def _make_example_pax_val_data():
_, example_spectra, _ = simulate_pax.simulate_from_presets(
5.0,
"schlappa",
"ag",
1000,
schlappa_performance.SCHLAPPA_PARAMETERS["energy_loss"],
)
return np.mean(example_spectra["y"], axis=0)
def _run_deconvolution_set(log10_num_electrons):
parameters = pax_simulation_pipeline.DEFAULT_PARAMETERS
impulse_response, pax_spectra, xray_xy = simulate_pax.simulate_from_presets(
log10_num_electrons,
"schlappa",
"ag",
parameters["simulations"],
parameters["energy_loss"],
)
regularizer_widths = parameters["regularizer_widths"]
iterations = 1e5
results = Parallel(n_jobs=-1)(
delayed(_run_single_deconvolution)(
impulse_response, pax_spectra, xray_xy, regularizer_width, iterations
)
for regularizer_width in regularizer_widths
)
return results
def _run_cv(log10_num_electrons, rixs, photoemission, regularization_strengths,
parameters):
impulse_response, pax_spectra, xray_xy = simulate_pax.simulate_from_presets(
log10_num_electrons,
rixs,
photoemission,
parameters["simulations"],
parameters["energy_loss"],
)
deconvolver = deconvolvers.LRFisterGrid(
impulse_response["x"],
impulse_response["y"],
pax_spectra["x"],
parameters["regularizer_widths"],
parameters["iterations"],
xray_xy["y"],
parameters["cv_fold"],
)
deconvolver.fit(np.array(pax_spectra["y"]))
return deconvolver, pax_spectra
def run_ana():
"""Run analysis to create data used in this plot
"""
log10_num_electrons = 4.0
rixs_model = "schlappa"
photoemission_model = "ag"
num_simulations = 100
energy_loss = np.arange(-8, 10, 0.01)
regularization_strengths = np.logspace(-3, -1, 10)
iterations = 1000
impulse_response, pax_spectra, xray_xy = simulate_pax.simulate_from_presets(
log10_num_electrons,
rixs_model,
photoemission_model,
num_simulations,
energy_loss,
)
assess_convergence.run(
impulse_response, pax_spectra, xray_xy, regularization_strengths, iterations
)
def _run_single_regularizer(log10_num_electrons, rixs, photoemission,
regularizer_width, parameters):
"""Run deconvolution for a single input regularization strength
"""
impulse_response, pax_spectra, xray_xy = simulate_pax.simulate_from_presets(
log10_num_electrons,
rixs,
photoemission,
parameters["simulations"],
parameters["energy_loss"],
)
deconvolver = deconvolvers.LRFisterDeconvolve(
impulse_response["x"],
impulse_response["y"],
pax_spectra["x"],
regularizer_width,
iterations=parameters["iterations"],
ground_truth_y=xray_xy["y"],
)
deconvolver.fit(np.array(pax_spectra["y"]))
return deconvolver
def run_set(separation, log10_counts):
deconvolved_list = []
for i in range(NUM_SIMULATIONS):
impulse_response, pax_spectra, xray_xy = simulate_pax.simulate_from_presets(
log10_counts,
["i_doublet", separation],
"fermi",
1000,
np.arange(-0.2, 0.4, 0.002),
)
deconvolver = deconvolvers.LRFisterGrid(
impulse_response["x"],
impulse_response["y"],
pax_spectra["x"],
np.logspace(-4, -2, 10),
ITERATIONS,
xray_xy["y"],
)
_ = deconvolver.fit(np.array(pax_spectra["y"]))
deconvolved_list.append(deconvolver)
print(
f"Completed {str(log10_counts)} counts, {str(separation)} separation, {str(i)} iteration"
)
bootstrap_results = _run_bootstraps(
impulse_response,
pax_spectra,
xray_xy,
deconvolver.best_regularization_strength_,
)
to_save = {
"deconvolved": deconvolved_list,
"bootstraps": bootstrap_results,
"ground_truth": xray_xy,
}
file_name = ("simulated_results/doublet2_" + str(separation) + "_" +
str(log10_counts) + ".pickle")
with open(file_name, "wb") as f:
pickle.dump(to_save, f)