def test_unit_init_print():
"""This test ensures that the init file printing process work as intended. For this
purpose we generate random init file specifications import the resulting files,
write the specifications to another init file, import it again and comparing both
initialization dicts
"""
order = [
"GENERAL",
"CONSTANTS",
"EDUC",
"SIMULATION",
"SOLUTION",
"EXOG_PROC",
]
for _ in range(5):
random_init()
model_params_df, _ = read_model_params_init("test.soepy.pkl")
model_spec = read_model_spec_init("test.soepy.yml", model_params_df)
init_dict_flat = namedtuple_to_dict(model_spec)
init_dict = init_dict_flat_to_init_dict(init_dict_flat)
init_dict2 = read_init_file2("test.soepy.yml")
for key in order:
for subkey in init_dict[key].keys():
if not init_dict[key][subkey] == init_dict2[key][subkey]:
raise AssertionError()
def partiable_simulate(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
model_params_init_file_name,
model_spec_init_file_name,
is_expected=True,
):
# Read in model specification from yaml file
model_params_df, model_params = read_model_params_init(
model_params_init_file_name)
model_spec = read_model_spec_init(model_spec_init_file_name,
model_params_df)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
is_expected,
)
# Simulate agents experiences according to parameters in the model specification
df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
return df
def get_simulate_func(model_params_init_file_name, model_spec_init_file_name):
"""Create the simulation function, such that the state space creation is already
done ."""
# Read in model specification from yaml file
model_params_df, model_params = read_model_params_init(
model_params_init_file_name)
model_spec = read_model_spec_init(model_spec_init_file_name,
model_params_df)
# Get information concerning exogenous processes
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(model_spec,
model_spec.ft_exp_shares_file_name)
prob_exp_pt = gen_prob_init_exp_vector(model_spec,
model_spec.pt_exp_shares_file_name)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
# Create state space
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
partial_simulate = partial(
partiable_simulate,
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
)
return partial_simulate
def simulate(model_params_init_file_name, model_spec_init_file_name):
"""Create a data frame of individuals' simulated experiences."""
# Read in model specification from yaml file
model_params_df, model_params = read_model_params_init(
model_params_init_file_name)
model_spec = read_model_spec_init(model_spec_init_file_name,
model_params_df)
# Obtain model solution
states, indexer, covariates, emaxs = pyth_solve(model_params, model_spec)
# Simulate agents experiences according to parameters in the model specification
df = pyth_simulate(model_params, model_spec, states, indexer, emaxs,
covariates)
return df
def test_pyth_simulate(input_vault, test_id):
"""This test runs a random selection of test regression tests from
our regression test battery.
"""
(
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_func,
expected_df_sim_sol,
) = input_vault[test_id]
exog_educ_shares.to_pickle("test.soepy.educ.shares.pkl")
exog_child_age_shares.to_pickle("test.soepy.child.age.shares.pkl")
exog_child_info.to_pickle("test.soepy.child.pkl")
exog_partner_shares.to_pickle("test.soepy.partner.shares.pkl")
exog_exper_shares_pt.to_pickle("test.soepy.pt.exp.shares.pkl")
exog_exper_shares_ft.to_pickle("test.soepy.ft.exp.shares.pkl")
exog_partner_arrival_info.to_pickle("test.soepy.partner.arrival.pkl")
exog_partner_separation_info.to_pickle("test.soepy.partner.separation.pkl")
model_params_df, model_params = read_model_params_init(random_model_params_df)
model_spec = read_model_spec_init(model_spec_init_dict, model_params_df)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(
model_spec, model_spec.ft_exp_shares_file_name
)
prob_exp_pt = gen_prob_init_exp_vector(
model_spec, model_spec.pt_exp_shares_file_name
)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
calculated_df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
pd.testing.assert_series_equal(
calculated_df.sum(axis=0),
expected_df_sim_sol,
)
cleanup()
def test_unit_data_frame_shape():
"""This test ensures that the shape of the simulated data frame corresponds
to the random specifications of our initialization file.
