def lbfgs_wz(model, w, z, x, hook=None, maxiter=None):
def f(y):
_w, _z = ndict.unflatten_multiple(y, [w, z])
logpx, logpz = model.logpxz(_w, x, _z)
return - (logpx.sum() + logpz.sum())
def fprime(y):
_w, _z = ndict.unflatten_multiple(y, [w, z])
logpx, logpz, gw, gz = model.dlogpxz_dwz(_w, x, _z)
gwz = ndict.flatten_multiple((gw, gz))
return - gwz
t = [0, 0, time.time()]
def callback(wz):
if hook is None: return
_w, _z = ndict.unflatten_multiple(wz, (w, z))
t[1] += 1
hook(t[1], _w, _z)
x0 = ndict.flatten_multiple((w, z))
xn, f, d = scipy.optimize.fmin_l_bfgs_b(func=f, x0=x0, fprime=fprime, m=100, iprint=0, callback=callback, maxiter=maxiter)
#scipy.optimize.fmin_cg(f=f, x0=x0, fprime=fprime, full_output=True, callback=hook)
#scipy.optimize.fmin_ncg(f=f, x0=x0, fprime=fprime, full_output=True, callback=hook)
w, z = ndict.unflatten_multiple(xn, (w, z))
if d['warnflag'] is 2:
print 'warnflag:', d['warnflag']
print d['task']
return w, z
def lbfgs_wz(model, w, z, x, hook=None, maxiter=None):
def f(y):
_w, _z = ndict.unflatten_multiple(y, [w, z])
logpx, logpz = model.logpxz(_w, x, _z)
return -(logpx.sum() + logpz.sum())
def fprime(y):
_w, _z = ndict.unflatten_multiple(y, [w, z])
logpx, logpz, gw, gz = model.dlogpxz_dwz(_w, x, _z)
gwz = ndict.flatten_multiple((gw, gz))
return -gwz
t = [0, 0, time.time()]
def callback(wz):
if hook is None: return
_w, _z = ndict.unflatten_multiple(wz, (w, z))
t[1] += 1
hook(t[1], _w, _z)
x0 = ndict.flatten_multiple((w, z))
xn, f, d = scipy.optimize.fmin_l_bfgs_b(func=f,
x0=x0,
fprime=fprime,
m=100,
iprint=0,
callback=callback,
maxiter=maxiter)
#scipy.optimize.fmin_cg(f=f, x0=x0, fprime=fprime, full_output=True, callback=hook)
#scipy.optimize.fmin_ncg(f=f, x0=x0, fprime=fprime, full_output=True, callback=hook)
w, z = ndict.unflatten_multiple(xn, (w, z))
if d['warnflag'] is 2:
print 'warnflag:', d['warnflag']
print d['task']
return w, z
def fprime(y):
_w, _z = ndict.unflatten_multiple(y, [w, z])
logpx, logpz, gw, gz = model.dlogpxz_dwz(_w, x, _z)
gwz = ndict.flatten_multiple((gw, gz))
return -gwz
def fprime(y):
_w, _z = ndict.unflatten_multiple(y, [w, z])
logpx, logpz, gw, gz = model.dlogpxz_dwz(_w, x, _z)
gwz = ndict.flatten_multiple((gw, gz))
return - gwz