"num_periods": np.random.randint(3, 10),
"estimation": {
"maxfun": 0
},
}
params_spec, options_spec = generate_random_model(point_constr=constr)
base_sol_log, base_est_info_log = None, None
base_est_log = None
for is_parallel in [False, True]:
options_spec["program"]["threads"] = 1
options_spec["program"]["procs"] = 1
if is_parallel:
if IS_PARALLELISM_OMP:
options_spec["program"]["threads"] = np.random.randint(
2, 5)
if IS_PARALLELISM_MPI:
options_spec["program"]["procs"] = np.random.randint(2, 5)
respy_obj = RespyCls(params_spec, options_spec)
file_sim = respy_obj.get_attr("file_sim")
simulate_observed(respy_obj)
respy_obj.fit()
# Check for identical records
fname = file_sim + ".respy.sol"
if base_sol_log is None:
base_sol_log = open(fname, "r").read()
assert open(fname, "r").read() == base_sol_log
if base_est_info_log is None:
base_est_info_log = open("est.respy.info", "r").read()
assert open("est.respy.info", "r").read() == base_est_info_log
if base_est_log is None:
base_est_log = open("est.respy.log", "r").readlines()
compare_est_log(base_est_log)
def test_9(self):
""" This test just locks in the evaluation of the criterion function for the
original Keane & Wolpin data. We create an additional initialization files that
include numerous types and initial conditions.
"""
# This ensures that the experience effect is taken care of properly.
open(".restud.respy.scratch", "w").close()
kw_spec, result = random.choice([
("kw_data_one", 10.45950941513551),
("kw_data_two", 45.04552402391903),
("kw_data_three", 74.28253652773714),
("kw_data_one_types", 9.098738585839529),
("kw_data_one_initial", 7.965979149372883),
])
base_path = TEST_RESOURCES_DIR / kw_spec
# Evaluate criterion function at true values.
respy_obj = RespyCls(base_path.with_suffix(".csv"),
base_path.with_suffix(".json"))
respy_obj.unlock()
respy_obj.set_attr("maxfun", 0)
respy_obj.lock()
simulate_observed(respy_obj, is_missings=False)
_, val = respy_obj.fit()
np.testing.assert_allclose(val, result)
def test_5(self):
""" Test the scripts.
"""
# Constraints that ensure that two alternative initialization files can be used
# for the same simulated data.
for _ in range(10):
num_agents = np.random.randint(5, 100)
constr = {
"simulation": {
"agents": num_agents
},
"num_periods": np.random.randint(1, 4),
"edu_spec": {
"start": [7],
"max": 15,
"share": [1.0]
},
"estimation": {
"maxfun": 0,
"agents": num_agents
},
}
# Simulate a dataset
params_spec, options_spec = generate_random_model(
point_constr=constr)
respy_obj = RespyCls(params_spec, options_spec)
simulate_observed(respy_obj)
# Create output to process a baseline.
respy_obj.unlock()
respy_obj.set_attr("maxfun", 0)
respy_obj.lock()
respy_obj.fit()
# Potentially evaluate at different points.
params_spec, options_spec = generate_random_model(
point_constr=constr)
respy_obj = RespyCls(params_spec, options_spec)
single = np.random.choice([True, False])
scripts_check("estimate", respy_obj)
scripts_estimate(single, respy_obj)
def test_2(self):
"""Ensure that the evaluation of the criterion is equal across versions."""
max_draws = np.random.randint(10, 100)
# It seems to be important that max_draws and max_agents is the same
# number because otherwise some functions that read draws from a file
# to ensure compatibility of fortran and python versions won't work.
bound_constr = {"max_draws": max_draws, "max_agents": max_draws}
point_constr = {
"interpolation": {"flag": False},
"program": {"procs": 1, "threads": 1, "version": "python"},
"estimation": {"maxfun": 0},
}
params_spec, options_spec = generate_random_model(
point_constr=point_constr, bound_constr=bound_constr
)
respy_obj = RespyCls(params_spec, options_spec)
num_agents_sim, optim_paras = dist_class_attributes(
respy_obj, "num_agents_sim", "optim_paras"
)
type_shares = optim_paras["type_shares"]
# Simulate a dataset
simulate_observed(respy_obj)
# Iterate over alternative implementations
base_x, base_val = None, None
num_periods = options_spec["num_periods"]
write_draws(num_periods, max_draws)
write_types(type_shares, num_agents_sim)
for version in ["python", "fortran"]:
respy_obj.unlock()
respy_obj.set_attr("version", version)
respy_obj.lock()
x, val = respy_obj.fit()
# Check for the returned parameters.
if base_x is None:
base_x = x
np.testing.assert_allclose(base_x, x)
# Check for the value of the criterion function.
if base_val is None:
base_val = val
np.testing.assert_allclose(base_val, val)
def create_single(idx):
""" This function creates a single test.