"""
for _ in range(5):
constr = dict()
constr["AGENTS"] = np.random.randint(10, 100)
constr["PERIODS"] = np.random.randint(7, 10)
constr["EDUC_YEARS"] = [
0, np.random.randint(1, 2),
np.random.randint(3, 5)
]
random_init(constr)
model_params_df, model_params = read_model_params_init(
"test.soepy.pkl")
model_spec = read_model_spec_init("test.soepy.yml", model_params_df)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(
model_spec, model_spec.ft_exp_shares_file_name)
prob_exp_pt = gen_prob_init_exp_vector(
model_spec, model_spec.pt_exp_shares_file_name)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
# Count individuals with each educ level
counts = []
for i in [0, 1, 2]:
counts.append(
df[df["Education_Level"] == i]["Identifier"].nunique())
shape = (constr["AGENTS"] * constr["PERIODS"] -
counts[1] * constr["EDUC_YEARS"][1] -
counts[2] * constr["EDUC_YEARS"][2])
np.testing.assert_array_equal(df.shape[0], shape)
def test_coef_educ_level_specificity():
"""This test ensures that when parameters for a specific
education group are changed, the simulated data for the remaining education
groups does not change."""
constr = dict()
constr["AGENTS"] = 100000
constr["PERIODS"] = 10
random_init(constr)
model_params_base = pd.read_pickle("test.soepy.pkl")
# Draw random education level to change
educ_levels = ["low", "middle", "high"]
random_educ_level = random.choice([0, 1, 2])
param_to_change = f"gamma_f_{educ_levels[random_educ_level]}"
model_params_changed = model_params_base
model_params_changed.loc[("exp_returns_f", param_to_change), "value"] = (
model_params_changed.loc[("exp_returns_f", param_to_change), "value"] *
2)
data = []
for num_sim, i in enumerate([model_params_base, model_params_changed]):
model_params_df, model_params = read_model_params_init(i)
model_spec = read_model_spec_init("test.soepy.yml", model_params_df)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(
model_spec, model_spec.ft_exp_shares_file_name)
prob_exp_pt = gen_prob_init_exp_vector(
model_spec, model_spec.pt_exp_shares_file_name)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
data.append(df)
data_base = data[0]
data_changed = data[1]
for level in (0, 1, 2):
if level == random_educ_level:
continue
data_base_educ_level = data_base[data_base["Education_Level"] == level]
data_changed_educ_level = data_changed[data_changed["Education_Level"]
== level]
true_list = [
col for col in data_changed_educ_level.columns.values
if not pd.Series.equals(data_base_educ_level[col],
data_changed_educ_level[col])
]
pd.testing.assert_frame_equal(data_base_educ_level,
data_changed_educ_level)
def test_shares_according_to_initial_conditions():
"""This test ensures that the shares of individuals with particular characteristics
in the simulated data frame as determined by initial conditions correspond to the probabilities
specified in the init file.
"""
constr = dict()
constr["AGENTS"] = 500000
constr["EDUC_YEARS"] = [0, 0, 0]
constr["PERIODS"] = 2
constr["CHILD_AGE_INIT_MAX"] = 1
constr["INIT_EXP_MAX"] = 2
random_init(constr)
model_params_df, model_params = read_model_params_init("test.soepy.pkl")
model_spec = read_model_spec_init("test.soepy.yml", model_params_df)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(model_spec,
model_spec.ft_exp_shares_file_name)
prob_exp_pt = gen_prob_init_exp_vector(model_spec,
model_spec.pt_exp_shares_file_name)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
# Education level shares
simulated = (
df.groupby(["Education_Level"])["Identifier"].nunique().to_numpy() /
constr["AGENTS"])
np.testing.assert_almost_equal(simulated,
prob_educ_level,
decimal=2,
err_msg="Education level shares mismatch")
# Partner status in initial period
simulated = (df[df["Period"] == 0].groupby(
