apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-02 16:22:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions
GPy/GPy/models/state_space_main.py

3489 lines
135 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# -*- coding: utf-8 -*-
# Copyright (c) 2015, Alex Grigorevskiy
# Licensed under the BSD 3-clause license (see LICENSE.txt)
"""
Main functionality for state-space inference.
"""
import collections # for cheking whether a variable is iterable
import types # for cheking whether a variable is a function
import numpy as np
import scipy as sp
import scipy.linalg as linalg
try:
from . import state_space_setup
setup_available = True
except ImportError as e:
setup_available = False
print_verbose = False
try:
import state_space_cython
cython_code_available = True
if print_verbose:
print("state_space: cython is available")
except ImportError as e:
cython_code_available = False
#cython_code_available = False
# Use cython by default
use_cython = False
if setup_available:
use_cython = state_space_setup.use_cython
if print_verbose:
if use_cython:
print("state_space: cython is used")
else:
print("state_space: cython is NOT used")
class Dynamic_Callables_Python(object):
def f_a(self, k, m, A):
"""
p_a: function (k, x_{k-1}, A_{k}). Dynamic function.
k (iteration number), starts at 0
x_{k-1} State from the previous step
A_{k} Jacobian matrices of f_a. In the linear case it is exactly
A_{k}.
"""
raise NotImplemented("f_a is not implemented!")
def Ak(self, k, m, P): # returns state iteration matrix
"""
function (k, m, P) return Jacobian of dynamic function, it is passed
into p_a.
k (iteration number), starts at 0
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
"""
raise NotImplemented("Ak is not implemented!")
def Qk(self, k):
"""
function (k). Returns noise matrix of dynamic model on iteration k.
k (iteration number). starts at 0
"""
raise NotImplemented("Qk is not implemented!")
def Q_srk(self, k):
"""
function (k). Returns the square root of noise matrix of dynamic model
on iteration k.
k (iteration number). starts at 0
This function is implemented to use SVD prediction step.
"""
raise NotImplemented("Q_srk is not implemented!")
def dAk(self, k):
"""
function (k). Returns the derivative of A on iteration k.
k (iteration number). starts at 0
"""
raise NotImplemented("dAk is not implemented!")
def dQk(self, k):
"""
function (k). Returns the derivative of Q on iteration k.
k (iteration number). starts at 0
"""
raise NotImplemented("dQk is not implemented!")
def reset(self, compute_derivatives=False):
"""
Return the state of this object to the beginning of iteration
(to k eq. 0).
"""
raise NotImplemented("reset is not implemented!")
if use_cython:
Dynamic_Callables_Class = state_space_cython.Dynamic_Callables_Cython
else:
Dynamic_Callables_Class = Dynamic_Callables_Python
class Measurement_Callables_Python(object):
def f_h(self, k, m_pred, Hk):
"""
function (k, x_{k}, H_{k}). Measurement function.
k (iteration number), starts at 0
x_{k} state
H_{k} Jacobian matrices of f_h. In the linear case it is exactly
H_{k}.
"""
raise NotImplemented("f_a is not implemented!")
def Hk(self, k, m_pred, P_pred): # returns state iteration matrix
"""
function (k, m, P) return Jacobian of measurement function, it is
passed into p_h.
k (iteration number), starts at 0
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
"""
raise NotImplemented("Hk is not implemented!")
def Rk(self, k):
"""
function (k). Returns noise matrix of measurement equation
on iteration k.
k (iteration number). starts at 0
"""
raise NotImplemented("Rk is not implemented!")
def R_isrk(self, k):
"""
function (k). Returns the square root of the noise matrix of
measurement equation on iteration k.
k (iteration number). starts at 0
This function is implemented to use SVD prediction step.
"""
raise NotImplemented("Q_srk is not implemented!")
def dHk(self, k):
"""
function (k). Returns the derivative of H on iteration k.
k (iteration number). starts at 0
"""
raise NotImplemented("dAk is not implemented!")
def dRk(self, k):
"""
function (k). Returns the derivative of R on iteration k.
k (iteration number). starts at 0
"""
raise NotImplemented("dQk is not implemented!")
def reset(self, compute_derivatives=False):
"""
Return the state of this object to the beginning of iteration
(to k eq. 0)
"""
raise NotImplemented("reset is not implemented!")
if use_cython:
Measurement_Callables_Class = state_space_cython.\
Measurement_Callables_Cython
else:
Measurement_Callables_Class = Measurement_Callables_Python
class R_handling_Python(Measurement_Callables_Class):
"""
The calss handles noise matrix R.
"""
def __init__(self, R, index, R_time_var_index, unique_R_number, dR=None):
"""
Input:
---------------
R - array with noise on various steps. The result of preprocessing
the noise input.
index - for each step of Kalman filter contains the corresponding index
in the array.