"""
dirname = get_random_dirname(5)
os.mkdir(dirname)
os.chdir(dirname)
# The late import is required so a potentially just compiled FORTRAN implementation
# is recognized. This is important for the creation of the regression vault as we
# want to include FORTRAN use cases.
from respy import RespyCls
# We impose a couple of constraints that make the requests manageable.
np.random.seed(idx)
version = np.random.choice(["python", "fortran"])
# only choose from constraint optimizers because we always have some bounds
if version == "python":
optimizer = "SCIPY-LBFGSB"
else:
optimizer = "FORTE-BOBYQA"
constr = {
"program": {
"version": version
},
"preconditioning": {
"type": np.random.choice(["identity", "magnitudes"])
},
"estimation": {
"maxfun":
int(np.random.choice(range(6), p=[0.5, 0.1, 0.1, 0.1, 0.1, 0.1])),
"optimizer":
optimizer,
},
}
constr["flag_estimation"] = True
param_spec, options_spec = generate_random_model(point_constr=constr)
respy_obj = RespyCls(param_spec, options_spec)
simulate_observed(respy_obj)
crit_val = respy_obj.fit()[1]
# In rare instances, the value of the criterion function might be too large and thus
# printed as a string. This occurred in the past, when the gradient preconditioning
# had zero probability observations. We now generate random initialization files
# with smaller gradient step sizes.
if not isinstance(crit_val, float):
raise AssertionError(" ... value of criterion function too large.")
# Cleanup of temporary directories.from
os.chdir("../")
shutil.rmtree(dirname)
return respy_obj.attr, crit_val
def check_single(tests, idx):
""" This function checks a single test from the dictionary.
"""
# Distribute test information.
attr, crit_val = tests[idx]
if not IS_PARALLELISM_OMP or not IS_FORTRAN:
attr["num_threads"] = 1
if not IS_PARALLELISM_MPI or not IS_FORTRAN:
attr["num_procs"] = 1
if not IS_FORTRAN:
attr["version"] = "python"
# In the past we also had the problem that some of the testing machines report
# selective failures when the regression vault was created on another machine.
msg = " ... test is known to fail on this machine"
if "zeus" in socket.gethostname() and idx in []:
print(msg)
return None
if "acropolis" in socket.gethostname() and idx in []:
print(msg)
return None
if "pontos" in socket.gethostname() and idx in []:
print(msg)
return None
# We need to create an temporary directory, so the multiprocessing does not
# interfere with any of the files that are printed and used during the small
# estimation request.
dirname = get_random_dirname(5)
os.mkdir(dirname)
os.chdir(dirname)
# The late import is required so a potentially just compiled FORTRAN implementation
# is recognized. This is important for the creation of the regression vault as we
# want to include FORTRAN use cases.
from respy import RespyCls
params_spec = _params_spec_from_attributes(attr)
options_spec = _options_spec_from_attributes(attr)
respy_obj = RespyCls(params_spec, options_spec)
simulate_observed(respy_obj)
est_val = respy_obj.fit()[1]
is_success = np.isclose(est_val, crit_val, rtol=TOL, atol=TOL)
# Cleanup of temporary directories.from
os.chdir("../")
shutil.rmtree(dirname)
return is_success
def test_3(self):
""" Testing whether the a simulated dataset and the evaluation of the criterion function
are the same for a tiny delta and a myopic agent.