["Education_Level"])["Partner_Indicator"].mean().to_numpy())
np.testing.assert_almost_equal(simulated,
prob_partner_present,
decimal=2,
err_msg="Partner shares mismatch")
# Child ages in initial period
simulated = (df[df["Period"] == 0].groupby(
["Education_Level"])["Age_Youngest_Child"].value_counts(
normalize=True).sort_index(ascending=True).to_numpy())
prob_child_age_flat = [
item for sublist in prob_child_age for item in sublist
]
np.testing.assert_almost_equal(simulated,
prob_child_age_flat,
decimal=2,
err_msg="Child age shares mismatch")
# Experience in initial period
# Part-time
simulated = (df[df["Period"] == 0].groupby(
["Education_Level"])["Experience_Part_Time"].value_counts(
normalize=True).sort_index(ascending=True).to_numpy())
prob_exp_pt_flat = [item for sublist in prob_exp_pt for item in sublist]
np.testing.assert_almost_equal(
simulated,
prob_exp_pt_flat,
decimal=2,
err_msg="Part-time experience shares mismatch",
)
# Full-time
simulated = (df[df["Period"] == 0].groupby(
["Education_Level"])["Experience_Full_Time"].value_counts(
normalize=True).sort_index(ascending=True).to_numpy())
prob_exp_ft_flat = [item for sublist in prob_exp_ft for item in sublist]
np.testing.assert_almost_equal(
simulated,
prob_exp_ft_flat,
decimal=2,
err_msg="Full-time experience shares mismatch",
)
def test_no_children_no_exp():
"""This test ensures that
i) child age equals -1 in the entire simulates sample,
equivalent to no kid is ever born, if the probability to get a child is zero
for all periods
ii) initial experience is zero if so specified in constraint"""
expected = 0
is_expected = False
constr = {
"AGENTS": 200,
"PERIODS": 10,
"CHILD_AGE_INIT_MAX": -1,
"INIT_EXP_MAX": 0
}
random_init(constr)
model_params_df, model_params = read_model_params_init("test.soepy.pkl")
model_spec = read_model_spec_init("test.soepy.yml", model_params_df)
# Set probability of having children to zero for all periods
prob_child = np.full((model_spec.num_periods, 3), 0.00)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(model_spec,
model_spec.ft_exp_shares_file_name)
prob_exp_pt = gen_prob_init_exp_vector(model_spec,
model_spec.pt_exp_shares_file_name)
prob_partner = gen_prob_partner(model_spec)
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
np.testing.assert_equal(sum(df.dropna()["Age_Youngest_Child"] != -1),
expected)
np.testing.assert_equal(
sum(df[df["Period"] == 0].dropna()["Experience_Part_Time"] != 0),
expected)
np.testing.assert_equal(
sum(df[df["Period"] == 0].dropna()["Experience_Full_Time"] != 0),
expected)
def test_pyth_simulate(input_vault, test_id, is_expected):
"""This test runs a random selection of test regression tests from
our regression test battery.
"""
(
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_func,
expected_df_sim_sol,
) = input_vault[test_id]
model_params_df, model_params = read_model_params_init(random_model_params_df)
model_spec = read_model_spec_init(model_spec_init_dict, model_params_df)
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Calculate utility components
log_wage_systematic, non_consumption_utilities = calculate_utility_components(
model_params, model_spec, states, covariates, is_expected
)
for edu_ind, edu_type in enumerate(["low", "middle", "high"]):
# Now test the full and part time level, 1 and 2 for the wages
relevant_states_ind = (
(states[:, 3] == 1) & (states[:, 4] == 2) & (states[:, 1] == edu_ind)
)
gamma_0 = random_model_params_df.loc[
("const_wage_eq", f"gamma_0_{edu_type}"), "value"
]
gamma_f = random_model_params_df.loc[
("exp_returns_f", f"gamma_f_{edu_type}"), "value"
]
if is_expected:
gamma_p = (
random_model_params_df.loc[
("exp_returns_p_bias", f"gamma_p_bias_{edu_type}"), "value"
]
* gamma_f
)
else:
gamma_p = random_model_params_df.loc[
