File size: 6,320 Bytes
11c937f 85d39f9 693c63e 0b6eb3e 11c937f | 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 | import numpy as np
import matplotlib.pyplot as plt
from metrics import compare_all_fast_with_curves
COLORS = {
"ALSA": "C0", # blue
"LISA": "C3", # red
}
METRICS = {
"MSE": ["mse", "MSE", "mse_mean"],
"Spectral (JS/KL)": ["spectral_kl", "spectral_js", "spectral_div", "spec_kl", "spectral"],
"ACF-MSE": ["acf_mse", "acf_l2", "acf_err", "acf"],
"MMD$^2$ (RFF)": ["mmd2_rff", "mmd_rff", "mmd2", "mmd"],
}
styles = {
"Truth": dict(color="k", lw=2.3, ls="-"),
"NLSA": dict(color="0.2", lw=2.0, ls="--"),
"ALSA": dict(color=COLORS["ALSA"], lw=2.0, ls="-.", alpha=0.95),
"LISA": dict(color=COLORS["LISA"], lw=2.1, ls=":", alpha=0.95),
}
def standardize_global(F_tX: np.ndarray, eps: float = 1e-12):
mu = F_tX.mean(axis=0, keepdims=True)
sd = F_tX.std(axis=0, keepdims=True) + eps
return (F_tX - mu) / sd, mu, sd
def plot_3d_phase_multi(
bg: np.ndarray,
truth: np.ndarray,
preds: dict[str, np.ndarray],
*,
title: str,
elev: float = 20,
azim: float = 35,
bg_stride: int = 5,
traj_stride: int = 1,
max_bg_points: int = 20000,
styles: dict[str, dict] | None = None,
bg_style: dict | None = None,
axis_names: tuple[str, str, str] = ("x", "y", "z"),
):
"""
bg: (Tbg, 3) background attractor (e.g. train)
truth: (T, 3)
preds: name -> (T, 3)
styles: dict mapping curve name -> matplotlib kwargs
e.g. styles["Truth"] = {...}, styles["NLSA"] = {...}
keys should match preds keys.
"""
bg = np.asarray(bg)
truth = np.asarray(truth)
if bg.ndim != 2 or truth.ndim != 2 or bg.shape[1] != 3 or truth.shape[1] != 3:
raise ValueError(f"bg and truth must be (T,3). Got bg={bg.shape}, truth={truth.shape}")
for k, v in preds.items():
v = np.asarray(v)
if v.ndim != 2 or v.shape[1] != 3:
raise ValueError(f"pred '{k}' must be (T,3). Got {v.shape}")
preds[k] = v
if styles is None:
styles = {}
if bg_style is None:
bg_style = dict(color="0.75", lw=0.8, alpha=0.35)
# subsample background for speed
if bg.shape[0] > max_bg_points:
idx = np.linspace(0, bg.shape[0] - 1, max_bg_points).astype(int)
bgp = bg[idx]
else:
bgp = bg
bgp = bgp[::max(1, int(bg_stride))]
t = truth[::max(1, int(traj_stride))]
fig = plt.figure(figsize=(10, 8))
ax = fig.add_subplot(111, projection="3d")
ax.plot(bgp[:, 0], bgp[:, 1], bgp[:, 2], label="Background (train)", **bg_style)
# Truth styling
truth_style = dict(color="k", lw=2.2)
truth_style.update(styles.get("Truth", {}))
ax.plot(t[:, 0], t[:, 1], t[:, 2], label="Truth", **truth_style)
# Predictions styling (fallback to cycle if not provided)
cycle = plt.rcParams["axes.prop_cycle"].by_key().get("color", ["C0", "C1", "C2", "C3", "C4"])
for i, (name, P) in enumerate(preds.items()):
p = P[::max(1, int(traj_stride))]
style = dict(color=cycle[i % len(cycle)], lw=1.8, alpha=0.95)
style.update(styles.get(name, {}))
ax.plot(p[:, 0], p[:, 1], p[:, 2], label=name, **style)
# mark start point
ax.scatter(t[0, 0], t[0, 1], t[0, 2], color=truth_style.get("color", "k"), s=60)
ax.set_title(title)
ax.set_xlabel(axis_names[0])
ax.set_ylabel(axis_names[1])
ax.set_zlabel(axis_names[2])
ax.view_init(elev=elev, azim=azim)
ax.legend(frameon=False, loc="upper left")
plt.tight_layout()
plt.show()
return None
def make_task(F_test: np.ndarray, a_start: int, ell_ctx: int, steps: int):
"""
Returns:
prefix: (ell_ctx, D)
truth : (steps, D)
"""
assert a_start >= ell_ctx
assert a_start + steps <= F_test.shape[0]
prefix = F_test[a_start - ell_ctx: a_start, :]
truth = F_test[a_start: a_start + steps, :]
return prefix, truth
def pick_metric_key(available_keys: list[str], candidates: list[str]) -> str:
"""Pick the first existing key from a list of candidate names."""
for c in candidates:
if c in available_keys:
return c
# fallback: substring match
for c in candidates:
for k in available_keys:
if c.lower() in k.lower():
return k
raise KeyError(f"None of candidates {candidates} found in keys: {available_keys}")
def eval_multistart(
model_name: str,
predictor_fn, # takes prefix -> (steps,D)
F_test: np.ndarray,
*,
starts: np.ndarray,
ell_ctx: int,
steps: int,
burn_in_metrics: int = 0,
mmd_sample: int = 1024,
seed: int = 0,
dt=0.01,
):
"""
Returns:
scalars_mean: dict[str,float]
scalars_std : dict[str,float]
curves_mean : dict[str,np.ndarray]
"""
scalar_keys = None
scalars_acc = {}
curves_acc = {"mse_by_horizon": [], "mse_per_feature": []}
for i, a_start in enumerate(starts):
prefix, truth = make_task(F_test, int(a_start), ell_ctx, steps)
pred = predictor_fn(prefix)
s, c = compare_all_fast_with_curves(
truth, pred,
burn_in=burn_in_metrics,
acf_max_lag=min(200, steps - 2),
mmd_sample=min(mmd_sample, steps),
seed=seed + i,
include_mmd_rff=True,
dt=dt,
)
if scalar_keys is None:
scalar_keys = list(s.keys())
for k in scalar_keys:
scalars_acc[k] = []
for k in scalar_keys:
scalars_acc[k].append(s[k])
curves_acc["mse_by_horizon"].append(c["mse_by_horizon"])
curves_acc["mse_per_feature"].append(c["mse_per_feature"])
scalars_mean = {k: float(np.nanmean(v)) for k, v in scalars_acc.items()}
scalars_std = {k: float(np.nanstd(v)) for k, v in scalars_acc.items()}
curves_mean = {
"mse_by_horizon": np.nanmean(np.stack(curves_acc["mse_by_horizon"], axis=0), axis=0),
"mse_per_feature": np.nanmean(np.stack(curves_acc["mse_per_feature"], axis=0), axis=0),
}
print(f"\n[{model_name}] ℓ={ell_ctx} mean±std over {len(starts)} starts:")
for k in scalars_mean.keys():
print(f" {k:16s}: {scalars_mean[k]:.6g} ± {scalars_std[k]:.3g}")
return scalars_mean, scalars_std, curves_mean
|