Skip to content
GitLab
Commit 85d8c6c9 authored by Tiago Peixoto's avatar Tiago Peixoto
Browse files

Improve test_inference_mcmc.py

parent 48b231bd
Hide whitespace changes
Inline Side-by-side
doc/test_inference_mcmc.py
View file @ 85d8c6c9
......@@ -8,6 +8,25 @@ import numpy.random
from numpy.random import randint
import scipy.stats
try:
from data_cache import FSDataCache
cache = FSDataCache()
def get_lock(name, params):
return cache.get_lock(name, params, block=False)
def get_cache(name, params):
return cache.get(name, params)
def put_cache(name, params, val):
return cache.put(name, params, val)
except ImportError:
from contextlib import contextmanager
@contextmanager
def get_lock(name, params):
return None
def get_cache(name, params):
raise KeyError()
def put_cache(name, params, val):
pass
numpy.random.seed(43)
seed_rng(43)
......@@ -39,7 +58,7 @@ for directed in [True, False]:
plot(res)
gca().set_ylim(-.1, .1)
savefig("test_mcmc_reverse_prob_directed%s.pdf" % directed)
savefig("test_mcmc/test_mcmc_reverse_prob_directed%s.pdf" % directed)
for directed in [True, False]:
clf()
......@@ -88,12 +107,12 @@ for directed in [True, False]:
(v, p))
clf()
plot(res)
savefig("test_mcmc_move_prob_directed%s_v%d.pdf" % (directed, int(v)))
savefig("test_mcmc/test_mcmc_move_prob_directed%s_v%d.pdf" % (directed, int(v)))
plot(res)
gca().set_ylim(-.1, .1)
savefig("test_mcmc_move_prob_directed%s.pdf" % directed)
savefig("test_mcmc/test_mcmc_move_prob_directed%s.pdf" % directed)
g = graph_union(complete_graph(4), complete_graph(4))
g.add_edge(3, 4)
......@@ -103,108 +122,201 @@ for i in range(3 * 8):
t = vs[randint(4) + 4]
g.add_edge(s, t)
for directed in [False, True]:
g.set_directed(directed)
for nested in [False, True]:
if nested:
state = minimize_nested_blockmodel_dl(g, deg_corr=True)
else:
state = minimize_blockmodel_dl(g, deg_corr=True)
hists = {}
cs = list(reversed([numpy.inf, 1, 0.1, 0.01, "gibbs", -numpy.inf, -1]))
inits = [1, "N"]
for init in inits:
if nested:
bs = [g.get_vertices()] if init == "N" else [zeros(1)]
state = state.copy(bs=bs + [zeros(1)] * 6, sampling=True)
else:
state = state.copy(B=g.num_vertices() if init == "N" else 1)
for i, c in enumerate(cs):
if c != "gibbs":
mcmc_args=dict(beta=.3, c=abs(c), niter=40)
else:
mcmc_args=dict(beta=.3, niter=40)
if nested:
get_B = lambda s: [sl.get_nonempty_B() for sl in s.levels]
else:
get_B = lambda s: [s.get_nonempty_B()]
hists[(c, init)] = mcmc_equilibrate(state,
mcmc_args=mcmc_args,
gibbs=c=="gibbs",
multiflip = c != "gibbs" and c < 0,
wait=6000,
nbreaks=6,
callback=get_B,
verbose=(1, "c = %s " % str(c)) if verbose else False,
history=True)
output = open("test_mcmc_nested%s_directed%s-output" % (nested, directed), "w")
for c1 in cs:
for init1 in inits:
for c2 in cs:
for init2 in inits:
try:
if (c2, init2) < (c1, init1):
continue
except TypeError:
pass
Ss1 = array(list(zip(*hists[(c1, init1)]))[2])
Ss2 = array(list(zip(*hists[(c2, init2)]))[2])
# add very small normal noise, to solve discreteness issue
Ss1 += numpy.random.normal(0, 1e-2, len(Ss1))
