def prepare_XT(X,T, T_scaler=T_scaler, gen_plots=False):
nCMP = 21
X_multi = view_as_windows(X, (1,nCMP,X.shape[2],X.shape[3])).squeeze().reshape((-1, nCMP, X.shape[2], X.shape[3]))
X_multi = np.swapaxes(X_multi,1,3)
X_multi = np.swapaxes(X_multi,1,2)
T_multi = view_as_windows(T, (1, nCMP, T.shape[2])).squeeze().reshape((-1, nCMP, T.shape[2]))[:,nCMP//2,:].squeeze()
X_scaled_multi = X_multi
T_scaled_multi = T_scaler.transform(T_multi)
# extract central CMPs for singleCMP network
X_scaled = X_scaled_multi[:,:,:,nCMP//2:nCMP//2+1]
T_scaled = T_scaled_multi
#%%
if gen_plots:
plt_nb_T(T_multi, dx=c.jgx*c.dx, dz=c.jlogz*c.dx, fname="../latex/Fig/T_multi")
plt_nb_T(1e3*T_scaled, dx=c.jgx*dx, dz=c.jlogz*c.dx, fname="../latex/Fig/T_scaled")
#%%
# show single training sample
sample_reveal = nCMP
plt_nb_T(1e3*np.concatenate((np.squeeze(X_scaled_multi[sample_reveal,:,:,-1]), np.flipud(np.squeeze(X_scaled_multi[sample_reveal,:,:,0]))), axis=0),
title="CMP first | CMP last", dx=200, dz=1e3*dt*c.jdt,
origin_in_middle=True, ylabel="Time(s)", fname="../latex/Fig/X_scaled", cbar_label = "")
print(np.shape(1e3*T_scaled[sample_reveal-(nCMP+1)//2:sample_reveal+(nCMP-1)//2:nCMP]))
plt_nb_T(1e3*T_multi[sample_reveal-(nCMP-1)//2:sample_reveal+(nCMP-1)//2,:],
dx=100, dz=c.dx*c.jlogz,
title="scaled velocity logs")
return X_scaled_multi, T_scaled_multi
def scale_X_data(X_data_test=X_data, X_scaler=None):
generate_scaler = False
if X_scaler == None:
generate_scaler = True
# custom scaling of X_data
X_scaler = StandardScaler() # MinMaxScaler([-1, 1])
print(np.shape(X_data))
X_data_cut=X_data[:,:-3,:,:]
X_matrix = X_data_cut.reshape([X_data_cut.shape[0], -1])
X_scaler.fit(X_matrix)
plt_nb_T(np.flipud(1e3*X_scaler.scale_.reshape([X_data_cut.shape[1], -1])), title="Scale",ylabel="Time(s)", dx=200, dz=1e3*dt*jdt, cbar_label="", fname="../latex/Fig/X_scale")
plt_nb_T(np.flipud(1e3*X_scaler.mean_.reshape([X_data_cut.shape[1], -1])), title="Mean", ylabel="Time(s)", dx=200, dz=1e3*dt*jdt, cbar_label="", fname="../latex/Fig/X_mean")
plt_nb_T(np.flipud(np.log(X_scaler.var_.reshape([X_data_cut.shape[1], -1]))), title="Variance", ylabel="Time(s)", dx=200, dz=1e3*dt*jdt, cbar_label="", fname="../latex/Fig/X_log_var")
X_data_test = X_data_test[:,:-3,:,:]
X_matrix_test = X_data_test.reshape([X_data_test.shape[0], -1])
X_data_test_matrix_scaled = X_scaler.transform(X_matrix_test)
X_data_test_scaled = X_data_test_matrix_scaled.reshape(X_data_test.shape)
X_data_test_scaled = np.clip(X_data_test_scaled, a_min=-1, a_max=1)
#plt_nb_T(1e3*np.squeeze(np.percentile(abs(X_data_test), axis=0, q=10)), title="Percentile")
if generate_scaler:
return X_data_test_scaled, X_scaler
return X_data_test_scaled
def show_model_generation():
stretch_X_train = c.stretch_X_train
vel = generate_model(stretch_X=stretch_X_train,
training_flag=False,
crop_flag=False,
distort_flag=False,
random_state_number=randint(10000))
vel = rsf_to_np("marmvel.hh")
plt_nb_T(aug_flip(vel), dx=4, dz=4, fname="../latex/Fig/marm_aug")
vel = generate_model(stretch_X=stretch_X_train,
