Spaces:
Sleeping
Sleeping
File size: 14,737 Bytes
f270024 3b01b0f f270024 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 |
#!/usr/bin/env python
# coding: utf-8
# In[2]:
import time
import torch
import warnings
import numpy as np
import gradio as gr
import matplotlib.pyplot as plt
# Import Burgers' equation components
from data_burgers import exact_solution as exact_solution_burgers
from model_io_burgers import load_model
from model_v2 import Encoder, Decoder, Propagator_concat as Propagator, Model
from LSTM_model import AE_Encoder, AE_Decoder, AE_Model, PytorchLSTM
# Import Advection-Diffusion components
from data_adv_dif import exact_solution as exact_solution_adv_dif
from model_io_adv_dif import load_model as load_model_adv_dif
from model_adv_dif import Encoder as Encoder2D, Decoder as Decoder2D, Propagator_concat as Propagator2D, Model as Model2D
warnings.filterwarnings("ignore")
# ========== Burgers' Equation Setup ==========
def get_burgers_model(input_dim, latent_dim):
encoder = Encoder(input_dim, latent_dim)
decoder = Decoder(latent_dim, input_dim)
propagator = Propagator(latent_dim)
return Model(encoder, decoder, propagator)
flexi_prop_model = get_burgers_model(128, 2)
checkpoint = torch.load("../1d_viscous_burgers/FlexiPropagator_2025-02-01-10-28-34_3e9656b5_best.pt", map_location='cpu')
flexi_prop_model.load_state_dict(checkpoint['model_state_dict'])
flexi_prop_model.eval()
# AE LSTM models
ae_encoder = AE_Encoder(128)
ae_decoder = AE_Decoder(2, 128)
ae_model = AE_Model(ae_encoder, ae_decoder)
lstm_model = PytorchLSTM()
ae_encoder.load_state_dict(torch.load("../1d_viscous_burgers/LSTM_model/ae_encoder_weights.pth", map_location='cpu'))
ae_decoder.load_state_dict(torch.load("../1d_viscous_burgers/LSTM_model/ae_decoder_weights.pth", map_location='cpu'))
ae_model.load_state_dict(torch.load("../1d_viscous_burgers/LSTM_model/ae_model.pth", map_location='cpu'))
lstm_model.load_state_dict(torch.load("../1d_viscous_burgers/LSTM_model/lstm_weights.pth", map_location='cpu'))
# ========== Helper Functions Burgers ==========
def exacts_equals_timewindow(t_0, Re, time_window=40):
dt = 2 / 500
solutions = [exact_solution_burgers(Re, t) for t in (t_0 + np.arange(0, time_window) * dt)]
solns = torch.tensor(solutions, dtype=torch.float32)[None, :, :]
latents = ae_encoder(solns)
re_normalized = Re / 1000
re_repeated = torch.ones(1, time_window, 1) * re_normalized
return torch.cat((latents, re_repeated), dim=2), latents, solns
# Precompute contour plots
z1_vals = np.linspace(-10, 0.5, 200)
z2_vals = np.linspace(5, 32, 200)
Z1, Z2 = np.meshgrid(z1_vals, z2_vals)
latent_grid = np.stack([Z1.ravel(), Z2.ravel()], axis=1)
# Convert to tensor for decoding
latent_tensors = torch.tensor(latent_grid, dtype=torch.float32)
# Decode latent vectors and compute properties
with torch.no_grad():
decoded_signals = flexi_prop_model.decoder(latent_tensors)
sharpness = []
peak_positions = []
x_vals = np.linspace(0, 2, decoded_signals.shape[1])
dx = x_vals[1] - x_vals[0]
for signal in decoded_signals.numpy():
grad_u = np.gradient(signal, dx)
sharpness.append(np.max(np.abs(grad_u)))
peak_positions.append(x_vals[np.argmax(signal)])