R_time_var_index - another index in the array R. Computed earlier and
is passed here.
unique_R_number - number of unique noise matrices below which square
roots are cached and above which they are computed each time.
dR: 3D array[:, :, param_num]
derivative of R. Derivative is supported only when R do not change
over time
Output:
--------------
Object which has two necessary functions:
f_R(k)
inv_R_square_root(k)
"""
self.R = R
self.index = np.asarray(index, np.int_)
self.R_time_var_index = int(R_time_var_index)
self.dR = dR
if (len(np.unique(index)) > unique_R_number):
self.svd_each_time = True
else:
self.svd_each_time = False
self.R_square_root = {}
def Rk(self, k):
return self.R[:, :, self.index[self.R_time_var_index, k]]
def dRk(self, k):
if self.dR is None:
raise ValueError("dR derivative is None")
return self.dR # the same dirivative on each iteration
def R_isrk(self, k):
"""
Function returns the inverse square root of R matrix on step k.
"""
ind = int(self.index[self.R_time_var_index, k])
R = self.R[:, :, ind]
if (R.shape[0] == 1): # 1-D case handle simplier. No storage
# of the result, just compute it each time.
inv_square_root = np.sqrt(1.0/R)
else:
if self.svd_each_time:
(U, S, Vh) = sp.linalg.svd(R, full_matrices=False,
compute_uv=True, overwrite_a=False,
check_finite=True)
inv_square_root = U * 1.0/np.sqrt(S)
else:
if ind in self.R_square_root:
inv_square_root = self.R_square_root[ind]
else:
(U, S, Vh) = sp.linalg.svd(R, full_matrices=False,
compute_uv=True,
overwrite_a=False,
check_finite=True)
inv_square_root = U * 1.0/np.sqrt(S)
self.R_square_root[ind] = inv_square_root
return inv_square_root
if use_cython:
R_handling_Class = state_space_cython.R_handling_Cython
else:
R_handling_Class = R_handling_Python
class Std_Measurement_Callables_Python(R_handling_Class):
def __init__(self, H, H_time_var_index, R, index, R_time_var_index,
unique_R_number, dH=None, dR=None):
super(Std_Measurement_Callables_Python,
self).__init__(R, index, R_time_var_index, unique_R_number, dR)
self.H = H
self.H_time_var_index = int(H_time_var_index)
self.dH = dH
def f_h(self, k, m, H):
"""
function (k, x_{k}, H_{k}). Measurement function.
k (iteration number), starts at 0
x_{k} state
H_{k} Jacobian matrices of f_h. In the linear case it is exactly
H_{k}.
"""
return np.dot(H, m)
def Hk(self, k, m_pred, P_pred): # returns state iteration matrix
"""
function (k, m, P) return Jacobian of measurement function, it is
passed into p_h.
k (iteration number), starts at 0
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
"""
return self.H[:, :, self.index[self.H_time_var_index, k]]
def dHk(self, k):
if self.dH is None:
raise ValueError("dH derivative is None")
return self.dH # the same dirivative on each iteration
if use_cython:
Std_Measurement_Callables_Class = state_space_cython.\
Std_Measurement_Callables_Cython
else:
Std_Measurement_Callables_Class = Std_Measurement_Callables_Python
class Q_handling_Python(Dynamic_Callables_Class):
def __init__(self, Q, index, Q_time_var_index, unique_Q_number, dQ=None):
"""
Input:
---------------
R - array with noise on various steps. The result of preprocessing
the noise input.
index - for each step of Kalman filter contains the corresponding index
in the array.
R_time_var_index - another index in the array R. Computed earlier and
passed here.
unique_R_number - number of unique noise matrices below which square
roots are cached and above which they are computed each time.
dQ: 3D array[:, :, param_num]
derivative of Q. Derivative is supported only when Q do not
change over time
Output:
--------------
Object which has two necessary functions:
f_R(k)
inv_R_square_root(k)
"""
self.Q = Q
self.index = np.asarray(index, np.int_)
self.Q_time_var_index = Q_time_var_index
self.dQ = dQ
if (len(np.unique(index)) > unique_Q_number):
self.svd_each_time = True
else:
self.svd_each_time = False
self.Q_square_root = {}
def Qk(self, k):
"""
function (k). Returns noise matrix of dynamic model on iteration k.
k (iteration number). starts at 0
"""
return self.Q[:, :, self.index[self.Q_time_var_index, k]]
def dQk(self, k):
if self.dQ is None:
raise ValueError("dQ derivative is None")
return self.dQ # the same dirivative on each iteration
def Q_srk(self, k):
"""
function (k). Returns the square root of noise matrix of dynamic model
on iteration k.
k (iteration number). starts at 0
This function is implemented to use SVD prediction step.