Ss2 += numpy.random.normal(0, 1e-2, len(Ss2))
D, p = scipy.stats.ks_2samp(Ss1, Ss2)
D_c = 1.63 * sqrt((len(Ss1) + len(Ss2)) / (len(Ss1) * len(Ss2)))
print("nested:", nested, "directed:", directed, "c1:", c1,
"init1:", init1, "c2:", c2, "init2:", init2, "D:",
D, "D_c:", D_c, "p-value:", p,
file=output)
if p < .001:
print(("Warning, distributions for directed=%s (c1, c2) = " +
"(%s, %s) are not the same, with a p-value: %g (D=%g, D_c=%g)") %
(str(directed), str((c1, init1)),
str((c2, init2)), p, D, D_c),
file=output)
output.flush()
for cum in [True, False]:
figure(figsize=(10 * 4/3, 10))
bins = None
for c in cs:
for init in inits:
hist = hists[(c,init)]
if cum:
h = histogram(list(zip(*hist))[2], 1000000,
density=True)
y = numpy.cumsum(h[0])
y /= y[-1]
plot(h[-1][:-1], y, "-", label="c=%s, init=%s" % (str(c), init))
if c != numpy.inf:
hist = hists[(numpy.inf, 1)]
h2 = histogram(list(zip(*hist))[2], bins=h[-1],
density=True)
y2 = numpy.cumsum(h2[0])
y2 /= y2[-1]
res = abs(y - y2)
i = res.argmax()
axvline(h[-1][i], color="grey")
else:
h = histogram(list(zip(*hist))[2], 40, density=True)
plot(h[-1][:-1], h[0], "s-", label="c=%s, init=%s" % (str(c), init))
gca().set_yscale("log")
if cum:
legend(loc="lower right")
else:
legend(loc="best")
savefig("test_mcmc_nested%s_directed%s-cum%s.pdf" % (nested, directed, str(cum)))
print("OK")
\ No newline at end of file
g_small = g
for wait in [5000, 25000, 50000, 100000]:
for g, name in [(g_small, "small"), (collection.data["lesmis"], "lesmis")]:
for directed in [False, True]:
g.set_directed(directed)
for nested in [False, True]:
params = dict(name=name, directed=directed, nested=nested,
wait=wait)
with get_lock("mcmc_test", params) as lock:
if lock is None:
continue
cs = list(reversed([numpy.inf, 0.01, "gibbs", -numpy.inf, -0.1]))
inits = [1, "N"]
try:
hists = get_cache("hists", params)
except KeyError:
if nested:
state = minimize_nested_blockmodel_dl(g, deg_corr=True)
else:
state = minimize_blockmodel_dl(g, deg_corr=True)
hists = {}
for init in inits:
if nested:
bs = [g.get_vertices()] if init == "N" else [zeros(1)]
state = state.copy(bs=bs + [zeros(1)] * 6, sampling=True)
else:
state = state.copy(B=g.num_vertices() if init == "N" else 1)
step = 500
for i, c in enumerate(cs):
if c != "gibbs":
mcmc_args=dict(beta=.3, c=abs(c), niter=step)
else:
mcmc_args=dict(beta=.3, niter=step)
if name == "lesmis":
mcmc_args["beta"] = .9
if nested:
get_B = lambda s: [s.levels[0].get_Be()]
else:
get_B = lambda s: [s.get_Be()]
hists[(c, init)] = mcmc_equilibrate(state,
mcmc_args=mcmc_args,
gibbs=c=="gibbs",
multiflip = c != "gibbs" and c < 0,
force_niter=wait,
callback=get_B,
verbose=(1, "c = %s " % str(c)) if verbose else False,
history=True)
put_cache("hists", params, hists)
output = open(f"test_mcmc/test_mcmc_{name}_nested{nested}_directed{directed}-wait{wait}-output", "w")
for c1 in cs:
for init1 in inits:
for c2 in cs:
for init2 in inits:
try:
if (c2, init2) < (c1, init1):
continue
except TypeError:
pass
Ss1 = array(list(zip(*hists[(c1, init1)]))[2])
