def make_phase_1():
for d, k in [(3, 3), (3, 5), (3, 8), (5, 3), (8, 3)]:
dst = get_dirname(d, 80 * d * k, k, 0.8, makedir=True)
print(dst)
for _ in tqdm(range(100)):
ds = bmcc.GaussianMixture(
n=80 * d * k, k=k, d=d, r=0.8, alpha=40, df=d,
symmetric=False, shuffle=False)
ds.save(os.path.join(dst, str(uuid.uuid4())))
def run_evaluate(args):
"""Evaluate samples"""
path, method_name = args
result_dir = path.replace(BASE_DIR, os.path.join(RESULT_DIR, method_name))
eval_dir = path.replace(BASE_DIR, os.path.join(EVAL_DIR, method_name))
# Ignore if already run
if os.path.exists(eval_dir + "_scores.json"):
return
# Load dataset
dataset = bmcc.GaussianMixture(path, load=True)
hist = np.load(result_dir)['hist']
if hist.shape[0] < 2:
with open(eval_dir + "_scores.json", "w") as f:
f.write(json.dumps({"errored": "no iterations saved"}))
return
# If procedure terminates before 2000it (400 once thinned), use 2nd half of
# samples
bi_base = min(int(hist.shape[0] / 2), 200)
# Base evaluation
# We don't care about the oracle matrix for now, so skip computation
res = bmcc.LstsqResult(dataset.data, hist, burn_in=bi_base)
res.evaluate(
dataset.assignments,
oracle=dataset.oracle,
oracle_matrix=None)
# save trace
save_fig(res.trace(), eval_dir + "_trace")
# Save scores
scores = {
"rand": res.rand_best,
"nmi": res.nmi_best,
"oracle_rand": res.oracle_rand,
"oracle_nmi": res.oracle_nmi,
"num_clusters": int(res.num_clusters[res.best_idx]),
"best_idx": int(res.best_idx),
"aggregation": res.aggregation_best,
"segregation": res.segregation_best,
"oracle_aggregation": res.oracle_aggregation,
"oracle_segregation": res.oracle_segregation,
"iterations": hist.shape[0]
}
with open(eval_dir + "_scores.json", "w") as f:
f.write(json.dumps(scores))
def make_phase_2():
for d in [3, 4, 5, 6, 8, 10, 12, 15, 18, 21]:
for n in [600, 800, 1000]:
dst = get_dirname(d, n, 3, 1.0, makedir=True)
print(dst)
for _ in tqdm(range(100)):
ds = bmcc.GaussianMixture(
n=n, k=3, d=d, r=1.0, alpha=40, df=d,
symmetric=False, shuffle=False)
ds.save(os.path.join(dst, str(uuid.uuid4())))
def run_sample(args):
"""Run MCMC sampling"""
# Unpack
path, method_name = args
dst = path.replace(
BASE_DIR, os.path.join(RESULT_DIR, method_name))
# Ignore if test already run (file present)
if os.path.exists(dst):
return
# Load dataset
dataset = bmcc.GaussianMixture(path, load=True)
# Fetch method
method = METHODS[method_name]
cm = bmcc.NormalWishart(
df=dataset.d,
scale=np.identity(dataset.d) if SCALE_MATRIX else None)
mm = method["mixture"](dataset.k)
# Create model
model = bmcc.BayesianMixture(
data=dataset.data,
sampler=method["sampler"],
component_model=cm,
mixture_model=mm,
assignments=np.zeros(dataset.n).astype(np.uint16),
thinning=5)
# Run iterations (break on exceeding limit)
try:
for i in range(5000):
model.iter()
if np.max(model.assignments) > CLUSTERS_LIMIT:
break
except Exception as e:
print("Exception in {} / {}:".format(method_name, path))
print(e)
np.savez(dst, hist=model.hist)
def make_phase_3():
for k in [5, 10, 15, 20, 30, 40, 50, 60, 70, 80, 100]:
dst = get_dirname(3, 50 * k, k, 1.0, makedir=True)
print(dst)
for _ in tqdm(range(100)):
# means = [
# (40 * k)**(1/3) * np.random.uniform(
# low=-0.5, high=0.5, size=3)
# for _ in range(k)
# ]
means = [
(40 * k)**(1 / 3) * np.array(x)
for x in ghalton.Halton(3).get(k)
]
ds = bmcc.GaussianMixture(
n=50 * k, k=k, d=3, r=1.0, df=3,
symmetric=False, shuffle=False, means=means)
ds.save(os.path.join(dst, str(uuid.uuid4())))
from scipy.stats import poisson
import time
from tqdm import tqdm
import bmcc
# Settings
ITERATIONS = 500
THINNING = 5
POINTS = 500
BURN_IN = 50
# Create dataset
dataset = bmcc.GaussianMixture(n=POINTS,
k=3,
d=3,
r=0.7,
alpha=40,
df=3,
symmetric=False,
shuffle=False)
def hybrid(*args, **kwargs):
for _ in range(5):
bmcc.gibbs(*args, **kwargs)
bmcc.split_merge(*args, **kwargs)
mm = bmcc.MFM(gamma=1, prior=lambda k: poisson.logpmf(k, 3))
# mm = bmcc.DPM(alpha=1, use_eb=False)
cm = bmcc.NormalWishart(df=3)
import bmcc
dataset = bmcc.GaussianMixture(n=1000,
k=3,
d=3,
r=0.7,
alpha=40,
df=3,
symmetric=False,
shuffle=False)
dataset.plot_actual(plot=True)
dataset.save("tmp.npz")
dataset = bmcc.GaussianMixture("tmp.npz", load=True)
dataset.plot_actual(plot=True)