Skip to content

Energy Fn Vis

Source https://github.com/soran-ghaderi/torchebm/blob/master/examples/visualization/basic/energy_fn_vis.py.

  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
#!/usr/bin/env python
import os
import torch
import matplotlib.pyplot as plt
import numpy as np

# Add parent directory to path to import torchebm
import sys

sys.path.append("../../..")

from torchebm.core import (
    GaussianEnergy,
    DoubleWellEnergy,
    RosenbrockEnergy,
    RastriginEnergy,
    AckleyEnergy,
)
from torchebm.utils.visualization import (
    plot_2d_energy_landscape,
    plot_3d_energy_landscape,
    plot_samples_on_energy,
    plot_sample_trajectories,
)
from torchebm.samplers.langevin_dynamics import LangevinDynamics

# Create output directory if it doesn't exist
os.makedirs("../../../docs/assets/images/e_functions", exist_ok=True)

# Set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")

# Generate 2D visualizations for all energy functions

# Gaussian Energy
gaussian_energy = GaussianEnergy(
    mean=torch.tensor([0.0, 0.0]), cov=torch.tensor([[1.0, 0.5], [0.5, 1.0]])
)
fig = plot_2d_energy_landscape(
    energy_fn=gaussian_energy,
    title="Gaussian Energy",
    device=device,
    save_path="../../../docs/assets/images/e_functions/gaussian.png",
)
plt.close(fig)

# Double Well Energy
double_well_energy = DoubleWellEnergy(barrier_height=2.0)
fig = plot_2d_energy_landscape(
    energy_fn=double_well_energy,
    title="Double Well Energy",
    device=device,
    save_path="../../../docs/assets/images/e_functions/double_well.png",
)
plt.close(fig)

# Rosenbrock Energy
rosenbrock_energy = RosenbrockEnergy(a=1.0, b=100.0)
fig = plot_2d_energy_landscape(
    energy_fn=rosenbrock_energy,
    x_range=(-2, 2),
    y_range=(-1, 3),
    title="Rosenbrock Energy",
    device=device,
    save_path="../../../docs/assets/images/e_functions/rosenbrock.png",
)
plt.close(fig)

# Rastrigin Energy
rastrigin_energy = RastriginEnergy(a=10.0)
fig = plot_2d_energy_landscape(
    energy_fn=rastrigin_energy,
    x_range=(-5.12, 5.12),
    y_range=(-5.12, 5.12),
    title="Rastrigin Energy",
    device=device,
    save_path="../../../docs/assets/images/e_functions/rastrigin.png",
)
plt.close(fig)

# Ackley Energy
ackley_energy = AckleyEnergy(a=20.0, b=0.2, c=2 * np.pi)
fig = plot_2d_energy_landscape(
    energy_fn=ackley_energy,
    x_range=(-5, 5),
    y_range=(-5, 5),
    title="Ackley Energy",
    device=device,
    save_path="../../../docs/assets/images/e_functions/ackley.png",
)
plt.close(fig)

# Generate sampling visualizations

# Gaussian sampling
sampler = LangevinDynamics(
    energy_function=gaussian_energy, step_size=0.01, noise_scale=1.0
)

# Initial states far from the mean
initial_states = torch.tensor(
    [[-3.0, -3.0], [3.0, 3.0], [-3.0, 3.0], [3.0, -3.0]], dtype=torch.float32
)

# Run sampling with trajectory tracking
samples, trajectories = sampler.sample(
    x=initial_states, n_steps=200, return_trajectory=True
)

# Plot trajectories
fig = plot_sample_trajectories(
    trajectories=trajectories,
    energy_fn=gaussian_energy,
    title="Sampling Trajectories on Gaussian Energy",
    device=device,
    save_path="../docs/assets/images/e_functions/gaussian_trajectories.png",
)
plt.close(fig)

# Double well sampling
sampler = LangevinDynamics(
    energy_function=double_well_energy,
    step_size=0.01,
    noise_scale=1.0,
)

# Initial states in the middle of the barrier
initial_states = torch.tensor(
    [[0.0, 0.0], [0.5, 0.5], [-0.5, -0.5], [0.0, 0.5]], dtype=torch.float32
)

# Run sampling with trajectory tracking
samples, trajectories = sampler.sample(
    x=initial_states, n_steps=300, return_trajectory=True
)

# Plot trajectories
fig = plot_sample_trajectories(
    trajectories=trajectories,
    energy_fn=double_well_energy,
    title="Sampling Trajectories on Double Well Energy",
    device=device,
    save_path="../docs/assets/images/e_functions/double_well_trajectories.png",
)
plt.close(fig)

# Rastrigin sampling
sampler = LangevinDynamics(
    energy_function=rastrigin_energy,
    step_size=0.005,
    noise_scale=0.5,
)

# Initial states far from origin
initial_states = torch.tensor(
    [[4.0, 4.0], [-4.0, -4.0], [4.0, -4.0], [-4.0, 4.0]], dtype=torch.float32
)

# Run sampling with trajectory tracking
samples, trajectories = sampler.sample(
    x=initial_states, n_steps=400, return_trajectory=True
)

# Plot trajectories
fig = plot_sample_trajectories(
    trajectories=trajectories,
    energy_fn=rastrigin_energy,
    title="Sampling Trajectories on Rastrigin Energy",
    x_range=(-5.12, 5.12),
    y_range=(-5.12, 5.12),
    device=device,
    save_path="../docs/assets/images/e_functions/rastrigin_trajectories.png",
)
plt.close(fig)

print("All visualizations generated successfully.")