("exp_returns_p", f"gamma_p_{edu_type}"), "value"
]
wage_calc = gamma_0 + gamma_f * np.log(3) + gamma_p * np.log(2)
np.testing.assert_array_equal(
log_wage_systematic[relevant_states_ind],
np.full(log_wage_systematic[relevant_states_ind].shape, wage_calc),
)
def update_solve_objectes():
vault_file = TEST_RESOURCES_DIR / "regression_vault.soepy.pkl"
vault = {}
with open(vault_file, "rb") as file:
tests_sim_func = pickle.load(file)
for i in range(0, 100):
print(i)
(
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_func,
expected_df_sim_sol,
) = tests_sim_func[i]
exog_educ_shares.to_pickle("test.soepy.educ.shares.pkl")
exog_child_age_shares.to_pickle("test.soepy.child.age.shares.pkl")
exog_child_info.to_pickle("test.soepy.child.pkl")
exog_partner_shares.to_pickle("test.soepy.partner.shares.pkl")
exog_exper_shares_pt.to_pickle("test.soepy.pt.exp.shares.pkl")
exog_exper_shares_ft.to_pickle("test.soepy.ft.exp.shares.pkl")
exog_partner_arrival_info.to_pickle("test.soepy.partner.arrival.pkl")
exog_partner_separation_info.to_pickle("test.soepy.partner.separation.pkl")
model_params_df, model_params = read_model_params_init(random_model_params_df)
model_spec = read_model_spec_init(model_spec_init_dict, model_params_df)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(
model_spec, model_spec.ft_exp_shares_file_name
)
prob_exp_pt = gen_prob_init_exp_vector(
model_spec, model_spec.pt_exp_shares_file_name
)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
# Create state space
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
calculated_df_sim_sol = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
vault[i] = (
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_func,
calculated_df_sim_sol.sum(axis=0),
)
with open(vault_file, "wb") as file:
pickle.dump(vault, file)
cleanup(options="regression")
def input_data():
out = {}
vault = TEST_RESOURCES_DIR / "regression_vault.soepy.pkl"
with open(vault, "rb") as file:
tests = pickle.load(file)
(
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_sol,
expected_df_sim,
) = tests[0]
exog_educ_shares.to_pickle("test.soepy.educ.shares.pkl")
exog_child_age_shares.to_pickle("test.soepy.child.age.shares.pkl")
exog_child_info.to_pickle("test.soepy.child.pkl")
exog_partner_shares.to_pickle("test.soepy.partner.shares.pkl")
exog_exper_shares_pt.to_pickle("test.soepy.pt.exp.shares.pkl")
exog_exper_shares_ft.to_pickle("test.soepy.ft.exp.shares.pkl")
exog_partner_arrival_info.to_pickle("test.soepy.partner.arrival.pkl")
exog_partner_separation_info.to_pickle("test.soepy.partner.separation.pkl")
model_params_df, model_params = read_model_params_init(
random_model_params_df)
for name, tax in [("splitted", True), ("individual", False)]:
# Standard tax_spltting is true
if tax:
pass
else:
model_spec_init_dict["TAXES_TRANSFERS"]["tax_splitting"] = tax
model_spec = read_model_spec_init(model_spec_init_dict,
model_params_df)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(
model_spec, model_spec.ft_exp_shares_file_name)
prob_exp_pt = gen_prob_init_exp_vector(
model_spec, model_spec.pt_exp_shares_file_name)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
prob_partner[:, :, 0, 1] = 0
prob_partner[:, :, 0, 0] = 1
prob_partner_present[:] = 0
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
calculated_df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
out[name] = create_disc_sum_av_utility(
calculated_df, model_spec_init_dict["CONSTANTS"]["delta"])
# Check if really all are single at any time
assert (calculated_df["Male_Wages"] == 0).all()
out["regression_disc_sum"] = -0.10754078488166594
return out
def simulate(model_params_init_file_name,
model_spec_init_file_name,
is_expected=True):
"""Create a data frame of individuals' simulated experiences."""