Ss2 = array(list(zip(*hists[(c2, init2)]))[2])
# add very small normal noise, to solve discreteness issue
Ss1 += numpy.random.normal(0, 1e-2, len(Ss1))
Ss2 += numpy.random.normal(0, 1e-2, len(Ss2))
D, p = scipy.stats.ks_2samp(Ss1, Ss2)
D_c = 1.63 * sqrt((len(Ss1) + len(Ss2)) / (len(Ss1) * len(Ss2)))
print("nested:", nested, "directed:", directed, "c1:", c1,
"init1:", init1, "c2:", c2, "init2:", init2, "D:",
D, "D_c:", D_c, "p-value:", p,
file=output)
if p < .001:
print(("Warning, distributions for directed=%s (c1, c2) = " +
"(%s, %s) are not the same, with a p-value: %g (D=%g, D_c=%g)") %
(str(directed), str((c1, init1)),
str((c2, init2)), p, D, D_c),
file=output)
output.flush()
for cum in [True, False]:
for Be in [True, False]:
figure(figsize=(10 * 4/3, 10))
hc = []
for c in cs:
for init in inits:
hist_i = hists[(c,init)]
vals = list(zip(*hist_i))
if Be:
hc += vals[-1]
else:
hc += vals[2]
bins = linspace(min(hc), max(hc), 40)
for c in cs:
for init in inits:
hist_i = hists[(c,init)]
vals = list(zip(*hist_i))
if Be:
vals = vals[-1]
else:
vals = vals[2]
if cum:
h = histogram(vals, 1000000, density=True)
y = numpy.cumsum(h[0])
y /= y[-1]
plot(h[-1][:-1], y, "-", lw=.5,
label="c=%s, init=%s" % (str(c), init))
if c != numpy.inf:
hist_i = hists[(numpy.inf, 1)]
h2 = histogram(list(zip(*hist_i))[2], bins=h[-1],
density=True)
y2 = numpy.cumsum(h2[0])
y2 /= y2[-1]
res = abs(y - y2)
i = res.argmax()
axvline(h[-1][i], color="grey")
else:
hist(vals, bins=bins, density=True, #log=True,
histtype="step", label="c=%s, init=%s" % (str(c), init))
if cum:
legend(loc="lower right")
else:
legend(loc="best")
savefig(f"test_mcmc/test_mcmc_{name}_nested{nested}_directed{directed}-Be{Be}-cum{cum}-wait{wait}.pdf")
if not cum:
figure(figsize=(10 * 4/3, 10))
vals = []
for c in cs:
for init in inits:
hist_i = hists[(c,init)]
vals_i = list(zip(*hist_i))
if Be:
vals += vals_i[-1]
else:
vals += vals_i[2]
x = linspace(min(vals), max(vals), 1000)
for c in cs:
for init in inits:
hist_i = hists[(c,init)]
vals = list(zip(*hist_i))
if Be:
vals = vals[-1]
else:
vals = vals[2]
kernel = scipy.stats.gaussian_kde(vals)
y = kernel(x)
plot(x, y, "-", linewidth=.4,
label="c=%s, init=%s" % (str(c), init))
legend(loc="best")
savefig(f"test_mcmc/test_mcmc_{name}_nested{nested}_directed{directed}-Be{Be}-kde-wait{wait}.pdf")
figure(figsize=(10 * 4/3, 10))
ymean = zeros(len(x))
count = 0
for c in cs:
for init in inits:
hist_i = hists[(c,init)]
vals = list(zip(*hist_i))
if Be:
vals = vals[-1]
else:
vals = vals[2]
kernel = scipy.stats.gaussian_kde(vals)
ymean += kernel(x)
count += 1
ymean /= count
for c in cs:
for init in inits:
hist_i = hists[(c,init)]
vals = list(zip(*hist_i))
if Be:
vals = vals[-1]
else:
vals = vals[2]
kernel = scipy.stats.gaussian_kde(vals)
y = kernel(x) - ymean
plot(x, y, "-", linewidth=1.5,
label="c=%s, init=%s" % (str(c), init))
legend(loc="best")
savefig(f"test_mcmc/test_mcmc_{name}_nested{nested}_directed{directed}-Be{Be}-kde-res-wait{wait}.pdf")
print("OK")
\ No newline at end of file
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment