def const():
X, PFCs, Y, C, N = load_data(500,
1000,
eta,
quality,
pad,
x_dim=4,
momentum_scale=momentum_scale,
n=1000000,
max_particle_select=150,
frac=0.1)
consts[0] = marginal(None, DNN.predict([X, DNN.shuffle(Y)]))
consts[1] = marginal(None, EFN.predict([PFCs[:, :, :3], EFN.shuffle(Y)]))
consts[2] = marginal(None, PFN.predict([PFCs[:, :, :3], PFN.shuffle(Y)]))
consts[3] = marginal(None,
PFN_pid.predict([PFCs[:, :, :4],
PFN.shuffle(Y)]))
def get_values(pt_low, pt_high):
X_test, PFCs_test, Y_test, C_test, N_test = load_data(
pt_low,
pt_high,
eta,
quality,
pad,
x_dim=4,
momentum_scale=momentum_scale,
n=1000000,
max_particle_select=150,
frac=0.1)
test = (X_test, PFCs_test, Y_test, C_test)
num_events = X_test.shape[0]
pt = (pt_high + pt_low) / 2
predictions = np.array(
(dnn_predict(test, DNN, consts[0]), efn_predict(test, EFN, consts[1]),
pfn_predict(test, PFN,
consts[2]), pfn_pid_predict(test, PFN_pid, consts[3]),
cms_predict(test, None, consts[4])))
means, mean_uncertainties = [], []
resolutions, resolution_uncertainties = [], []
MIs = []
for i in range(5):
mean, mean_uncertainty = [*norm.fit(predictions[i, 0])]
resolution, resolution_uncertainty = [*norm.fit(predictions[i, 1])]
means.append(mean / pt)
mean_uncertainties.append(mean_uncertainty / pt)
resolutions.append(resolution / pt)
resolution_uncertainties.append(resolution_uncertainty / pt)
MIs.append(predictions[i, 3])
return means, mean_uncertainties, resolutions, resolution_uncertainties, MIs, num_events
d_multiplier = param_dict["d_multiplier"]
# Dataset Parameters
cache_dir = dataset_dict["cache_dir"]
momentum_scale = dataset_dict["momentum_scale"]
n = dataset_dict["n"]
pad = dataset_dict["pad"]
pt_lower, pt_upper = dataset_dict["pt_lower"], dataset_dict["pt_upper"]
eta = dataset_dict["eta"]
quality = dataset_dict["quality"]
# #############################
# ########## DATASET ##########
# #############################
X, PFCs, Y, C, N = load_data(cache_dir, pt_lower, pt_upper, eta, quality, pad, x_dim = x_dim, momentum_scale = momentum_scale, n = n, max_particle_select = 150, amount = dataset_dict["amount"])
X_test, PFCs_test, Y_test, C_test, N_test = load_data(cache_dir, pt_lower, pt_upper, eta, quality, pad, x_dim = x_dim, momentum_scale = momentum_scale, n = 50)
print(X.shape, PFCs.shape, Y.shape)
# ############################
# ########## MODELS ##########
# ############################
MI_histories = []
retrain_points = []
for train_count in range(retrain + 1):
print("TRAINING %d" % (train_count))
# Pretain
n = dataset_dict["n"]
pad = dataset_dict["pad"]
pt_lower, pt_upper = dataset_dict["pt_lower"], dataset_dict["pt_upper"]
eta = dataset_dict["eta"]
quality = dataset_dict["quality"]
# #############################
# ########## DATASET ##########
# #############################
X, Y, C, N = load_data(cache_dir,
pt_lower,
pt_upper,
eta,
quality,
pad,
momentum_scale=momentum_scale,
n=n,
max_particle_select=None,
amount=dataset_dict["amount"],
return_pfcs=False)
X_test, Y_test, C_test, N_test = load_data(cache_dir,
pt_lower,
pt_upper,
eta,
quality,
pad,
momentum_scale=momentum_scale,
n=50,
return_pfcs=False)
pt_lower, pt_upper = 695, 705
eta = 2.4
quality = 2
epochs = 150
d_multiplier = 0.0
# #############################
# ########## DATASET ##########
# #############################
X_test, PFCs_test, Y_test, C_test, N_test = load_data(
cache_dir,
pt_lower,
pt_upper,
eta,
quality,
pad,
x_dim=4,
momentum_scale=momentum_scale,
n=n,
max_particle_select=150)
test = (X_test, PFCs_test, Y_test, C_test)
plt.hist(N_test,
bins=25,
histtype='step',
color="red",
label="# of Particles",
density=True)
plt.xlabel(r"$N$")
plt.ylabel("Density")