# Read in model specification from yaml file
model_params_df, model_params = read_model_params_init(
model_params_init_file_name)
model_spec = read_model_spec_init(model_spec_init_file_name,
model_params_df)
# Get information concerning exogenous processes
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = gen_prob_child_init_age_vector(model_spec)
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(model_spec,
model_spec.ft_exp_shares_file_name)
prob_exp_pt = gen_prob_init_exp_vector(model_spec,
model_spec.pt_exp_shares_file_name)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
# Create state space
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
is_expected,
)
# Simulate agents experiences according to parameters in the model specification
df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
return df
def input_data():
vault = TEST_RESOURCES_DIR / "regression_vault.soepy.pkl"
with open(vault, "rb") as file:
tests = pickle.load(file)
(
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_func,
expected_df_sim_sol,
) = tests[0]
exog_educ_shares.to_pickle("test.soepy.educ.shares.pkl")
exog_child_age_shares.to_pickle("test.soepy.child.age.shares.pkl")
exog_child_info.to_pickle("test.soepy.child.pkl")
exog_partner_shares.to_pickle("test.soepy.partner.shares.pkl")
exog_exper_shares_pt.to_pickle("test.soepy.pt.exp.shares.pkl")
exog_exper_shares_ft.to_pickle("test.soepy.ft.exp.shares.pkl")
exog_partner_arrival_info.to_pickle("test.soepy.partner.arrival.pkl")
exog_partner_separation_info.to_pickle("test.soepy.partner.separation.pkl")
model_params_df, model_params = read_model_params_init(
random_model_params_df)
model_spec = read_model_spec_init(model_spec_init_dict, model_params_df)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
states, indexer = pyth_create_state_space(model_spec)
# Create objects that depend only on the state space
covariates = construct_covariates(states, model_spec)
child_age_update_rule = define_child_age_update_rule(model_spec, states)
child_state_indexes = create_child_indexes(states, indexer, model_spec,
child_age_update_rule)
attrs_spec = ["seed_emax", "num_periods", "num_draws_emax"]
draws_emax = draw_disturbances(
*[getattr(model_spec, attr) for attr in attrs_spec], model_params)
log_wage_systematic, non_consumption_utilities = calculate_utility_components(
model_params, model_spec, states, covariates, True)
non_employment_benefits = calculate_non_employment_benefits(
model_spec, states, log_wage_systematic)
deductions_spec = np.array(model_spec.ssc_deductions)
tax_splitting = model_spec.tax_splitting
non_employment_consumption_resources = (
calculate_non_employment_consumption_resources(
deductions_spec,
model_spec.tax_params,
covariates[:, 1],
non_employment_benefits,
tax_splitting,
))
# Solve the model in a backward induction procedure
# Error term for continuation values is integrated out
# numerically in a Monte Carlo procedure
emaxs = pyth_backward_induction(
model_spec,
states,
child_state_indexes,
log_wage_systematic,
non_consumption_utilities,
draws_emax,
covariates,
prob_child,
prob_partner,
non_employment_consumption_resources,
deductions_spec,
)
return states, emaxs, child_state_indexes, prob_child, prob_partner
def input_data():
vault = TEST_RESOURCES_DIR / "regression_vault.soepy.pkl"
with open(vault, "rb") as file:
tests = pickle.load(file)
(
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_func,
expected_df_sim_sol,
) = tests[0]
exog_educ_shares.to_pickle("test.soepy.educ.shares.pkl")
exog_child_age_shares.to_pickle("test.soepy.child.age.shares.pkl")
exog_child_info.to_pickle("test.soepy.child.pkl")
exog_partner_shares.to_pickle("test.soepy.partner.shares.pkl")
exog_exper_shares_pt.to_pickle("test.soepy.pt.exp.shares.pkl")
exog_exper_shares_ft.to_pickle("test.soepy.ft.exp.shares.pkl")
