def mstep(self, gamma, x, u, weights=None, use_prior=False):
aux = []
if weights:
for _w, _gamma in zip(weights, gamma):
aux.append(_w[:, None] * _gamma)
gamma = aux
xs, ys, ws = [], [], []
for _x, _u, _w in zip(x, u, gamma):
_feat = self.featurize(_x)
xs.append(np.hstack((_feat, np.ones((_feat.shape[0], 1)))))
ys.append(_u)
ws.append(_w)
_cov = np.zeros((self.nb_states, self.dm_act, self.dm_act))
for k in range(self.nb_states):
coef_, sigma = linear_regression(Xs=np.vstack(xs), ys=np.vstack(ys),
weights=np.vstack(ws)[:, k], fit_intercept=False,
**self.prior if use_prior else {})
self.K[k, ...] = coef_[:, :self.dm_feat]
self.kff[k, ...] = coef_[:, -1]
_cov[k, ...] = sigma
# usage = sum([_gamma.sum(0) for _gamma in gamma])
# unused = np.where(usage < 1)[0]
# used = np.where(usage > 1)[0]
# if len(unused) > 0:
# for k in unused:
# i = npr.choice(used)
# self.K[k] = self.K[i] + 0.01 * npr.randn(*self.K[i].shape)
# self.kff[k] = self.kff[i] + 0.01 * npr.randn(*self.kff[i].shape)
# _cov[k] = _cov[i]
self.cov = _cov
def initialize(self, x, u, **kwargs):
localize = kwargs.get('localize', True)
feat = [self.featurize(_x) for _x in x]
Ts = [_x.shape[0] for _x in x]
if localize:
from sklearn.cluster import KMeans
km = KMeans(self.nb_states)
km.fit(np.hstack((np.vstack(feat), np.vstack(u))))
zs = np.split(km.labels_, np.cumsum(Ts)[:-1])
else:
zs = [npr.choice(self.nb_states, size=T) for T in Ts]
_cov = np.zeros((self.nb_states, self.dm_act, self.dm_act))
for k in range(self.nb_states):
ts = [np.where(z == k)[0] for z in zs]
xs = [_feat[t, :] for t, _feat in zip(ts, feat)]
ys = [_u[t, :] for t, _u in zip(ts, u)]
coef_, intercept_, sigma = linear_regression(np.vstack(xs), np.vstack(ys),
weights=None, fit_intercept=True,
**self.prior)
self.K[k, ...] = coef_[:, :self.dm_feat]
self.kff[k, :] = intercept_
_cov[k, ...] = sigma
self.cov = _cov
def initialize(self, x, u, **kwargs):
localize = kwargs.get('localize', True)
Ts = [_x.shape[0] for _x in x]
if localize:
from sklearn.cluster import KMeans
km = KMeans(self.nb_states)
km.fit(np.hstack((np.vstack(x), np.vstack(u))))
zs = np.split(km.labels_, np.cumsum(Ts)[:-1])
zs = [z[:-1] for z in zs]
else:
zs = [npr.choice(self.nb_states, size=T - 1) for T in Ts]
_cov = np.zeros((self.nb_states, self.dm_obs, self.dm_obs))
for k in range(self.nb_states):
ts = [np.where(z == k)[0] for z in zs]
xs = [
np.hstack((_x[t, :], _u[t, :])) for t, _x, _u in zip(ts, x, u)
]
ys = [_x[t + 1, :] for t, _x in zip(ts, x)]
coef_, intercept_, sigma = linear_regression(np.vstack(xs),
np.vstack(ys),
weights=None,
fit_intercept=True,
**self.prior)
self.A[k, ...] = coef_[:, :self.dm_obs]
self.B[k, ...] = coef_[:, self.dm_obs:]
self.c[k, :] = intercept_
_cov[k, ...] = sigma
self.cov = _cov