def conditionedExpectedSufficientStats( self, ys, u, alphas, betas, forMStep=False ):
Ext1_xt1 = np.zeros( ( self.D_latent, self.D_latent ) )
Ext1_xt = np.zeros( ( self.D_latent, self.D_latent ) )
Ext_xt = np.zeros( ( self.D_latent, self.D_latent ) )
if( forMStep ):
Eut_ut = np.zeros( ( self.D_latent, self.D_latent ) )
Ext_ut = np.zeros( ( self.D_latent, self.D_latent ) )
Ext1_ut = np.zeros( ( self.D_latent, self.D_latent ) )
allT = 0
allM = 0
Eyt_yt = np.zeros( ( self.D_obs, self.D_obs ) )
Eyt_xt = np.zeros( ( self.D_obs, self.D_latent ) )
Ext_xt_y = np.zeros( ( self.D_latent, self.D_latent ) )
Ex0_x0 = np.zeros( ( self.D_latent, self.D_latent ) )
Ex0 = np.zeros( self.D_latent )
if( u is not None and u.ndim == 3 ):
# Multiple u's
assert len( ys ) == len( u )
it = zip( ys, alphas, betas, u )
else:
assert u is None or u.ndim == 2
if( u is None ):
J, _, _ = alphas[ 0 ][ 0 ]
u = np.zeros( ( len( alphas[ 0 ] ), J.shape[ 0 ] ) )
it = zip( ys, alphas, betas, itertools.repeat( u, len( ys ) ) )
for i, ( _ys, _alphas, _betas, _u ) in enumerate( it ):
uMask = np.isnan( _u )
_u = MaskedData( _u, uMask, None )
M, T, _ = _ys.shape
if( forMStep ):
allT += T - 1
allM += T * M
for t in range( 1, T ):
_Ext1_xt1, _Ext1_xt, _Ext_xt = self.expectedTransitionStatsBlock( t - 1, _alphas, _betas, ys=_ys, u=_u )
Ext1_xt1 += _Ext1_xt1
Ext1_xt += _Ext1_xt
Ext_xt += _Ext_xt
if( forMStep ):
J, h, _ = np.add( _alphas[ t - 1 ], _betas[ t - 1 ] )
J1, h1, _ = np.add( _alphas[ t ], _betas[ t ] )
Ext = Normal.natToStandard( J, h, fromPrecision=True )[ 0 ]
Ex1t = Normal.natToStandard( J1, h1, fromPrecision=True )[ 0 ]
Eut_ut += np.outer( _u[ t - 1 ], _u[ t - 1 ] )
Ext_ut += np.outer( Ext, _u[ t - 1 ] )
Ext1_ut += np.outer( Ex1t, _u[ t - 1 ] )
for t in range( T ):
_Eyt_yt, _Eyt_xt, _Ext_xt = self.expectedEmissionStats( t, _ys, _alphas, _betas, conditionOnY=True )
Eyt_yt += _Eyt_yt
Eyt_xt += _Eyt_xt
Ext_xt_y += _Ext_xt
_Ex0_x0, _Ex0 = self.expectedInitialStats( _alphas, _betas )
Ex0_x0 += _Ex0_x0
Ex0 += _Ex0
if( forMStep ):
return Ext1_xt1, Ext1_xt, Ext_xt, Eyt_yt, Eyt_xt, Ext_xt_y, Ex0_x0, Ex0, Eut_ut, Ext_ut, Ext1_ut, allT, allM
return Ext1_xt1, Ext1_xt, Ext_xt, Eyt_yt, Eyt_xt, Ext_xt_y, Ex0_x0, Ex0
def sampleStep( self, J, h ):
return Normal.unpackSingleSample( Normal.sample( params=Normal.natToStandard( J, h, fromPrecision=True ) ) )
def likelihoodStep( self, x, J, h ):
return Normal.log_likelihood( x, params=Normal.natToStandard( J, h, fromPrecision=True ) )
def natToStandard( cls, n1, n2, n3, n4, n5, n6, n7, n8 ):
A, sigma = Regression.natToStandard( n1, n2, n3 )
C, R = Regression.natToStandard( n4, n5, n6 )
mu0, sigma0 = Normal.natToStandard( n7, n8 )
return A, sigma, C, R, mu0, sigma0