"""
ind = self.index[self.Q_time_var_index, k]
Q = self.Q[:, :, ind]
if (Q.shape[0] == 1): # 1-D case handle simplier. No storage
# of the result, just compute it each time.
square_root = np.sqrt(Q)
else:
if self.svd_each_time:
(U, S, Vh) = sp.linalg.svd(Q, full_matrices=False,
compute_uv=True,
overwrite_a=False,
check_finite=True)
square_root = U * np.sqrt(S)
else:
if ind in self.Q_square_root:
square_root = self.Q_square_root[ind]
else:
(U, S, Vh) = sp.linalg.svd(Q, full_matrices=False,
compute_uv=True,
overwrite_a=False,
check_finite=True)
square_root = U * np.sqrt(S)
self.Q_square_root[ind] = square_root
return square_root
if use_cython:
Q_handling_Class = state_space_cython.Q_handling_Cython
else:
Q_handling_Class = Q_handling_Python
class Std_Dynamic_Callables_Python(Q_handling_Class):
def __init__(self, A, A_time_var_index, Q, index, Q_time_var_index,
unique_Q_number, dA=None, dQ=None):
super(Std_Dynamic_Callables_Python,
self).__init__(Q, index, Q_time_var_index, unique_Q_number, dQ)
self.A = A
self.A_time_var_index = np.asarray(A_time_var_index, np.int_)
self.dA = dA
def f_a(self, k, m, A):
"""
f_a: function (k, x_{k-1}, A_{k}). Dynamic function.
k (iteration number), starts at 0
x_{k-1} State from the previous step
A_{k} Jacobian matrices of f_a. In the linear case it is exactly
A_{k}.
"""
return np.dot(A, m)
def Ak(self, k, m_pred, P_pred): # returns state iteration matrix
"""
function (k, m, P) return Jacobian of measurement function, it is
passed into p_h.
k (iteration number), starts at 0
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
"""
return self.A[:, :, self.index[self.A_time_var_index, k]]
def dAk(self, k):
if self.dA is None:
raise ValueError("dA derivative is None")
return self.dA # the same dirivative on each iteration
def reset(self, compute_derivatives=False):
"""
Return the state of this object to the beginning of iteration
(to k eq. 0)
"""
return self
if use_cython:
Std_Dynamic_Callables_Class = state_space_cython.\
Std_Dynamic_Callables_Cython
else:
Std_Dynamic_Callables_Class = Std_Dynamic_Callables_Python
class AddMethodToClass(object):
def __init__(self, func=None, tp='staticmethod'):
"""
Input:
--------------
func: function to add
tp: string
Type of the method: normal, staticmethod, classmethod
"""
if func is None:
raise ValueError("Function can not be None")
self.func = func
self.tp = tp
def __get__(self, obj, klass=None, *args, **kwargs):
if self.tp == 'staticmethod':
return self.func
elif self.tp == 'normal':
def newfunc(obj, *args, **kwargs):
return self.func
elif self.tp == 'classmethod':
def newfunc(klass, *args, **kwargs):
return self.func
return newfunc
class DescreteStateSpaceMeta(type):
"""
Substitute necessary methods from cython.
"""
def __new__(typeclass, name, bases, attributes):
"""
After thos method the class object is created
"""
if use_cython:
if '_kalman_prediction_step_SVD' in attributes:
attributes['_kalman_prediction_step_SVD'] =\
AddMethodToClass(state_space_cython.
_kalman_prediction_step_SVD_Cython)
if '_kalman_update_step_SVD' in attributes:
attributes['_kalman_update_step_SVD'] =\
AddMethodToClass(state_space_cython.
_kalman_update_step_SVD_Cython)
if '_cont_discr_kalman_filter_raw' in attributes:
attributes['_cont_discr_kalman_filter_raw'] =\
AddMethodToClass(state_space_cython.
_cont_discr_kalman_filter_raw_Cython)
return super(DescreteStateSpaceMeta,
typeclass).__new__(typeclass, name, bases, attributes)
class DescreteStateSpace(object):
"""
This class implents state-space inference for linear and non-linear
state-space models.
Linear models are:
x_{k} = A_{k} * x_{k-1} + q_{k-1}; q_{k-1} ~ N(0, Q_{k-1})
y_{k} = H_{k} * x_{k} + r_{k}; r_{k-1} ~ N(0, R_{k})
Nonlinear:
x_{k} = f_a(k, x_{k-1}, A_{k}) + q_{k-1}; q_{k-1} ~ N(0, Q_{k-1})
y_{k} = f_h(k, x_{k}, H_{k}) + r_{k}; r_{k-1} ~ N(0, R_{k})
Here f_a and f_h are some functions of k (iteration number), x_{k-1} or
x_{k} (state value on certain iteration), A_{k} and H_{k} - Jacobian
matrices of f_a and f_h respectively. In the linear case they are exactly
A_{k} and H_{k}.