exog_partner_arrival_info.to_pickle("test.soepy.partner.arrival.pkl")
exog_partner_separation_info.to_pickle("test.soepy.partner.separation.pkl")
model_params_df, model_params = read_model_params_init(
random_model_params_df)
model_spec = read_model_spec_init(model_spec_init_dict, model_params_df)
prob_child = gen_prob_child_vector(model_spec)
prob_partner = gen_prob_partner(model_spec)
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
return (
covariates,
states,
non_employment_consumption_resources,
model_spec,
child_age_update_rule,
)
def input_data():
out = {}
vault = TEST_RESOURCES_DIR / "regression_vault.soepy.pkl"
with open(vault, "rb") as file:
tests = pickle.load(file)
(
model_spec_init_dict,
random_model_params_df,
exog_educ_shares,
exog_child_age_shares,
exog_partner_shares,
exog_exper_shares_pt,
exog_exper_shares_ft,
exog_child_info,
exog_partner_arrival_info,
exog_partner_separation_info,
expected_df_sim_sol,
expected_df_sim,
) = tests[6]
exog_educ_shares.to_pickle("test.soepy.educ.shares.pkl")
exog_child_age_shares.to_pickle("test.soepy.child.age.shares.pkl")
exog_child_info.to_pickle("test.soepy.child.pkl")
exog_partner_shares.to_pickle("test.soepy.partner.shares.pkl")
exog_exper_shares_pt.to_pickle("test.soepy.pt.exp.shares.pkl")
exog_exper_shares_ft.to_pickle("test.soepy.ft.exp.shares.pkl")
exog_partner_arrival_info.to_pickle("test.soepy.partner.arrival.pkl")
exog_partner_separation_info.to_pickle("test.soepy.partner.separation.pkl")
model_params_df, model_params = read_model_params_init(random_model_params_df)
ccc_under_3 = np.array(
model_spec_init_dict["TAXES_TRANSFERS"]["child_care_costs"]["under_3"]
)
ccc_3_to_6 = np.array(
model_spec_init_dict["TAXES_TRANSFERS"]["child_care_costs"]["3_to_6"]
)
for name, factor in [("original", 1), ("multiplied", 5)]:
model_spec_init_dict["TAXES_TRANSFERS"]["child_care_costs"]["under_3"] = (
ccc_under_3 * factor
)
model_spec_init_dict["TAXES_TRANSFERS"]["child_care_costs"]["3_to_6"] = (
ccc_3_to_6 * factor
)
params_ind_list = [
("disutil_work", "yes_kids_f_educ_middle"),
("disutil_work", "yes_kids_p_educ_middle"),
("disutil_work", "yes_kids_p_educ_low"),
("disutil_work", "yes_kids_p_educ_high"),
("disutil_work", "child_0_2_f"),
("disutil_work", "child_3_5_f"),
]
for param_ind in params_ind_list:
model_params_df.loc[param_ind, "value"] *= factor
model_spec = read_model_spec_init(model_spec_init_dict, model_params_df)
prob_educ_level = gen_prob_educ_level_vector(model_spec)
prob_child_age = []
prob_child_age_old = gen_prob_child_init_age_vector(model_spec)
for educ_level_list in prob_child_age_old:
educ_level_array = np.array(educ_level_list)
educ_level_array[0] = 1
educ_level_array[1:] = 0
prob_child_age += [educ_level_array]
prob_partner_present = gen_prob_partner_present_vector(model_spec)
prob_exp_ft = gen_prob_init_exp_vector(
model_spec, model_spec.ft_exp_shares_file_name
)
prob_exp_pt = gen_prob_init_exp_vector(
model_spec, model_spec.pt_exp_shares_file_name
)
prob_child = gen_prob_child_vector(model_spec)
prob_child[:, :] = 0
prob_partner = gen_prob_partner(model_spec)
prob_partner[:, 0, 1] = 0
prob_partner[:, 0, 0] = 1
prob_partner_present[:] = 0
# Create state space
(
states,
indexer,
covariates,
child_age_update_rule,
child_state_indexes,
) = create_state_space_objects(model_spec)
# Obtain model solution
non_employment_consumption_resources, emaxs = pyth_solve(
states,
covariates,
child_state_indexes,
model_params,
model_spec,
prob_child,
prob_partner,
False,
)
# Simulate
calculated_df = pyth_simulate(
model_params,
model_spec,
states,
indexer,
emaxs,
covariates,
non_employment_consumption_resources,
child_age_update_rule,
prob_educ_level,
prob_child_age,
prob_partner_present,
prob_exp_ft,
prob_exp_pt,
prob_child,
prob_partner,
is_expected=False,
)
out[name] = calculated_df
return out
