def fit(self, train, test, n_states, train_lag_time, repetition, args, outfile):
n_features = train[0].shape[1]
kwargs = dict(n_states=n_states, n_features=n_features,
n_init=args.n_init,
n_em_iter=args.n_em_iter, n_lqa_iter=args.n_lqa_iter,
fusion_prior=args.fusion_prior, thresh=args.thresh,
reversible_type=args.reversible_type, platform=args.platform)
model = GaussianFusionHMM(**kwargs)
start = time.time()
model.fit(train)
end = time.time()
result = {
'model': 'GaussianFusionHMM',
'timescales': (np.real(model.timescales_) * train_lag_time).tolist(),
'transmat': np.real(model.transmat_).tolist(),
'populations': np.real(model.populations_).tolist(),
'n_states': model.n_states,
'split': args.split,
'fusion_prior': args.fusion_prior,
'train_lag_time': train_lag_time,
'train_time': end - start,
'means': np.real(model.means_).tolist(),
'vars': np.real(model.vars_).tolist(),
'train_logprob': model.fit_logprob_[-1],
'n_train_observations': sum(len(t) for t in train),
'n_test_observations': sum(len(t) for t in test),
'train_logprobs': model.fit_logprob_,
#'test_lag_time': args.test_lag_time,
'cross_validation_fold': 0,
'cross_validation_nfolds': 1,
'repetition': repetition,
}
# model.transmat_ = contraction(model.transmat_, float(train_lag_time) / float(args.test_lag_time))
# Don't do any contraction -- train and test at the same lagtime
result['test_logprob'] = model.score(test)
result['test_lag_time'] = train_lag_time
if not np.all(np.isfinite(model.transmat_)):
print('Nonfinite numbers in transmat !!')
json.dump(result, outfile)
outfile.write('\n')
def test_plusmin():
# Set constants
n_hotstart = 3
n_em_iter = 3
n_experiments = 1
n_seq = 1
T = 2000
gamma = 512.
# Generate data
plusmin = PlusminModel()
data, hidden = plusmin.generate_dataset(n_seq, T)
n_features = plusmin.x_dim
n_components = plusmin.K
# Train MSLDS
mslds_scores = []
l = MetastableSwitchingLDS(n_components, n_features,
n_hotstart=n_hotstart, n_em_iter=n_em_iter,
n_experiments=n_experiments)
l.fit(data, gamma=gamma)
mslds_score = l.score(data)
print("gamma = %f" % gamma)
print("MSLDS Log-Likelihood = %f" % mslds_score)
print()
# Fit Gaussian HMM for comparison
g = GaussianFusionHMM(plusmin.K, plusmin.x_dim)
g.fit(data)
hmm_score = g.score(data)
print("HMM Log-Likelihood = %f" % hmm_score)
print()
# Plot sample from MSLDS
sim_xs, sim_Ss = l.sample(T, init_state=0, init_obs=plusmin.mus[0])
sim_xs = np.reshape(sim_xs, (n_seq, T, plusmin.x_dim))
plt.close('all')
plt.figure(1)
plt.plot(range(T), data[0], label="Observations")
plt.plot(range(T), sim_xs[0], label='Sampled Observations')
plt.legend()
plt.show()
def test_doublewell():
import pdb, traceback, sys
try:
n_components = 2
n_features = 1
n_em_iter = 1
n_experiments = 1
tol=1e-1
data = load_doublewell(random_state=0)['trajectories']
T = len(data[0])
# Fit MSLDS model
model = MetastableSwitchingLDS(n_components, n_features,
n_experiments=n_experiments, n_em_iter=n_em_iter)
model.fit(data, gamma=.1, tol=tol)
mslds_score = model.score(data)
