def run_conditional_estimation_iter(model, result_row, i_iter, config, valid_generator, test_generator, n_bins=10):
logger = logging.getLogger()
logger.info('-'*45)
logger.info(f'iter : {i_iter}')
flush(logger)
iter_directory = os.path.join(model.results_path, f'iter_{i_iter}')
os.makedirs(iter_directory, exist_ok=True)
logger.info('Generate testing data')
test_generator.reset()
X_test, y_test, w_test = test_generator.generate(*config.TRUE, n_samples=config.N_TESTING_SAMPLES)
# SUMMARIES
logger.info('Set up NLL computer')
compute_summaries = model.summary_computer(n_bins=n_bins)
compute_nll = NLLComputer(compute_summaries, valid_generator, X_test, w_test, config=config)
# MEASURE STAT/SYST VARIANCE
logger.info('MEASURE STAT/SYST VARIANCE')
conditional_results = make_conditional_estimation(compute_nll, config)
fname = os.path.join(iter_directory, "no_nuisance.csv")
conditional_estimate = pd.DataFrame(conditional_results)
conditional_estimate['i'] = i_iter
conditional_estimate.to_csv(fname)
return conditional_estimate
def run_estimation_iter(model, result_row, i_iter, config, valid_generator, test_generator, n_bins=10):
logger = logging.getLogger()
logger.info('-'*45)
logger.info(f'iter : {i_iter}')
flush(logger)
iter_directory = os.path.join(model.results_path, f'iter_{i_iter}')
os.makedirs(iter_directory, exist_ok=True)
result_row['i'] = i_iter
result_row['n_test_samples'] = config.N_TESTING_SAMPLES
suffix = f'-mu={config.TRUE.mu:1.2f}_rescale={config.TRUE.rescale}'
logger.info('Generate testing data')
test_generator.reset()
X_test, y_test, w_test = test_generator.generate(*config.TRUE, n_samples=config.N_TESTING_SAMPLES)
# PLOT SUMMARIES
evaluate_summary_computer(model, X_test, y_test, w_test, n_bins=n_bins, prefix='', suffix=suffix, directory=iter_directory)
logger.info('Set up NLL computer')
compute_summaries = model.summary_computer(n_bins=n_bins)
compute_nll = NLLComputer(compute_summaries, valid_generator, X_test, w_test, config=config)
# NLL PLOTS
plot_nll_around_min(compute_nll, config.TRUE, iter_directory, suffix)
# MINIMIZE NLL
logger.info('Prepare minuit minimizer')
minimizer = get_minimizer(compute_nll, config.CALIBRATED, config.CALIBRATED_ERROR)
result_row.update(evaluate_minuit(minimizer, config.TRUE, iter_directory, suffix=suffix))
return result_row.copy()
def _make_rescale_plot_clf(true_rescale, true_mu):
config = Config()
test_generator = Generator(seed=SEED)
X_test, y_test, w_test = test_generator.generate(
true_rescale, true_mu, n_samples=config.N_TESTING_SAMPLES)
i_cv = 0
model = load_some_NN(i_cv=i_cv, cuda=False)
compute_summaries = model.summary_computer(n_bins=N_BINS)
valid_generator = Generator(seed=SEED - 1)
compute_nll = NLLComputer(compute_summaries,
valid_generator,
X_test,
w_test,
config=config)
rescale_array = np.linspace(0.5, 3, 50)
nll_array = [compute_nll(rescale, true_mu) for rescale in rescale_array]
param_name = 'rescale'
p = plt.plot(rescale_array,
nll_array,
label=f'NLL {param_name}={true_rescale}')
plt.axvline(x=true_rescale,
linestyle='--',
color=p[0].get_color(),
label='true value')
def run_iter(compute_summaries, i_cv, i_iter, config, valid_generator,
test_generator, directory):
logger = logging.getLogger()
result_row = dict(i_cv=i_cv, i=i_iter)
iter_directory = os.path.join(directory, f'iter_{i_iter}')
os.makedirs(iter_directory, exist_ok=True)
logger.info(f"True Parameters = {config.TRUE}")
suffix = f'-mu={config.TRUE.mu:1.2f}_rescale={config.TRUE.rescale}'
X_test, y_test, w_test = test_generator.generate(
*config.TRUE, n_samples=config.N_TESTING_SAMPLES)
debug_label(y_test)
compute_nll = NLLComputer(compute_summaries,
valid_generator,
X_test,
w_test,
config=config)
plot_nll_around_min(compute_nll, config.TRUE, iter_directory, suffix)
logger.info('Prepare minuit minimizer')
minimizer = get_minimizer(compute_nll, config.CALIBRATED,
config.CALIBRATED_ERROR)
result_row.update(evaluate_minuit(minimizer, config.TRUE))
return result_row
def get_nll_computer(model, config, valid_generator, test_generator):
X_test, y_test, w_test = test_generator.generate(
*config.TRUE, n_samples=config.N_TESTING_SAMPLES)
compute_summaries = model.summary_computer(n_bins=N_BINS)
compute_nll = NLLComputer(compute_summaries,
valid_generator,
X_test,
w_test,
config=config)
return compute_nll
def run_iter(model, result_row, i_iter, config, valid_generator, test_generator, calib_rescale, n_bins=10):
logger = logging.getLogger()
