def fit_arhmm(x, affine=True):
print("Fitting Sticky ARHMM")
dynamics_hypparams = \
dict(nu_0=D_latent + 2,
S_0=np.eye(D_latent),
M_0=np.hstack((np.eye(D_latent), np.zeros((D_latent, int(affine))))),
K_0=np.eye(D_latent + affine),
affine=affine)
dynamics_hypparams = get_empirical_ar_params([x], dynamics_hypparams)
dynamics_distns = [
AutoRegression(
A=np.column_stack((0.99 * np.eye(D_latent),
np.zeros((D_latent, int(affine))))),
sigma=np.eye(D_latent),
**dynamics_hypparams)
for _ in range(args.K)]
init_distn = Gaussian(nu_0=D_latent + 2,
sigma_0=np.eye(D_latent),
mu_0=np.zeros(D_latent),
kappa_0=1.0)
arhmm = ARWeakLimitStickyHDPHMM(
init_state_distn='uniform',
init_emission_distn=init_distn,
obs_distns=dynamics_distns,
alpha=3.0, kappa=10.0, gamma=3.0)
arhmm.add_data(x)
lps = []
for _ in tqdm(range(args.N_samples)):
arhmm.resample_model()
lps.append(arhmm.log_likelihood())
z_init = arhmm.states_list[0].stateseq
z_init = np.concatenate(([0], z_init))
return arhmm, z_init
def fit_ar_pyhsmm_models(
train_datas,
val_datas,
test_datas,
K=2,
N_iters=2000,
seed=1,
lags=1,
affine=True,
alpha=10,
gamma=10,
kappa=100,
init_state_distn="uniform",
observations="ar",
):
"""
Fit datasets for multiple values
"""
npr.seed(seed)
assert (len(train_datas) > 0) and (type(train_datas) is list)
assert (len(val_datas) > 0) and (type(val_datas) is list)
assert (len(test_datas) > 0) and (type(test_datas) is list)
# Standard AR model (Scale resampling)
D_obs = train_datas[0].shape[1]
def evalute_model(model):
# train log log_likelihood
train_ll = model.log_likelihood()
# validation log log_likelihood
val_pll = 0
for data in val_datas:
val_pll += model.log_likelihood(data)
# Test log log_likelihood
test_pll = 0
for data in test_datas:
test_pll += model.log_likelihood(data)
return train_ll, val_pll, test_pll
# Construct a standard AR-HMM
obs_hypers = dict(
nu_0=D_obs + 2,
S_0=np.eye(D_obs),
M_0=np.hstack(
(np.eye(D_obs), np.zeros((D_obs, D_obs * (lags - 1) + affine)))),
K_0=np.eye(D_obs * lags + affine),
affine=affine,
)
obs_hypers = get_empirical_ar_params(train_datas, obs_hypers)
obs_distns = [
ardistributions.AutoRegression(**obs_hypers) for _ in range(K)
]
# ----------------
# Init Model Param
# ----------------
model = armodels.ARWeakLimitStickyHDPHMM(
# sampled from 1d finite pmf
alpha=alpha,
gamma=gamma,
init_state_distn=init_state_distn,
# create A, Sigma
obs_distns=obs_distns,
kappa=kappa,
)
# ----------------
# Add datasets
# ----------------
for data in train_datas:
model.add_data(data)
# ---------------------
# Initialize the states
# ---------------------
model.resample_states()
# ------------------------------
# Initialize log log_likelihood
# ------------------------------
init_val = evalute_model(model)
# -----------------------
# Fit with Gibbs sampling
# -----------------------
def sample(model):
tic = time.time()
model.resample_model()
timestep = time.time() - tic
return evalute_model(model), timestep
# ----------------------
# Run for each iteration
# ----------------------
# values at each timestep
vals, timesteps = zip(*[sample(model) for _ in trange(N_iters)])
lls_train, lls_val, lls_test = \
zip(*((init_val,) + vals))
timestamps = np.cumsum((0., ) + timesteps)
# calculate the states after N_iters
z = [mm.stateseq for mm in model.states_list]
return model, lls_train, lls_val, lls_test, timestamps, z
def fit_ar_separate_trans_pyhsmm_models(
train_datas,
val_datas,
test_datas,
K=2,
N_iters=2000,
seed=1,
lags=1,
affine=True,
alpha=10,
gamma=10,
kappa=100,
init_state_distn="uniform",
observations="ar",
):
"""
Fit model using separate transition matrices per
element in dictionary.
"""
npr.seed(seed)
assert type(train_datas) is defaultdict
datas_all = []
for _, datalist in train_datas.items():
print(len(datalist))
datas_all.extend(datalist)
print("Running for {} data chunks".format(len(datas_all)))
# Standard AR model (Scale resampling)
D_obs = datas_all[0].shape[1]
def evalute_model(model):
# train log log_likelihood
ll = model.log_likelihood()
# validation log log_likelihood
val_pll = 0
for data_id, data in val_datas.items():
val_pll += model.log_likelihood(group_id=data_id, data=data)
# Test log log_likelihood
test_pll = 0
for data_id, data in test_datas.items():
test_pll += model.log_likelihood(group_id=data_id, data=data)
return ll, val_pll, test_pll
# Construct a standard AR-HMM
obs_hypers = dict(
nu_0=D_obs + 2,
S_0=np.eye(D_obs),
M_0=np.hstack(
(np.eye(D_obs), np.zeros((D_obs, D_obs * (lags - 1) + affine)))),
K_0=np.eye(D_obs * lags + affine),
affine=affine,
)
obs_hypers = get_empirical_ar_params(datas_all, obs_hypers)
obs_distns = [
ardistributions.AutoRegression(**obs_hypers) for _ in range(K)
]
# free space
del datas_all
# Init Model Param
model = armodels.ARWeakLimitStickyHDPHMMSeparateTrans(
# sampled from 1d finite pmf
alpha=alpha,
gamma=gamma,
init_state_distn=init_state_distn,
# create A, Sigma
obs_distns=obs_distns,
kappa=kappa,
)
# free space for very large datasets
del obs_distns
# --------------
# Add datasets
# --------------
for group_id, datalist in train_datas.items():
for data in datalist:
model.add_data(group_id=group_id, data=data)
# free space for very large datasets
del train_datas
# ---------------------
# Initialize the states
# ---------------------
model.resample_states()
# ------------------------------
# Initialize log log_likelihood
# ------------------------------
init_val = evalute_model(model)
# -----------------------
# Fit with Gibbs sampling
# -----------------------
def sample(model):
tic = time.time()
# resample model
model.resample_model()
timestep = time.time() - tic
return evalute_model(model), timestep
# ----------------------
# Run for each iteration
# ----------------------
# values at each timestep
vals, timesteps = zip(*[sample(model) for _ in trange(N_iters)])
lls_train, lls_val, lls_test = \
zip(*((init_val,) + vals))
timestamps = np.cumsum((0., ) + timesteps)
# calculate the states after N_iters
z = [mm.stateseq for mm in model.states_list]
return model, lls_train, lls_val, lls_test, timestamps, z