import numpy as np
from gmmmml import (gmm, visualize, utils)
np.random.seed(101)
import logging
for i in range(10):
j = 0
while True:
print(i, j)
j += 1
try:
y, labels, target, kwds = utils.generate_data(N=1000,
K=35,
D=9,
center_box=(-10, 10))
except:
logging.exception("failed")
if j > 3:
raise
continue
else:
break
gmm_kwds = dict(threshold=1e-5,
expected_improvement_fraction=1e-5,
covariance_regularization=1e-2)
dirname = os.path.dirname(path_format)
if not os.path.exists(dirname):
os.makedirs(dirname, exist_ok=True)
common_kwds = dict(dirichlet_concentration=100, isotropy=100, psi=0.1,
scale=max_N/10)
for i, (n, d, k) in enumerate(permutations):
print(f"Generating data for N = {n}, D = {d}, K = {k}")
draws = []
for m in range(M):
try:
X, meta = utils.generate_data(N=n, D=d, K=k, **common_kwds)
except ValueError:
print(f"Failed to generate data for N={n}, D={d}, K={k}")
continue
assert np.all(np.isfinite(X))
draws.append((X, meta))
if plot and m < 1:
fig, ax = plt.subplots()
ax.scatter(X.T[0], X.T[1], s=1)
ax.set_title(f"N={n}; D={d}; K={k}")
Nu = np.unique((orders.T @ steps).flatten())
Du = np.logspace(np.log10(2), np.log10(max_D), 6).astype(int)
N, D = np.meshgrid(Nu, Du)
K = np.clip(np.round([np.random.uniform(2, 0.1 * n) for n in N]), 2, np.inf)
path_format = "data/{K:05d}_{N:06d}_{D:01d}.data"
dirname = os.path.dirname(path_format)
if not os.path.exists(dirname):
os.makedirs(dirname, exist_ok=True)
common_kwds = dict(dirichlet_concentration=100, isotropy=10, psi=0.01)
N, D, K = (ea.flatten().astype(int) for ea in (N, D, K))
for i, (n, d, k) in enumerate(zip(N, D, K)):
print(f"Generating data for N = {n}, D = {d}, K = {k}")
draws = []
for m in range(M):
draws.append(utils.generate_data(N=n, D=d, K=k, **common_kwds))
kwds = dict(N=n, D=d, K=k)
with open(path_format.format(**kwds), "wb") as fp:
pickle.dump(draws, fp)