logger.info('-'*45)
logger.info(f'iter : {i_iter}')
flush(logger)
iter_directory = os.path.join(model.results_path, f'iter_{i_iter}')
os.makedirs(iter_directory, exist_ok=True)
result_row['i'] = i_iter
result_row['n_test_samples'] = config.N_TESTING_SAMPLES
suffix = f'-mu={config.TRUE.mu:1.2f}_rescale={config.TRUE.rescale}'
logger.info('Generate testing data')
test_generator.reset()
X_test, y_test, w_test = test_generator.generate(*config.TRUE, n_samples=config.N_TESTING_SAMPLES)
# PLOT SUMMARIES
evaluate_summary_computer(model, X_test, y_test, w_test, n_bins=n_bins, prefix='', suffix=suffix, directory=iter_directory)
# CALIBRATION
rescale_mean, rescale_sigma = calib_rescale.predict(X_test, w_test)
logger.info('rescale = {} =vs= {} +/- {}'.format(config.TRUE.rescale, rescale_mean, rescale_sigma) )
config.CALIBRATED = Parameter(rescale_mean, config.CALIBRATED.interest_parameters)
config.CALIBRATED_ERROR = Parameter(rescale_sigma, config.CALIBRATED_ERROR.interest_parameters)
for name, value in config.CALIBRATED.items():
result_row[name+"_calib"] = value
for name, value in config.CALIBRATED_ERROR.items():
result_row[name+"_calib_error"] = value
logger.info('Set up NLL computer')
compute_summaries = ClassifierSummaryComputer(model, n_bins=n_bins)
compute_nll = NLLComputer(compute_summaries, valid_generator, X_test, w_test, config=config)
# NLL PLOTS
plot_nll_around_min(compute_nll, config.TRUE, iter_directory, suffix)
# MEASURE STAT/SYST VARIANCE
logger.info('MEASURE STAT/SYST VARIANCE')
conditional_results = make_conditional_estimation(compute_nll, config)
fname = os.path.join(iter_directory, "no_nuisance.csv")
conditional_estimate = pd.DataFrame(conditional_results)
conditional_estimate['i'] = i_iter
conditional_estimate.to_csv(fname)
# MINIMIZE NLL
logger.info('Prepare minuit minimizer')
minimizer = get_minimizer(compute_nll, config.CALIBRATED, config.CALIBRATED_ERROR)
result_row.update(evaluate_minuit(minimizer, config.TRUE))
return result_row.copy(), conditional_estimate
def main():
# BASIC SETUP
logger = set_logger()
args = GB_parse_args(
main_description=
"Training launcher for Gradient boosting on S3D2 benchmark")
logger.info(args)
flush(logger)
# Config
config = Config()
config.TRUE = Parameter(rescale=0.9, mu=0.1)
train_generator = Generator(SEED)
valid_generator = Generator(SEED + 1)
test_generator = Generator(SEED + 2)
X_test, y_test, w_test = test_generator.generate(
*config.TRUE, n_samples=config.N_TESTING_SAMPLES)
# for nuisance in p(nuisance | data)
nuisance_param_sample = [
param_generator().nuisance_parameters for _ in range(25)
]
average_list = []
variance_list = []
result_table = []
for nuisance_params in nuisance_param_sample:
logger.info(f"nuisance_params = {nuisance_params}")
estimator_values = []
for i_cv in range(N_ITER):
clf = build_model(args, i_cv)
parameters = Parameter(*nuisance_params,
config.CALIBRATED.interest_parameters)
print(parameters)
n_samples = config.N_TRAINING_SAMPLES
X_train, y_train, w_train = train_generator.generate(
*parameters, n_samples=n_samples)
logger.info(f"Training {clf.full_name}")
clf.fit(X_train, y_train, w_train)
compute_summaries = ClassifierSummaryComputer(clf, n_bins=10)
nll_computer = NLLComputer(compute_summaries,
valid_generator,
X_test,
w_test,
config=config)
compute_nll = lambda mu: nll_computer(*nuisance_params, mu)
minimizer = get_minimizer(compute_nll)
results = evaluate_minuit(minimizer,
[config.TRUE.interest_parameters])
estimator_values.append(results['mu'])
results['i_cv'] = i_cv
results.update(params_to_dict(parameters, suffix='true'))
result_table.append(results.copy())
average_list.append(np.mean(estimator_values))
variance_list.append(np.var(estimator_values))
model = build_model(args, 0)
model.set_info(DATA_NAME, BENCHMARK_NAME, 0)
save_directory = model.results_path
os.makedirs(save_directory, exist_ok=True)
result_table = pd.DataFrame(result_table)
result_table.to_csv(os.path.join(save_directory, 'results.csv'))
logger.info(f"average_list {average_list}")
logger.info(f"variance_list {variance_list}")
v_stat = np.mean(variance_list)
v_syst = np.var(average_list)
v_total = v_stat + v_syst
logger.info(f"V_stat = {v_stat}")
logger.info(f"V_syst = {v_syst}")
logger.info(f"V_total = {v_total}")
eval_dict = {"V_stat": v_stat, "V_syst": v_syst, "V_total": v_total}
eval_path = os.path.join(save_directory, 'info.json')
with open(eval_path, 'w') as f:
json.dump(eval_dict, f)