distort_flag=False,
random_state_number=c.random_state_number,
show_flag=True)
plt_nb_T(vel, fname="../latex/Fig/cropMarm")
N = np.shape(vel)
vel_example = elastic_transform(np.atleast_3d(vel),
alpha_deform,
sigma_deform,
random_state_number=c.random_state_number,
plot_name="Normal")
N = np.shape(vel)
vel_example = generate_model(stretch_X=stretch_X_train,
training_flag=True,
random_state_number=c.random_state_number,
show_flag=True)
vel1 = generate_model(stretch_X=stretch_X_train,
training_flag=False,
random_state_number=randint(10000))
vel2 = generate_model(stretch_X=stretch_X_train,
training_flag=False,
random_state_number=randint(10000))
vel3 = generate_model(stretch_X=stretch_X_train,
training_flag=False,
random_state_number=randint(10000))
vel4 = generate_model(stretch_X=stretch_X_train,
training_flag=False,
random_state_number=randint(10000))
plt_nb_T(np.concatenate((vel_example, vel1, vel2, vel3, vel4), axis=1),
fname="../latex/Fig/random_model_example")
def test_on_model(folder="marmvel1D",
net_dict=None,
prefix="singleCMP",
model_filename=None,
distort_flag=False,
stretch_X=None,
nCMP_max=nCMP,
generate_rsf_data_flag=True,
jgx=jgx, sxbeg=sxbeg, gxbeg=gxbeg,
X_scaler=X_scaler):
if model_filename==None:
model_filename=f"{folder}.hh"
fig_path = f"../latex/Fig/test_{prefix}_{folder}"
# expand model
model_output="vel_test.rsf"
print(model_output)
vel_test = generate_model(model_input=model_filename,
model_output=model_output,
stretch_X=stretch_X,
random_state_number=const.random_state_number,
distort_flag=distort_flag,
crop_flag=False)
# model data
if generate_rsf_data_flag:
cmd(f"mkdir {folder}")
cmd(f"cp {model_output} {folder}/{model_output}")
# check stability
print(f"you chose dt = {dt}, dt < {dx/np.max(vel_test):.4f} should be chosen for stability \n")
# force stability
assert dt < dx/np.max(vel_test)
generate_rsf_data(model_name=f"{folder}/vel_test.rsf",
shots_out=f"{folder}/shots_cmp_test.rsf",
logs_out=f"{folder}/logs_test.rsf")
# read data
X_data_test, T_data_test = read_rsf_to_np(shots_rsf=f"{folder}/shots_cmp_test.rsf",
logs_rsf=f"{folder}/logs_test.rsf")
# X_scaled
X_scaled = scale_X_data(X_data_test, X_scaler)
nCMP = int(net_dict["0"].input.shape[3])
X_scaled, T_data_test = make_multi_CMP_inputs(X_scaled, T_data_test, nCMP_max)
sample_reveal = nCMP_max+1
plt_nb_T(1e3*np.concatenate((np.squeeze(X_scaled[sample_reveal,:,:,-1]), np.flipud(np.squeeze(X_scaled[sample_reveal,:,:,0]))), axis=0),
title="CMP first | CMP last", dx=200, dz=1e3*dt*jdt,
origin_in_middle=True, ylabel="Time(s)", fname=f"{fig_path}_X_scaled", cbar_label = "")
if nCMP == 1:
X_scaled = X_scaled[:,:,:,nCMP_max//2:nCMP_max//2+1]
# predict with all networks and save average
T_pred_total = np.zeros_like(net_dict["0"].predict(X_scaled))
T_pred_dict = np.zeros((2*len(net_dict), T_pred_total.shape[0], T_pred_total.shape[1]))
iNet=0
for net in net_dict.values():
T_pred_tmp = net.predict(X_scaled)
T_pred_tmp = T_scaler.inverse_transform(T_pred_tmp)
T_pred_dict[iNet,:,:] = T_pred_tmp
T_pred_tmp = net.predict(np.flip(X_scaled, axis=3))
T_pred_tmp = T_scaler.inverse_transform(T_pred_tmp)
T_pred_dict[iNet+1,:,:] = T_pred_tmp
iNet += 2
T_pred = np.mean(T_pred_dict, axis=0)
variance = np.var(T_pred_dict, axis=0)
plt_nb_T(np.sqrt(variance), title="Standard deviation",
dx=jgx*dx, dz=jlogz*dx,
fname=f"{fig_path}_inverted_std_dev",
vmin=0.05, vmax=1)
# plt_nb_T(T_pred-T_data_test, title="Pred-True",
# dx=jgx*dx, dz=jlogz*dx,
# fname=f"{fig_path}_inverted_std_dev",
# vmin=-1, vmax=1)
plt_nb_T(T_pred, title=f"{prefix} estimate, NRMS={nrms(T_pred, T_data_test):.1f}%",
dx=jgx*dx, dz=jlogz*dx,
vmin=np.min(1e-3*T_data_test),
vmax=np.max(1e-3*T_data_test),
fname=f"{fig_path}_inverted")
plt_nb_T(T_data_test,
dx=jgx*dx, dz=jlogz*dx,
fname=f"{fig_path}_true",
title=f"True, R2 = {r2_score(T_pred.flatten(), T_data_test.flatten()):.2f}")
print(np.shape(1e3*T_scaled[sample_reveal-(nCMP+1)//2:sample_reveal+(nCMP-1)//2:nCMP]))
if distort_flag:
vel_alpha = (0.9+0.2*resize(np.random.rand(5,10), vel.shape))
#print(vel_alpha)
vel *= vel_alpha
# add water
# vel = np.concatenate((1500*np.ones((vel.shape[0], 20)), vel),
# axis=1)
#vel = ndimage.median_filter(vel, size=(7,3))
#vel = 1500 * np.ones_like(vel)
if verbose:
print(f"Writing to {model_output}")
np_to_rsf(vel, model_output)
return vel
vel = generate_model(stretch_X=stretch_X_train, training_flag=False, random_state_number=randint(10000))
plt_nb_T(vel)
#%%
# Stretched Marmousi
vel = generate_model(stretch_X=stretch_X_train, distort_flag = False)
plt_nb_T(vel, fname="../latex/Fig/cropMarm")
N = np.shape(vel)
#%%
# vel_example = elastic_transform(np.atleast_3d(vel), alpha_deform//2, sigma_deform,
# random_state_number=const.random_state_number, plot_name="Mild")
vel_example = elastic_transform(np.atleast_3d(vel), alpha_deform, sigma_deform,
random_state_number=const.random_state_number, plot_name="Normal")
T_multi = view_as_windows(T, (1, nCMP, T.shape[2])).squeeze().reshape((-1, nCMP, T.shape[2]))[:,nCMP//2,:].squeeze()
# create scaler for the outputs
T_scaler = StandardScaler().fit(T_multi)
scale = np.copy(T_scaler.scale_)
mean = np.copy(T_scaler.mean_)
np.save("scale", scale)
np.save("mean", mean)
#%%
T_scaler.scale_[:] = 1
T_scaler.mean_[:] = 0
# X has the format (model, CMP, offset, time)
plt_nb_T(X[1,:10, -1,:200], title="Common offset (600 m) gather", dx=c.dx*c.jgx*2, dz=1e3*dt*c.jdt,
origin_in_middle=True, ylabel="Time(s)", fname="../latex/Fig/X_short_offset", vmin=-1e-4, vmax=1e-4)
plt_nb_T(T[1,:10,:100], title="Model", dx=c.dx*c.jgx*2, dz=c.dx*c.jlogz,
origin_in_middle=True, ylabel="Time(s)", fname="../latex/Fig/X_short_offset")
#X=X[:,:-3,:,:]
#%%
def prepare_XT(X,T, T_scaler=T_scaler, gen_plots=False):
nCMP = 21
X_multi = view_as_windows(X, (1,nCMP,X.shape[2],X.shape[3])).squeeze().reshape((-1, nCMP, X.shape[2], X.shape[3]))
X_multi = np.swapaxes(X_multi,1,3)
X_multi = np.swapaxes(X_multi,1,2)
T_multi = view_as_windows(T, (1, nCMP, T.shape[2])).squeeze().reshape((-1, nCMP, T.shape[2]))[:,nCMP//2,:].squeeze()