Spaces:
Running on Zero
Running on Zero
File size: 5,093 Bytes
6fe4366 | 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 | """
Plotting helpers for OT energy evaluation and FM diagnostics.
All functions are side-effect free besides writing image files. They raise on
errors, so callers can handle exceptions (keeping compatibility with existing
try/except blocks in scripts).
"""
from __future__ import annotations
from typing import Iterable, Sequence
def _to_numpy1d(x: Iterable) -> "object":
"""Convert a 1D-like iterable (list/torch/tensor/numpy) to a numpy array."""
try:
import numpy as np
except Exception as e: # pragma: no cover
raise RuntimeError("numpy is required for plotting helpers") from e
# Torch tensor support without importing torch at module level
if hasattr(x, "detach") and hasattr(x, "cpu") and hasattr(x, "numpy"):
try:
return x.detach().cpu().numpy()
except Exception:
pass
# Numpy array or array-like
try:
return np.asarray(list(x), dtype=float)
except Exception:
# Best-effort fallback
return np.array(list(x), dtype=float)
def plot_fm_loss(
iterations: Sequence[int],
loss_mean: Sequence[float],
loss_std: Sequence[float],
*,
out_path: str,
show_std: bool = False,
sigma_k: float = 3.0,
dpi: int = 150,
title: str = "FM training loss vs. iteration",
):
import matplotlib.pyplot as plt
its = list(iterations)
lmean = _to_numpy1d(loss_mean)
lstd = _to_numpy1d(loss_std)
fig, ax = plt.subplots(figsize=(7.5, 4.5))
ax.set_title(title)
ax.set_xlabel("iteration")
ax.set_ylabel("loss (mean per selected pair)")
ax.plot(its, lmean, color="C0", label="loss")
if show_std:
ax.fill_between(its, lmean - sigma_k * lstd, lmean + sigma_k * lstd, color="C0", alpha=0.2, label=f"±{sigma_k}σ")
ax.legend(loc="best")
fig.tight_layout()
fig.savefig(out_path, dpi=int(dpi))
plt.close(fig)
def plot_fm_ediff(
iterations: Sequence[int],
ediff_mean: Sequence[float],
ediff_std: Sequence[float],
*,
out_path: str,
sigma_k: float = 3.0,
dpi: int = 150,
title: str = "E_data - E_noise vs. iteration",
):
import matplotlib.pyplot as plt
its = list(iterations)
edmean = _to_numpy1d(ediff_mean)
edstd = _to_numpy1d(ediff_std)
fig, ax = plt.subplots(figsize=(7.5, 4.5))
ax.set_title(title)
ax.set_xlabel("iteration")
ax.set_ylabel("energy difference (mean per selected pair)")
ax.plot(its, edmean, color="C1", label="ΔE mean")
ax.fill_between(its, edmean - sigma_k * edstd, edmean + sigma_k * edstd, color="C1", alpha=0.2, label=f"±{sigma_k}σ")
ax.axhline(0.0, color="k", linewidth=0.8, linestyle=":")
ax.legend(loc="best")
fig.tight_layout()
fig.savefig(out_path, dpi=int(dpi))
plt.close(fig)
def plot_interpolation_energy(
taus: Iterable[float],
energy_means: Sequence[float],
energy_stds: Sequence[float],
*,
mu_noise: float,
sd_noise: float,
mu_data: float,
sd_data: float,
out_path: str,
sigma_k: float = 3.0,
dpi: int = 150,
title: str = "GEM energy vs. interpolation time (t=0 in model)",
):
import matplotlib.pyplot as plt
t_np = _to_numpy1d(taus)
m_np = _to_numpy1d(energy_means)
s_np = _to_numpy1d(energy_stds)
fig, ax = plt.subplots(figsize=(7.5, 4.5))
ax.set_title(title)
ax.set_xlabel("interpolation time (0=noise, 1=data)")
ax.set_ylabel("energy")
ax.plot(t_np, m_np, label="interp mean", color="C0")
ax.fill_between(t_np, m_np - sigma_k * s_np, m_np + sigma_k * s_np, color="C0", alpha=0.2, label=f"interp ±{sigma_k}σ")
# Noise/data reference bands
ax.axhline(mu_noise, color="C1", linestyle="--", label="noise mean")
ax.fill_between([0, 1], mu_noise - sigma_k * sd_noise, mu_noise + sigma_k * sd_noise, color="C1", alpha=0.15, label=f"noise ±{sigma_k}σ")
ax.axhline(mu_data, color="C2", linestyle="--", label="data mean")
ax.fill_between([0, 1], mu_data - sigma_k * sd_data, mu_data + sigma_k * sd_data, color="C2", alpha=0.15, label=f"data ±{sigma_k}σ")
ax.legend(loc="best")
fig.tight_layout()
fig.savefig(out_path, dpi=int(dpi))
plt.close(fig)
def plot_interpolation_grad(
taus: Iterable[float],
grad_means: Sequence[float],
grad_stds: Sequence[float],
*,
out_path: str,
sigma_k: float = 3.0,
dpi: int = 150,
title: str = "||-∇_x E|| vs. interpolation time (t=0 in model)",
):
import matplotlib.pyplot as plt
t_np = _to_numpy1d(taus)
gm_np = _to_numpy1d(grad_means)
gs_np = _to_numpy1d(grad_stds)
fig, ax = plt.subplots(figsize=(7.5, 4.5))
ax.set_title(title)
ax.set_xlabel("interpolation time (0=noise, 1=data)")
ax.set_ylabel("gradient strength (mean per node/edge)")
ax.plot(t_np, gm_np, label="||-∇E|| mean", color="C3")
ax.fill_between(t_np, gm_np - sigma_k * gs_np, gm_np + sigma_k * gs_np, color="C3", alpha=0.2, label=f"±{sigma_k}σ")
ax.legend(loc="best")
fig.tight_layout()
fig.savefig(out_path, dpi=int(dpi))
plt.close(fig)
|