print("MSLDS Log-Likelihood = %f" % mslds_score)
# Fit Gaussian HMM for comparison
g = GaussianFusionHMM(n_components, n_features)
g.fit(data)
hmm_score = g.score(data)
print("HMM Log-Likelihood = %f" % hmm_score)
print()
# Plot sample from MSLDS
sim_xs, sim_Ss = model.sample(T, init_state=0)
plt.close('all')
plt.figure(1)
plt.plot(range(T), data[0], label="Observations")
plt.plot(range(T), sim_xs, label='Sampled Observations')
plt.legend()
plt.show()
except:
type, value, tb = sys.exc_info()
traceback.print_exc()
pdb.post_mortem(tb)
def test_viterbi():
data = [np.random.randn(1000, 3) + np.tile(np.sin(np.arange(1000)/100.0), (3,1)).T]
model1 = GaussianFusionHMM(n_states=2, n_features=3, platform='sklearn').fit(data)
model2 = GaussianFusionHMM(n_states=2, n_features=3, platform='cpu').fit(data)
model2.means_ = model1.means_
model2.vars_ = model1.vars_
model2.transmat_ = model1.transmat_
model2.populations_ = model1.populations_
logprob1, seq1 = model1.predict(data)
logprob2, seq2 = model2.predict(data)
np.testing.assert_almost_equal(logprob1, logprob2, decimal=3)
np.testing.assert_array_equal(seq1[0], seq2[0])
if PLOT:
import matplotlib.pyplot as pp
pp.plot(data[0][:, 0], label='data')
pp.plot(seq1[0], lw='5', label='viterbi')
pp.legend()
pp.show()
def fit(self, train, test, n_states, train_lag_time, fold, args, outfile):
kwargs = dict(n_states=n_states, n_features=self.n_features, n_em_iter=args.n_em_iter,
n_lqa_iter = args.n_lqa_iter, fusion_prior=args.fusion_prior,
thresh=args.thresh, reversible_type=args.reversible_type,
platform=args.platform)
print(kwargs)
model = GaussianFusionHMM(**kwargs)
start = time.time()
model.fit(train)
end = time.time()
result = {
'model': 'GaussianFusionHMM',
'timescales': (np.real(model.timescales_()) * train_lag_time).tolist(),
'transmat': np.real(model.transmat_).tolist(),
'populations': np.real(model.populations_).tolist(),
'n_states': model.n_states,
'split': args.split,
'fusion_prior': args.fusion_prior,
'train_lag_time': train_lag_time,
'train_time': end - start,
'means': np.real(model.means_).tolist(),
'vars': np.real(model.vars_).tolist(),
'train_logprob': model.fit_logprob_[-1],
'n_train_observations': sum(len(t) for t in train),
'n_test_observations': sum(len(t) for t in test),
'train_logprobs': model.fit_logprob_,
#'test_lag_time': args.test_lag_time,
'cross_validation_fold': fold,
'cross_validation_nfolds': args.n_cv,
}
# model.transmat_ = contraction(model.transmat_, float(train_lag_time) / float(args.test_lag_time))
# Don't do any contraction -- train and test at the same lagtime
result['test_logprob'] = model.score(test)
result['test_lag_time'] = train_lag_time
if not np.all(np.isfinite(model.transmat_)):
print('Nonfinite numbers in transmat !!')
json.dump(result, outfile)
outfile.write('\n')
def test_muller_potential():
import pdb, traceback, sys
try:
# Set constants
n_hotstart = 3
n_em_iter = 3
n_experiments = 1
n_seq = 1
num_trajs = 1
T = 2500
sim_T = 2500
gamma = 200.
