def test_coin_sample_post(self):
"""Test sampling from posterior distribution"""
outdir = 'test/tmp/test_hmm/test_coin_sample_post/'
make_clean_dir(outdir)
model = make_coin_model()
# sample states and data
ndata = 100
states = list(islice(hmm.sample_hmm_states(model), ndata))
data = list(hmm.sample_hmm_data(model, states))
model.prob_emission = (
lambda pos, state: model.prob_emission_data(state, data[pos]))
p = Gnuplot()
p.enableOutput(False)
p.plot(states, style="lines")
probs = hmm.get_posterior_probs(model, len(data))
states2 = [exp(probs[i][1]) for i in xrange(len(data))]
p.plot(util.vadds(states2, 1.5), style="lines", miny=-1, maxy=12)
for i in range(2, 10):
states2 = hmm.sample_posterior(model, ndata)
self.assertTrue(stats.corr(states, states2) > .5)
p.plot(util.vadds(states2, 1.5 * i),
style="lines",
miny=-1,
maxy=12)
p.enableOutput(True)
p.save(outdir + 'plot.png')
def test_coin_sample_post(self):
"""Test sampling from posterior distribution"""
outdir = 'test/tmp/test_hmm/test_coin_sample_post/'
make_clean_dir(outdir)
model = make_coin_model()
# sample states and data
ndata = 100
states = list(islice(hmm.sample_hmm_states(model), ndata))
data = list(hmm.sample_hmm_data(model, states))
model.prob_emission = (lambda pos, state:
model.prob_emission_data(state, data[pos]))
p = Gnuplot()
p.enableOutput(False)
p.plot(states, style="lines")
probs = hmm.get_posterior_probs(model, len(data))
states2 = [exp(probs[i][1]) for i in xrange(len(data))]
p.plot(util.vadds(states2, 1.5), style="lines", miny=-1, maxy=12)
for i in range(2, 10):
states2 = hmm.sample_posterior(model, ndata)
self.assertTrue(stats.corr(states, states2) > .5)
p.plot(util.vadds(states2, 1.5*i), style="lines", miny=-1, maxy=12)
p.enableOutput(True)
p.save(outdir + 'plot.png')
def test_norecomb_plot(self):
k = 50
n = 1e4
rho = 1.5e-8 * .0001
rho2 = 1.5e-8 * 10
mu = 2.5e-8 * 100
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(ntimes=20)
arghmm.discretize_arg(arg, times)
tree = arg.get_marginal_tree(0)
treelib.draw_tree_names(tree.get_tree(), minlen=5, scale=4e-4)
# get thread
new_name = "n%d" % (k - 1)
keep = ["n%d" % i for i in range(k - 1)]
arglib.subarg_by_leaf_names(arg, keep)
arg.set_ancestral()
arg.prune()
model = arghmm.ArgHmm(arg,
seqs,
new_name=new_name,
times=times,
rho=rho2,
mu=mu)
print "states", len(model.states[0])
print "muts", len(muts)
# simulate a new thread
states = list(islice(hmm.sample_hmm_states(model), 0, arg.end))
data = list(hmm.sample_hmm_data(model, states))
seqs[new_name] = "".join(data)
#alignlib.print_align(seqs)
thread = [
model.times[model.states[i][s][1]] for i, s in enumerate(states)
]
p = plot(thread, style="lines")
probs = arghmm.get_posterior_probs(model, length, verbose=True)
print "done"
high = list(arghmm.iter_posterior_times(model, probs, .75))
low = list(arghmm.iter_posterior_times(model, probs, .25))
p.plot(high, style="lines")
p.plot(low, style="lines")
pause()
def test_norecomb_plot(self):
k = 50
n = 1e4
rho = 1.5e-8 * .0001
rho2 = 1.5e-8 * 10
mu = 2.5e-8 * 100
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(ntimes=20)
arghmm.discretize_arg(arg, times)
tree = arg.get_marginal_tree(0)
treelib.draw_tree_names(tree.get_tree(), minlen=5, scale=4e-4)
# get thread
new_name = "n%d" % (k-1)
keep = ["n%d" % i for i in range(k-1)]
arglib.subarg_by_leaf_names(arg, keep)
arg.set_ancestral()
arg.prune()
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times,
rho=rho2, mu=mu)
print "states", len(model.states[0])
print "muts", len(muts)
# simulate a new thread
states = list(islice(hmm.sample_hmm_states(model), 0, arg.end))
data = list(hmm.sample_hmm_data(model, states))
seqs[new_name] = "".join(data)
#alignlib.print_align(seqs)
thread = [model.times[model.states[i][s][1]]
for i, s in enumerate(states)]
p = plot(thread, style="lines")
probs = arghmm.get_posterior_probs(model, length, verbose=True)
print "done"
high = list(arghmm.iter_posterior_times(model, probs, .75))
low = list(arghmm.iter_posterior_times(model, probs, .25))
p.plot(high, style="lines")
p.plot(low, style="lines")
pause()
def test_coin_post(self):
"""Test that posterior decoding."""
model = make_coin_model()
# sample states and data
ndata = 100
states = list(islice(hmm.sample_hmm_states(model), ndata))
data = list(hmm.sample_hmm_data(model, states))
model.prob_emission = (
lambda pos, state: model.prob_emission_data(state, data[pos]))
probs = hmm.get_posterior_probs(model, len(data))
for col in probs:
p = sum(map(exp, col))
self.assertAlmostEqual(p, 1.0)
def test_coin_post(self):
"""Test that posterior decoding."""
model = make_coin_model()
# sample states and data
ndata = 100
states = list(islice(hmm.sample_hmm_states(model), ndata))
data = list(hmm.sample_hmm_data(model, states))
model.prob_emission = (lambda pos, state:
model.prob_emission_data(state, data[pos]))
probs = hmm.get_posterior_probs(model, len(data))
for col in probs:
p = sum(map(exp, col))
self.assertAlmostEqual(p, 1.0)
def test_coin(self):
"""Test that viterbi and posterior coding work well."""
outdir = 'test/tmp/test_hmm/test_coin/'
make_clean_dir(outdir)
model = make_coin_model()
# sample states
ndata = 100
states = list(islice(hmm.sample_hmm_states(model), ndata))
p = Gnuplot()
p.enableOutput(False)
p.plot(states, style="lines")
# sample data
data = list(hmm.sample_hmm_data(model, states))
# viterbi
model.prob_emission = (
lambda pos, state: model.prob_emission_data(state, data[pos]))
states2 = hmm.viterbi(model, len(data))
# posterior
probs = hmm.get_posterior_probs(model, len(data))
states3 = [exp(probs[i][1]) for i in xrange(len(data))]
# assert that inferences correlates with true state
self.assertTrue(stats.corr(states, states2) > .5)
self.assertTrue(stats.corr(states, states3) > .5)
# plot inference
p.plot(util.vadds(states2, 1.5), style="lines", miny=-1, maxy=4)
p.plot(util.vadds(states3, 2.5), style="lines", miny=-1, maxy=4)
p.enableOutput(True)
p.save(outdir + 'plot.png')
def test_coin(self):
"""Test that viterbi and posterior coding work well."""
outdir = 'test/tmp/test_hmm/test_coin/'
make_clean_dir(outdir)
model = make_coin_model()
# sample states
ndata = 100
states = list(islice(hmm.sample_hmm_states(model), ndata))
p = Gnuplot()
p.enableOutput(False)
p.plot(states, style="lines")
# sample data
data = list(hmm.sample_hmm_data(model, states))
# viterbi
model.prob_emission = (lambda pos, state:
model.prob_emission_data(state, data[pos]))
states2 = hmm.viterbi(model, len(data))
# posterior
probs = hmm.get_posterior_probs(model, len(data))
states3 = [exp(probs[i][1]) for i in xrange(len(data))]
# assert that inferences correlates with true state
self.assertTrue(stats.corr(states, states2) > .5)
self.assertTrue(stats.corr(states, states3) > .5)
# plot inference
p.plot(util.vadds(states2, 1.5), style="lines", miny=-1, maxy=4)
p.plot(util.vadds(states3, 2.5), style="lines", miny=-1, maxy=4)
p.enableOutput(True)
p.save(outdir + 'plot.png')