sharpness = np.array(sharpness).reshape(Z1.shape)
peak_positions = np.array(peak_positions).reshape(Z1.shape)
def plot_burgers_comparison(Re, tau, t_0):
dt = 2.0 / 500.0
t_final = t_0 + tau * dt
x_exact = exact_solution_burgers(Re, t_final)
tau_tensor, Re_tensor, xt = torch.tensor([tau]).float()[:, None], torch.tensor([Re]).float()[:, None], torch.tensor([exact_solution_burgers(Re, t_0)]).float()[:, None]
with torch.no_grad():
_, x_hat_tau, *_ = flexi_prop_model(xt, tau_tensor, Re_tensor)
latent_for_lstm, *_ = exacts_equals_timewindow(t_0, Re)
with torch.no_grad():
for _ in range(40, tau):
pred = lstm_model(latent_for_lstm)
pred_with_re = torch.cat((pred, torch.tensor([[Re / 1000]], dtype=torch.float32)), dim=1)
latent_for_lstm = torch.cat((latent_for_lstm[:, 1:, :], pred_with_re.unsqueeze(0)), dim=1)
final_pred_high_dim = ae_decoder(pred.unsqueeze(0))
fig, ax = plt.subplots(figsize=(9, 5))
ax.plot(xt.squeeze(), '--', linewidth=3, alpha=0.5, color="C0")
ax.plot(x_hat_tau.squeeze(), 'D', markersize=5, color="C2")
ax.plot(final_pred_high_dim.squeeze().detach().numpy(), '^', markersize=5, color="C1")
ax.plot(x_exact.squeeze(), linewidth=2, alpha=0.5, color="Black")
ax.set_title(f"Comparison ($t_0$={t_0:.2f} β $t_f$={t_final:.2f}), Ο={tau}", fontsize=14)
ax.legend(["Initial", "Flexi-Prop", "AE LSTM", "True"])
return fig
def burgers_update(Re, tau, t0):
fig1 = plot_burgers_comparison(Re, tau, t0)
# Timing calculations
start = time.time()
_ = flexi_prop_model(torch.randn(1, 1, 128), torch.tensor([[tau]]), torch.tensor([[Re]]))
flexi_time = time.time() - start
start = time.time()
latent_for_lstm, _, _ = exacts_equals_timewindow(t0, Re, 40)
encode_time = time.time() - start
start = time.time()
with torch.no_grad():
for _ in range(40, tau):
pred = lstm_model(latent_for_lstm)
pred_with_re = torch.cat((pred, torch.tensor([[Re / 1000]], dtype=torch.float32)), dim=1)
latent_for_lstm = torch.cat((latent_for_lstm[:, 1:, :], pred_with_re.unsqueeze(0)), dim=1)
recursion_time = time.time() - start
start = time.time()
final_pred_high_dim = ae_decoder(pred.unsqueeze(0))
decode_time = time.time() - start
ae_lstm_total_time = encode_time + recursion_time + decode_time
time_ratio = ae_lstm_total_time / flexi_time
# Time plot
fig, ax = plt.subplots(figsize=(11, 6))
ax.bar(["Flexi-Prop", "AE LSTM (Encode)", "AE LSTM (Recursion)", "AE LSTM (Decode)", "AE LSTM (Total)"],
[flexi_time, encode_time, recursion_time, decode_time, ae_lstm_total_time],
color=["C0", "C1", "C2", "C3", "C4"])
ax.set_ylabel("Time (s)", fontsize=14)
ax.set_title("Computation Time Comparison", fontsize=14)
ax.grid(alpha=0.3)
# Latent space visualization
latent_fig = plot_latent_interpretation(Re, tau, t0)
return fig1, fig, time_ratio, latent_fig
def plot_latent_interpretation(Re, tau, t_0):
tau_tensor = torch.tensor([tau]).float()[:, None]
Re_tensor = torch.tensor([Re]).float()[:, None]
x_t = exact_solution_burgers(Re, t_0)
xt = torch.tensor([x_t]).float()[:, None]
with torch.no_grad():
_, _, _, _, z_tau = flexi_prop_model(xt, tau_tensor, Re_tensor)
z_tau = z_tau.squeeze().numpy()
fig, axes = plt.subplots(1, 2, figsize=(9, 3))
# Sharpness Plot
c1 = axes[0].pcolormesh(Z1, Z2, sharpness, cmap='plasma', shading='gouraud')
axes[0].scatter(z_tau[0], z_tau[1], color='red', marker='o', s=50, label="Current State")
axes[0].set_ylabel("$Z_2$", fontsize=14)
axes[0].set_title("Sharpness Encoding", fontsize=14)
fig.colorbar(c1, ax=axes[0])
axes[0].legend()
# Peak Position Plot
c2 = axes[1].pcolormesh(Z1, Z2, peak_positions, cmap='viridis', shading='gouraud')
axes[1].scatter(z_tau[0], z_tau[1], color='red', marker='o', s=50, label="Current State")
axes[1].set_title("Peak position Encoding", fontsize=14)
fig.colorbar(c2, ax=axes[1], label="Peak Position")
# Remove redundant y-axis labels on the second plot for better aesthetics
axes[1].set_yticklabels([])
# Set a single x-axis label centered below both plots
fig.supxlabel("$Z_1$", fontsize=14)
return fig
# ========== Advection-Diffusion Setup ==========
def get_adv_dif_model(latent_dim, output_dim):
encoder = Encoder2D(latent_dim)
decoder = Decoder2D(latent_dim)
propagator = Propagator2D(latent_dim)
return Model2D(encoder, decoder, propagator)
adv_dif_model = get_adv_dif_model(3, 128)
adv_dif_model, _, _, _ = load_model_adv_dif(
"../2D_adv_dif/FlexiPropagator_2D_2025-01-30-12-11-01_0aee8fb0_best.pt",
adv_dif_model
)
def generate_3d_visualization(Re, t_0, tau):
dt = 2 / 500
t = t_0 + tau * dt
U_initial = exact_solution_adv_dif(Re, t_0)
U_evolved = exact_solution_adv_dif(Re, t)
if np.isnan(U_initial).any() or np.isnan(U_evolved).any():
return None
fig3d = plt.figure(figsize=(12, 5))
ax3d = fig3d.add_subplot(111, projection='3d')
x_vals = np.linspace(-2, 2, U_initial.shape[1])
y_vals = np.linspace(-2, 2, U_initial.shape[0])
X, Y = np.meshgrid(x_vals, y_vals)
surf1 = ax3d.plot_surface(X, Y, U_initial, cmap="viridis", alpha=0.6, label="Initial")
surf2 = ax3d.plot_surface(X, Y, U_evolved, cmap="plasma", alpha=0.8, label="Evolved")
ax3d.set_xlim(-3, 3)
ax3d.set_xlabel("x")
ax3d.set_ylabel("y")
ax3d.set_zlabel("u(x,y,t)")
ax3d.view_init(elev=25, azim=-45)
ax3d.set_box_aspect((2,1,1))
fig3d.colorbar(surf1, ax=ax3d, shrink=0.5, label="Initial")
fig3d.colorbar(surf2, ax=ax3d, shrink=0.5, label="Evolved")
ax3d.set_title(f"Solution Evolution\nInitial ($t_0$={t_0:.2f}) vs Evolved ($t_f$={t:.2f})")
plt.tight_layout()
plt.close(fig3d)
return fig3d
def adv_dif_comparison(Re, t_0, tau):
dt = 2 / 500
exact_initial = exact_solution_adv_dif(Re, t_0)
exact_final = exact_solution_adv_dif(Re, t_0 + tau * dt)
if np.isnan(exact_initial).any() or np.isnan(exact_final).any():
return None
x_in = torch.tensor(exact_initial, dtype=torch.float32)[None, None, :, :]
Re_in = torch.tensor([[Re]], dtype=torch.float32)
tau_in = torch.tensor([[tau]], dtype=torch.float32)
with torch.no_grad():
x_hat, x_hat_tau, *_ = adv_dif_model(x_in, tau_in, Re_in)
pred = x_hat_tau.squeeze().numpy()
if pred.shape != exact_final.shape:
return None
mse = np.square(pred - exact_final)
fig, axs = plt.subplots(1, 3, figsize=(15, 4))
for ax, (data, title) in zip(axs, [(pred, "Model Prediction"),
(exact_final, "Exact Solution"),
(mse, "MSE Error")]):
if title == "MSE Error":
im = ax.imshow(data, cmap="viridis", vmin=0, vmax=1e-2)
plt.colorbar(im, ax=ax, fraction=0.075)
else:
im = ax.imshow(data, cmap="jet")
ax.set_title(title)
ax.axis("off")
plt.tight_layout()
plt.close(fig)
return fig
def update_initial_plot(Re, t_0):
exact_initial = exact_solution_adv_dif(Re, t_0)
fig, ax = plt.subplots(figsize=(5, 5))
im = ax.imshow(exact_initial, cmap='jet')
plt.colorbar(im, ax=ax)
ax.set_title('Initial State')
return fig
# ========== Gradio Interface ==========
with gr.Blocks(title="Flexi-Propagator: PDE Prediction Suite") as app:
gr.Markdown("# Flexi-Propagator: Unified PDE Prediction Interface")
with gr.Tabs():
# 1D Burgers' Equation Tab
with gr.Tab("1D Burgers' Equation"):
gr.Markdown(r"""
## π Flexi-Propagator: Single-Shot Prediction for Nonlinear PDEs
**Governing Equation (1D Burgers' Equation):**
$$
\frac{\partial u}{\partial t} + u \frac{\partial u}{\partial x} = \nu \frac{\partial^2 u}{\partial x^2}
$$
**Key Advantages:**
βοΈ **60-150Γ faster** than AE-LSTM baselines
βοΈ **Parametric control**: Embeds system parameters in latent space
**Physically Interpretable Latent Space - Disentanglement:**
<div align="left">
$$
Z_1 \text{ Encodes Peak Location, } Z_2 \text{ Predominantly Encodes Re (Sharpness)}
$$
</div>
""")
with gr.Row():
with gr.Column():
re_burgers = gr.Slider(425, 2350, 1040, label="Reynolds Number")
tau_burgers = gr.Slider(150, 450, 315, label="Time Steps (Ο)")
t0_burgers = gr.Number(0.4, label="Initial Time")
latent_plot = gr.Plot(label="Latent Space Dynamics")
with gr.Column():
burgers_plot = gr.Plot()
time_plot = gr.Plot()
ratio_out = gr.Number(label="Time Ratio (Flexi Prop/AE LSTM)")
# with gr.Row():
# latent_plot = gr.Plot(label="Latent Space Dynamics")
re_burgers.change(burgers_update, [re_burgers, tau_burgers, t0_burgers],
[burgers_plot, time_plot, ratio_out, latent_plot])
tau_burgers.change(burgers_update, [re_burgers, tau_burgers, t0_burgers],
[burgers_plot, time_plot, ratio_out, latent_plot])
t0_burgers.change(burgers_update, [re_burgers, tau_burgers, t0_burgers],
[burgers_plot, time_plot, ratio_out, latent_plot])
# 2D Advection-Diffusion Tab
with gr.Tab("2D Advection-Diffusion"):
gr.Markdown(r"""
## πͺοΈ 2D Advection-Diffusion Visualization
**Governing Equation:**
$$
\frac{\partial u}{\partial t} + c \frac{\partial u}{\partial x} = \nu \left( \frac{\partial^2 u}{\partial x^2} + \frac{\partial^2 u}{\partial y^2} \right)
$$
""")
with gr.Row():
with gr.Column(scale=1):
re_adv = gr.Slider(1, 10, 9, label="Reynolds Number (Re)")
t0_adv = gr.Number(0.45, label="Initial Time")
tau_adv = gr.Slider(150, 425, 225, label="Tau (Ο)")
initial_plot_adv = gr.Plot(label="Initial State")
with gr.Column(scale=3):
with gr.Row():
three_d_plot_adv = gr.Plot(label="3D Evolution")
with gr.Row():
comparison_plots_adv = gr.Plot(label="Model Comparison")
def adv_update(Re, t0, tau):
return (
generate_3d_visualization(Re, t0, tau),
adv_dif_comparison(Re, t0, tau),
update_initial_plot(Re, t0)
)
for component in [re_adv, t0_adv, tau_adv]:
component.change(adv_update, [re_adv, t0_adv, tau_adv],
[three_d_plot_adv, comparison_plots_adv, initial_plot_adv])
app.load(lambda: adv_update(8, 0.35, 225),
outputs=[three_d_plot_adv, comparison_plots_adv, initial_plot_adv])
app.launch()
# In[ ]:
|