# Generate data
warnings.filterwarnings("ignore", category=DeprecationWarning)
muller = MullerModel()
data, trajectory, start = \
muller.generate_dataset(n_seq, num_trajs, T)
n_features = muller.x_dim
n_components = muller.K
# Train MSLDS
model = MetastableSwitchingLDS(n_components, n_features,
n_hotstart=n_hotstart, n_em_iter=n_em_iter,
n_experiments=n_experiments)
model.fit(data, gamma=gamma)
mslds_score = model.score(data)
print("MSLDS Log-Likelihood = %f" % mslds_score)
# Fit Gaussian HMM for comparison
g = GaussianFusionHMM(n_components, n_features)
g.fit(data)
hmm_score = g.score(data)
print("HMM Log-Likelihood = %f" % hmm_score)
# Clear Display
plt.cla()
plt.plot(trajectory[start:, 0], trajectory[start:, 1], color='k')
plt.scatter(model.means_[:, 0], model.means_[:, 1],
color='r', zorder=10)
plt.scatter(data[0][:, 0], data[0][:, 1],
edgecolor='none', facecolor='k', zorder=1)
Delta = 0.5
minx = min(data[0][:, 0])
maxx = max(data[0][:, 0])
miny = min(data[0][:, 1])
maxy = max(data[0][:, 1])
sim_xs, sim_Ss = model.sample(sim_T, init_state=0,
init_obs=model.means_[0])
minx = min(min(sim_xs[:, 0]), minx) - Delta
maxx = max(max(sim_xs[:, 0]), maxx) + Delta
miny = min(min(sim_xs[:, 1]), miny) - Delta
maxy = max(max(sim_xs[:, 1]), maxy) + Delta
plt.scatter(sim_xs[:, 0], sim_xs[:, 1], edgecolor='none',
zorder=5, facecolor='g')
plt.plot(sim_xs[:, 0], sim_xs[:, 1], zorder=5, color='g')
MullerForce.plot(ax=plt.gca(), minx=minx, maxx=maxx,
miny=miny, maxy=maxy)
plt.show()
except:
type, value, tb = sys.exc_info()
traceback.print_exc()
pdb.post_mortem(tb)
def test_alanine_dipeptide():
import pdb, traceback, sys
warnings.filterwarnings("ignore",
category=DeprecationWarning)
try:
b = fetch_alanine_dipeptide()
trajs = b.trajectories
n_seq = len(trajs)
n_frames = trajs[0].n_frames
n_atoms = trajs[0].n_atoms
n_features = n_atoms * 3
sim_T = 1000
data_home = get_data_home()
data_dir = join(data_home, TARGET_DIRECTORY_ALANINE)
top = md.load(join(data_dir, 'ala2.pdb'))
n_components = 2
# Superpose m
data = []
for traj in trajs:
traj.superpose(top)
Z = traj.xyz
Z = np.reshape(Z, (len(Z), n_features), order='F')
data.append(Z)
# Fit MSLDS model
n_experiments = 1
n_em_iter = 1
tol = 1e-1
model = MetastableSwitchingLDS(n_components,
n_features, n_experiments=n_experiments,
n_em_iter=n_em_iter)
model.fit(data, gamma=.1, tol=tol, verbose=True)
mslds_score = model.score(data)
print("MSLDS Log-Likelihood = %f" % mslds_score)
# Fit Gaussian HMM for comparison
g = GaussianFusionHMM(n_components, n_features)
g.fit(data)
hmm_score = g.score(data)
print("HMM Log-Likelihood = %f" % hmm_score)
print()
# Generate a trajectory from learned model.
sample_traj, hidden_states = model.sample(sim_T)
states = []
for k in range(n_components):
states.append([])
# Presort the data into the metastable wells
for k in range(n_components):
for i in range(len(trajs)):
traj = trajs[i]
Z = traj.xyz
Z = np.reshape(Z, (len(Z), n_features), order='F')
logprob = log_multivariate_normal_density(Z,
np.array(model.means_),
np.array(model.covars_), covariance_type='full')
assignments = np.argmax(logprob, axis=1)
#probs = np.max(logprob, axis=1)
# pick structures that have highest log probability in state
s = traj[assignments == k]
states[k].append(s)
# Pick frame from original trajectories closest to current sample
gen_traj = None
for t in range(sim_T):
h = hidden_states[t]
best_logprob = -np.inf
best_frame = None
for i in range(len(trajs)):
if t > 0:
states[h][i].superpose(gen_traj, t-1)
Z = states[h][i].xyz
Z = np.reshape(Z, (len(Z), n_features), order='F')
mean = sample_traj[t]
logprobs = log_multivariate_normal_density(Z,
mean, model.Qs_[h], covariance_type='full')
ind = np.argmax(logprobs, axis=0)
logprob = logprobs[ind]
if logprob > best_log_prob:
logprob = best_logprob
best_frame = states[h][i][ind]
if t == 0:
gen_traj = best_frame
else:
gen_traj = gen_traj.join(frame)
gen_traj.save('%s.xtc' % self.out)
gen_traj[0].save('%s.xtc.pdb' % self.out)
except:
type, value, tb = sys.exc_info()
traceback.print_exc()
pdb.post_mortem(tb)