"""
constr = {"estimation": {"maxfun": 0}}
params_spec, options_spec = generate_random_model(point_constr=constr,
myopic=True)
respy_obj = RespyCls(params_spec, options_spec)
optim_paras, num_agents_sim, edu_spec = dist_class_attributes(
respy_obj, "optim_paras", "num_agents_sim", "edu_spec")
write_types(optim_paras["type_shares"], num_agents_sim)
write_edu_start(edu_spec, num_agents_sim)
write_lagged_start(num_agents_sim)
# Iterate over alternative discount rates.
base_data, base_val = None, None
for delta in [0.00, 0.000001]:
respy_obj = RespyCls(params_spec, options_spec)
respy_obj.unlock()
respy_obj.attr["optim_paras"]["delta"] = np.array([delta])
respy_obj.lock()
simulate_observed(respy_obj)
# This parts checks the equality of simulated dataset for the different
# versions of the code.
data_frame = pd.read_csv("data.respy.dat", delim_whitespace=True)
if base_data is None:
base_data = data_frame.copy()
assert_frame_equal(base_data, data_frame)
# This part checks the equality of an evaluation of the criterion function.
_, crit_val = respy_obj.fit()
if base_val is None:
base_val = crit_val
np.testing.assert_allclose(base_val,
crit_val,
rtol=1e-03,
atol=1e-03)
def run(hours):
start, timeout = datetime.now(), timedelta(hours=hours)
count = 0
while True:
print("COUNT", count)
count += 1
# Generate random initialization file
constr = {
"program": {
"version": "fortran"
},
"estimation": {
"maxfun": np.random.randint(0, 50),
"optimizer": "FORT-BOBYQA",
},
}
params_spec, options_spec = generate_random_model(point_constr=constr)
base = None
for is_parallel in [True, False]:
if is_parallel is False:
options_spec["program"]["threads"] = 1
options_spec["program"]["procs"] = 1
else:
if IS_PARALLELISM_OMP:
options_spec["program"]["threads"] = np.random.randint(
2, 5)
if IS_PARALLELISM_MPI:
options_spec["program"]["procs"] = np.random.randint(2, 5)
respy_obj = RespyCls(params_spec, options_spec)
respy_obj = simulate_observed(respy_obj)
_, crit_val = respy_obj.fit()
if base is None:
base = crit_val
np.testing.assert_equal(base, crit_val)
# Timeout.
if timeout < datetime.now() - start:
break
def test_1(self):
"""Ensure that it makes no difference whether the
criterion function is evaluated in parallel or not.
"""
# Generate random initialization file
constr = {
"program": {
"version": "fortran"
},
"estimation": {
"maxfun": np.random.randint(0, 50)
},
}
params_spec, options_spec = generate_random_model(point_constr=constr)
# If delta is a not fixed, we need to ensure a bound-constraint optimizer.
# However, this is not the standard flag_estimation as the number of function
# evaluation is possibly much larger to detect and differences in the updates of
# the optimizer steps depending on the implementation.
if params_spec.loc[("delta", "delta"), "fixed"] is False:
options_spec["estimation"]["optimizer"] = "FORT-BOBYQA"
base = None
for is_parallel in [True, False]:
options_spec["program"]["threads"] = 1
options_spec["program"]["procs"] = 1
if is_parallel:
if IS_PARALLELISM_OMP:
options_spec["program"]["threads"] = np.random.randint(
2, 5)
if IS_PARALLELISM_MPI:
options_spec["program"]["procs"] = np.random.randint(2, 5)
respy_obj = RespyCls(params_spec, options_spec)
respy_obj = simulate_observed(respy_obj)
_, crit_val = respy_obj.fit()
if base is None:
base = crit_val
np.testing.assert_equal(base, crit_val)
def test_single_regression(regression_vault, index):
"""Run a single regression test."""
attr, crit_val = regression_vault[index]
if not IS_PARALLELISM_OMP or not IS_FORTRAN:
attr["num_threads"] = 1
if not IS_PARALLELISM_MPI or not IS_FORTRAN:
attr["num_procs"] = 1
if not IS_FORTRAN:
attr["version"] = "python"
if attr["optimizer_used"] not in OPT_EST_PYTH:
attr["optimizer_used"] = OPT_EST_PYTH[2]
params_spec = _params_spec_from_attributes(attr)
options_spec = _options_spec_from_attributes(attr)
respy_obj = RespyCls(params_spec, options_spec)
simulate_observed(respy_obj)
est_val = respy_obj.fit()[1]
assert np.isclose(est_val, crit_val, rtol=TOL, atol=TOL)
def run(request, is_compile, is_background, is_strict, num_procs):
""" Run the regression tests.
"""
if is_compile:
compile_package(True)
# We can set up a multiprocessing pool right away.
mp_pool = mp.Pool(num_procs)
# The late import is required so a potentially just compiled FORTRAN implementation
# is recognized. This is important for the creation of the regression vault as we
# want to include FORTRAN use cases.
from respy import RespyCls
# Process command line arguments
is_creation = False
is_investigation, is_check = False, False
num_tests, idx = None, None
if request[0] == "create":
is_creation, num_tests = True, int(request[1])
elif request[0] == "check":
is_check, num_tests = True, int(request[1])
elif request[0] == "investigate":
is_investigation, idx = True, int(request[1])
else:
raise AssertionError("request in [create, check. investigate]")
if num_tests is not None:
assert num_tests > 0
if idx is not None:
assert idx >= 0
if is_investigation:
fname = TEST_RESOURCES_DIR / "regression_vault.pickle"
with open(fname, "rb") as p:
tests = pickle.load(p)
attr, crit_val = tests[idx]
params_spec = _params_spec_from_attributes(attr)
options_spec = _options_spec_from_attributes(attr)
respy_obj = RespyCls(params_spec, options_spec)
simulate_observed(respy_obj)
result = respy_obj.fit()[1]
np.testing.assert_almost_equal(result, crit_val, decimal=DECIMALS)
if is_creation:
# We maintain the separate execution in the case of a single processor for
# debugging purposes. The error messages are generally much more informative.
if num_procs == 1:
tests = []
for idx in range(num_tests):
tests += [create_single(idx)]
else:
tests = mp_pool.map(create_single, range(num_tests))
with open(TEST_RESOURCES_DIR / "regression_vault.pickle", "wb") as p:
pickle.dump(tests, p)
return
if is_check:
fname = TEST_RESOURCES_DIR / "regression_vault.pickle"
with open(fname, "rb") as p:
tests = pickle.load(p)
run_single = partial(check_single, tests)
indices = list(range(num_tests))
# We maintain the separate execution in the case of a single processor for
# debugging purposes. The error messages are generally much more informative.
if num_procs == 1:
ret = []
for index in indices:
ret += [run_single(index)]
# We need an early termination if a strict test run is requested.
if is_strict and (False in ret):
break
else:
ret = []
for chunk in get_chunks(indices, num_procs):
ret += mp_pool.map(run_single, chunk)
# We need an early termination if a strict test run is requested. So we
# check whether there are any failures in the last batch.
if is_strict and (False in ret):
break
# This allows to call this test from another script, that runs other tests as
# well.
idx_failures = [i for i, x in enumerate(ret) if x not in [True, None]]
is_failure = False in ret
if len(idx_failures) > 0:
is_failure = True
if not is_background:
send_notification(
"regression", is_failed=is_failure, idx_failures=idx_failures
)
return not is_failure
def run_robustness_test(seed, is_investigation):
"""Run a single robustness test."""
passed = True
error_message = None
np.random.seed(seed)
old_dir = os.getcwd()
t = str(time())[-6:]
if is_investigation is True:
new_dir = join(old_dir, str(seed))
if exists(new_dir):
rmtree(new_dir)
os.mkdir(new_dir)
else:
new_dir = join(old_dir, str(seed) + "_" + t)
os.mkdir(new_dir)
for file in ["career_data.respy.dat", "career_data.respy.pkl"]:
copy(join(old_dir, file), join(new_dir, file))
os.chdir(new_dir)
# We need to impose some constraints so that the random initialization file does
# meet the structure of the empirical dataset. We need to be particularly careful
# with the construction of the maximum level of schooling as we need to rule out
# that anyone in the estimation sample has a value larger then the specified maximum
# value.
version = np.random.choice(["python", "fortran"])
if version == "python":
max_periods = 3
else:
max_periods = 10
num_periods = np.random.randint(1, max_periods)
agents = np.random.randint(500, 1372 + 1)
edu_start = np.random.choice(range(7, 12))
constr = {
"num_periods": num_periods,
"edu_spec": {
"start": [int(edu_start)],
"max": np.random.randint(edu_start + num_periods, 30),
},
"estimation": {
"file": "career_data.respy.dat",
"agents": agents,
"maxfun": np.random.randint(1, 5),
},
"program": {
"version": version
},
}
if version == "fortran":
constr["estimation"]["optimizer"] = "FORT-BOBYQA"
if version == "python":
constr["estimation"]["optimizer"] = "SCIPY-LBFGSB"
params_spec, options_spec = generate_random_model(point_constr=constr)
try:
respy_obj = RespyCls(params_spec, options_spec)
if is_investigation:
write_out_model_spec(respy_obj.attr, str(seed))
respy_obj.fit()
except:
tb = traceback.format_exc()
passed = False
error_message = str(tb)
os.chdir(old_dir)
if is_investigation is False:
rmtree(new_dir)
return passed, error_message
def test_2(self):
"""Compare results from an evaluation of the criterion function at the initial
values."""
args = generate_constraints_dict()
params_spec, options_spec = generate_random_model(**args)
params_spec, options_spec = adjust_model_spec(params_spec, options_spec)
max_draws = args["bound_constr"]["max_draws"]
# At this point, the random initialization file does only provide diagonal
# covariances.
cov_sampled = np.random.uniform(0, 0.01, size=(4, 4)) + np.diag(
np.random.uniform(1.0, 1.5, size=4)
)
chol = np.linalg.cholesky(cov_sampled)
coeffs = chol[np.tril_indices(4)]
params_spec.loc["shocks", "para"] = coeffs
params_spec.loc["shocks", "upper"] = np.nan
params_spec.loc["shocks", "lower"] = np.nan
respy_obj = RespyCls(params_spec, options_spec)
# This flag aligns the random components between the RESTUD program and RESPY
# package. The existence of the file leads to the RESTUD program to write out
# the random components.
(
optim_paras,
edu_spec,
num_agents_est,
num_periods,
num_draws_emax,
num_draws_prob,
tau,
num_agents_sim,
) = dist_class_attributes(
respy_obj,
"optim_paras",
"edu_spec",
"num_agents_est",
"num_periods",
"num_draws_emax",
"num_draws_prob",
"tau",
"num_agents_sim",
)
shocks_cholesky = optim_paras["shocks_cholesky"]
cov = np.matmul(shocks_cholesky, shocks_cholesky.T)
# Simulate sample model using RESTUD code.
transform_respy_to_restud_sim(
optim_paras, edu_spec, num_agents_sim, num_periods, num_draws_emax, cov
)
open(".restud.testing.scratch", "a").close()
cmd = str(TEST_RESOURCES_BUILD / "kw_dp3asim")
subprocess.check_call(cmd, shell=True)
transform_respy_to_restud_est(
optim_paras,
edu_spec,
num_agents_est,
num_draws_prob,
tau,
num_periods,
num_draws_emax,
cov,
)
filenames = ["in.txt", TEST_RESOURCES_DIR / "in_bottom.txt"]
with open("in1.txt", "w") as outfile:
for fname in filenames:
with open(fname) as infile:
outfile.write(infile.read())
draws_standard = np.random.multivariate_normal(
np.zeros(4), np.identity(4), (num_periods, max_draws)
)
with open(".draws.respy.test", "w") as file_:
for period in range(num_periods):
for i in range(max_draws):
fmt = " {0:15.10f} {1:15.10f} {2:15.10f} {3:15.10f}\n"
line = fmt.format(*draws_standard[period, i, :])
file_.write(line)
# We always need the seed.txt
shutil.copy(str(TEST_RESOURCES_DIR / "seed.txt"), "seed.txt")
cmd = str(TEST_RESOURCES_BUILD / "kw_dpml4a")
subprocess.check_call(cmd, shell=True)
Path("seed.txt").unlink()
with open("output1.txt", "r") as searchfile:
# Search file for strings, trim lines and save as variables
for line in searchfile:
if "OLD LOGLF=" in line:
stat = float(shlex.split(line)[2])
break
# Now we also evaluate the criterion function with the RESPY package.
restud_sample_to_respy()
respy_obj = respy.RespyCls(params_spec, options_spec)
respy_obj.attr["file_est"] = "ftest.respy.dat"
open(".restud.respy.scratch", "a").close()
_, val = respy_obj.fit()
Path(".restud.respy.scratch").unlink()
# This ensure that the two values are within 1% of the RESPY value.
np.testing.assert_allclose(
abs(stat), abs(val * num_agents_est), rtol=0.01, atol=0.00
)
def test_10(self):
""" This test ensures that the order of the initial schooling level specified in
the initialization files does not matter for the simulation of a dataset and
subsequent evaluation of the criterion function.
Warning
-------
This test fails if types have the identical intercept as no unique ordering is
determined than.
"""
point_constr = {
"estimation": {
"maxfun": 0
},
# We cannot allow for interpolation as the order of states within each
# period changes and thus the prediction model is altered even if the same
# state identifier is used.
"interpolation": {
"flag": False
},
}
params_spec, options_spec = generate_random_model(
point_constr=point_constr)
respy_obj = RespyCls(params_spec, options_spec)
edu_baseline_spec, num_types, num_paras, optim_paras = dist_class_attributes(
respy_obj, "edu_spec", "num_types", "num_paras", "optim_paras")
# We want to randomly shuffle the list of initial schooling but need to maintain
# the order of the shares.
edu_shuffled_start = np.random.permutation(
edu_baseline_spec["start"]).tolist()
edu_shuffled_share, edu_shuffled_lagged = [], []
for start in edu_shuffled_start:
idx = edu_baseline_spec["start"].index(start)
edu_shuffled_lagged += [edu_baseline_spec["lagged"][idx]]
edu_shuffled_share += [edu_baseline_spec["share"][idx]]
edu_shuffled_spec = copy.deepcopy(edu_baseline_spec)
edu_shuffled_spec["lagged"] = edu_shuffled_lagged
edu_shuffled_spec["start"] = edu_shuffled_start
edu_shuffled_spec["share"] = edu_shuffled_share
# We are only looking at a single evaluation as otherwise the reordering affects
# the optimizer that is trying better parameter values one-by-one. The
# reordering might also violate the bounds.
for i in range(53, num_paras):
optim_paras["paras_bounds"][i] = [None, None]
optim_paras["paras_fixed"][i] = False
# We need to ensure that the baseline type is still in the first position.
types_order = [0] + np.random.permutation(range(1, num_types)).tolist()
type_shares = []
for i in range(num_types):
lower, upper = i * 2, (i + 1) * 2
type_shares += [optim_paras["type_shares"][lower:upper].tolist()]
optim_paras_baseline = copy.deepcopy(optim_paras)
optim_paras_shuffled = copy.deepcopy(optim_paras)
list_ = [
optim_paras["type_shifts"][i, :].tolist() for i in types_order
]
optim_paras_shuffled["type_shifts"] = np.array(list_)
list_ = [type_shares[i] for i in types_order]
optim_paras_shuffled["type_shares"] = np.array(list_).flatten()
base_data, base_val = None, None
k = 0
for optim_paras in [optim_paras_baseline, optim_paras_shuffled]:
for edu_spec in [edu_baseline_spec, edu_shuffled_spec]:
respy_obj.unlock()
respy_obj.set_attr("edu_spec", edu_spec)
respy_obj.lock()
# There is some more work to do to update the coefficients as we
# distinguish between the economic and optimization version of the
# parameters.
x = get_optim_paras(optim_paras, num_paras, "all", True)
shocks_cholesky, _ = extract_cholesky(x)
shocks_coeffs = cholesky_to_coeffs(shocks_cholesky)
x[43:53] = shocks_coeffs
respy_obj.update_optim_paras(x)
respy_obj.reset()
simulate_observed(respy_obj)
# This part checks the equality of simulated dataset.
data_frame = pd.read_csv("data.respy.dat",
delim_whitespace=True)
if base_data is None:
base_data = data_frame.copy()
assert_frame_equal(base_data, data_frame)
# This part checks the equality of a single function evaluation.
_, val = respy_obj.fit()
if base_val is None:
base_val = val
np.testing.assert_almost_equal(base_val, val)
respy_obj.reset()
k += 1
def test_3(self):
"""Ensure that the log looks exactly the same for different versions."""
max_draws = np.random.randint(10, 100)
bound_constr = {"max_draws": max_draws, "max_agents": max_draws}
point_constr = {
"interpolation": {"flag": False},
"program": {"procs": 1, "threads": 1, "version": "python"},
"estimation": {"maxfun": 0},
}
params_spec, options_spec = generate_random_model(
point_constr=point_constr, bound_constr=bound_constr
)
respy_obj = RespyCls(params_spec, options_spec)
num_agents_sim, optim_paras, file_sim = dist_class_attributes(
respy_obj, "num_agents_sim", "optim_paras", "file_sim"
)
# Iterate over alternative implementations
base_sol_log, base_est_info, base_est_log = None, None, None
base_sim_log = None
type_shares = respy_obj.attr["optim_paras"]["type_shares"]
num_periods = options_spec["num_periods"]
edu_spec = options_spec["edu_spec"]
write_draws(num_periods, max_draws)
write_types(type_shares, num_agents_sim)
write_edu_start(edu_spec, num_agents_sim)
write_lagged_start(num_agents_sim)
for version in ["fortran", "python"]:
respy_obj.unlock()
respy_obj.set_attr("version", version)
respy_obj.lock()
simulate_observed(respy_obj)
# Check for identical logging
fname = file_sim + ".respy.sol"
if base_sol_log is None:
base_sol_log = open(fname, "r").read()
assert open(fname, "r").read() == base_sol_log
# Check for identical logging
fname = file_sim + ".respy.sim"
if base_sim_log is None:
base_sim_log = open(fname, "r").read()
assert open(fname, "r").read() == base_sim_log
respy_obj.fit()
if base_est_info is None:
base_est_info = open("est.respy.info", "r").read()
assert open("est.respy.info", "r").read() == base_est_info
if base_est_log is None:
base_est_log = open("est.respy.log", "r").readlines()
compare_est_log(base_est_log)
def test_8(self):
""" We ensure that the number of initial conditions does not matter for the
evaluation of the criterion function if a weight of one is put on the first
group.
"""
num_agents = np.random.randint(5, 100)
constr = {
"simulation": {
"agents": num_agents
},
"num_periods": np.random.randint(1, 4),
"edu_spec": {
"max": np.random.randint(15, 25, size=1).tolist()[0]
},
"estimation": {
"maxfun": 0,
"agents": num_agents
},
"interpolation": {
"flag": False
},
}
params_spec, options_spec = generate_random_model(point_constr=constr)
respy_obj = RespyCls(params_spec, options_spec)
simulate_observed(respy_obj)
base_val, edu_start_base = (None,
np.random.randint(1, 5,
size=1).tolist()[0])
# We need to ensure that the initial lagged activity always has the same
# distribution.
edu_lagged_base = np.random.uniform(size=5).tolist()
for num_edu_start in [1, np.random.choice([2, 3, 4]).tolist()]:
# We always need to ensure that a weight of one is on the first level of
# initial schooling.
options_spec["edu_spec"]["share"] = [
1.0
] + [0.0] * (num_edu_start - 1)
options_spec["edu_spec"][
"lagged"] = edu_lagged_base[:num_edu_start]
# We need to make sure that the baseline level of initial schooling is
# always included. At the same time we cannot have any duplicates.
edu_start = np.random.choice(range(1, 10),
size=num_edu_start,
replace=False).tolist()
if edu_start_base in edu_start:
edu_start.remove(edu_start_base)
edu_start.insert(0, edu_start_base)
else:
edu_start[0] = edu_start_base
options_spec["edu_spec"]["start"] = edu_start
respy_obj = RespyCls(params_spec, options_spec)
simulate_observed(respy_obj)
_, val = respy_obj.fit()
if base_val is None:
base_val = val
np.testing.assert_almost_equal(base_val, val)
def test_1(self):
""" Testing the equality of an evaluation of the criterion function for a random
request.
"""
# Run evaluation for multiple random requests.
is_deterministic = np.random.choice([True, False], p=[0.10, 0.9])
is_interpolated = bool(np.random.choice([True, False], p=[0.10, 0.9]))
is_myopic = np.random.choice([True, False], p=[0.10, 0.9])
max_draws = np.random.randint(11, 100)
num_agents = np.random.randint(10, max_draws)
bound_constr = {"max_draws": max_draws}
point_constr = {
"interpolation": {"flag": is_interpolated},
"program": {"procs": 1, "threads": 1, "version": "python"},
"estimation": {"maxfun": 0, "agents": num_agents},
"simulation": {"agents": num_agents},
"num_periods": np.random.randint(1, 5),
}
num_types = np.random.randint(2, 5)
if is_interpolated:
point_constr["num_periods"] = np.random.randint(3, 5)
params_spec, options_spec = generate_random_model(
bound_constr=bound_constr,
point_constr=point_constr,
deterministic=is_deterministic,
myopic=is_myopic,
num_types=num_types,
)
edu_spec = options_spec["edu_spec"]
num_periods = point_constr["num_periods"]
# The use of the interpolation routines is a another special case. Constructing
# a request that actually involves the use of the interpolation routine is a
# little involved as the number of interpolation points needs to be lower than
# the actual number of states. And to know the number of states each period, I
# need to construct the whole state space.
if is_interpolated:
state_space = StateSpace(
num_periods, num_types, edu_spec["start"], edu_spec["max"]
)
max_states_period = state_space.states_per_period.max()
options_spec["interpolation"]["points"] = np.random.randint(
10, max_states_period
)
# Write out random components and interpolation grid to align the three
# implementations.
write_draws(num_periods, max_draws)
respy_obj = RespyCls(params_spec, options_spec)
write_interpolation_grid(respy_obj)
type_shares = respy_obj.attr["optim_paras"]["type_shares"]
write_types(type_shares, num_agents)
write_edu_start(edu_spec, num_agents)
write_lagged_start(num_agents)
# Clean evaluations based on interpolation grid,
base_val, base_data = None, None
for version in ["python", "fortran"]:
respy_obj = RespyCls(params_spec, options_spec)
# Modify the version of the program for the different requests.
respy_obj.unlock()
respy_obj.set_attr("version", version)
respy_obj.lock()
# Solve the model
respy_obj = simulate_observed(respy_obj)
# This parts checks the equality of simulated dataset for the different
# versions of the code.
data_frame = pd.read_csv("data.respy.dat", delim_whitespace=True)
if base_data is None:
base_data = data_frame.copy()
assert_frame_equal(base_data, data_frame)
# This part checks the equality of an evaluation of the criterion function.
_, crit_val = respy_obj.fit()
if base_val is None:
base_val = crit_val
np.testing.assert_allclose(base_val, crit_val, rtol=1e-05, atol=1e-06)
# We know even more for the deterministic case.
if is_deterministic:
assert crit_val in [-1.0, 0.0]
def run_estimation(which):
""" Run an estimation with the respective release.
"""
os.chdir(which)
import numpy as np
from respy import RespyCls
from respy.pre_processing.model_processing import write_init_file
# We need to make sure that the function simulate_observed() is imported from the original
# package. Otherwise dependencies might not work properly.
import respy
sys.path.insert(0, os.path.dirname(respy.__file__) + "/tests")
from respy.tests.codes.auxiliary import simulate_observed
init_dict = json.load(open("init_dict.respy.json", "r"))
# There was a change in the setup for releases after 1.00. This is only required when
# comparing to v1.0.0.
if "1.0.0" in sys.executable:
init_dict["SHOCKS"]["fixed"] = np.array(init_dict["SHOCKS"]["fixed"])
write_init_file(init_dict)
respy_obj = RespyCls("test.respy.ini")
# This flag ensures a clean switch to the synthetic simulation for cases where the
# simulate_observed() was changed in between releases.
if os.path.exists("../.simulate_observed.cfg"):
respy_obj.simulate()
else:
simulate_observed(respy_obj)
# This flag ensures that the change in the truncation of the wage variable has no effect. We
# simply copy the dataset from the new release to the old.
if ("2.0.0.dev20" in sys.executable) and ("/new" in os.getcwd()):
fnames = glob.glob("data.respy.*")
for fname in fnames:
shutil.copy("../old/" + fname, ".")
# Moving from 2.0.0.dev17 to 2.0.0.dev18 breaks the equality because the simulated datasets
# differ. So, we just copy the one from old. However, this is only relevant if 2.0.0.dev18 is
# the candidate.
if ("2.0.0.dev18" in sys.executable) and ("/new" in os.getcwd()):
os.chdir("../old")
files = glob.glob("data.respy.*")
for file in files:
shutil.copy(file, "../new")
os.chdir("../new")
_, crit_val = respy_obj.fit()
# There was a bug in version 1.0 which might lead to crit_val not to actually take the lowest
# value that was visited by the optimizer. So, we reprocess the log file again to be sure.
if "1.0.0" in sys.executable:
crit_val = 1e10
with open("est.respy.log") as infile:
for line in infile.readlines():
list_ = shlex.split(line)
# Skip empty lines
if not list_:
continue
# Process candidate value
if list_[0] == "Criterion":
try:
value = float(list_[1])
if value < crit_val:
crit_val = value
except ValueError:
pass
pkl.dump(crit_val, open("crit_val.respy.pkl", "wb"))
os.chdir("../")