Currently two nonlinear Gaussian filter algorithms are implemented:
Extended Kalman Filter (EKF), Statistically linearized Filter (SLF), which
implementations are very similar.
"""
__metaclass__ = DescreteStateSpaceMeta
@staticmethod
def _reshape_input_data(shape, desired_dim=3):
"""
Static function returns the column-wise shape for for an input shape.
Input:
--------------
shape: tuple
Shape of an input array, so that it is always a column.
desired_dim: int
desired shape of output. For Y data it should be 3
(sample_no, dimension, ts_no). For X data - 2 (sample_no, 1)
Output:
--------------
new_shape: tuple
New shape of the measurements array. Idea is that samples are
along dimension 0, sample dimension - dimension 1, different
time series - dimension 2.
old_shape: tuple or None
If the shape has been modified, return old shape, otherwise
None.
"""
if (len(shape) > 3):
raise ValueError("""Input array is not supposed to be more
than 3 dimensional.""")
if (len(shape) > desired_dim):
raise ValueError("Input array shape is more than desired shape.")
elif len(shape) == 1:
if (desired_dim == 3):
return ((shape[0], 1, 1), shape) # last dimension is the
# time serime_series_no
elif (desired_dim == 2):
return ((shape[0], 1), shape)
elif len(shape) == 2:
if (desired_dim == 3):
return ((shape[1], 1, 1), shape) if (shape[0] == 1) else\
((shape[0], shape[1], 1), shape) # convert to column
# vector
elif (desired_dim == 2):
return ((shape[1], 1), shape) if (shape[0] == 1) else\
((shape[0], shape[1]), None) # convert to column vector
else: # len(shape) == 3
return (shape, None) # do nothing
@classmethod
def kalman_filter(cls, p_A, p_Q, p_H, p_R, Y, index=None, m_init=None,
P_init=None, p_kalman_filter_type='regular',
calc_log_likelihood=False,
calc_grad_log_likelihood=False, grad_params_no=None,
grad_calc_params=None):
"""
This function implements the basic Kalman Filter algorithm
These notations for the State-Space model are assumed:
x_{k} = A_{k} * x_{k-1} + q_{k-1}; q_{k-1} ~ N(0, Q_{k-1})
y_{k} = H_{k} * x_{k} + r_{k}; r_{k-1} ~ N(0, R_{k})
Returns estimated filter distributions x_{k} ~ N(m_{k}, P(k))
Current Features:
----------------------------------------
1) The function generaly do not modify the passed parameters. If
it happens then it is an error. There are several exeprions: scalars
can be modified into a matrix, in some rare cases shapes of
the derivatives matrices may be changed, it is ignored for now.
2) Copies of p_A, p_Q, index are created in memory to be used later
in smoother. References to copies are kept in "matrs_for_smoother"
return parameter.
3) Function support "multiple time series mode" which means that exactly
the same State-Space model is used to filter several sets of measurements.
In this case third dimension of Y should include these state-space measurements
Log_likelihood and Grad_log_likelihood have the corresponding dimensions then.
4) Calculation of Grad_log_likelihood is not supported if matrices A,Q,
H, or R changes over time. (later may be changed)
5) Measurement may include missing values. In this case update step is
not done for this measurement. (later may be changed)
Input:
-----------------
p_A: scalar, square matrix, 3D array
A_{k} in the model. If matrix then A_{k} = A - constant.
If it is 3D array then A_{k} = p_A[:,:, index[0,k]]
p_Q: scalar, square symmetric matrix, 3D array
Q_{k-1} in the model. If matrix then Q_{k-1} = Q - constant.
If it is 3D array then Q_{k-1} = p_Q[:,:, index[1,k]]
p_H: scalar, matrix (measurement_dim, state_dim) , 3D array
H_{k} in the model. If matrix then H_{k} = H - constant.
If it is 3D array then H_{k} = p_Q[:,:, index[2,k]]
p_R: scalar, square symmetric matrix, 3D array
R_{k} in the model. If matrix then R_{k} = R - constant.
If it is 3D array then R_{k} = p_R[:,:, index[3,k]]
Y: matrix or vector or 3D array
Data. If Y is matrix then samples are along 0-th dimension and
features along the 1-st. If 3D array then third dimension
correspond to "multiple time series mode".
index: vector
Which indices (on 3-rd dimension) from arrays p_A, p_Q,p_H, p_R to use
on every time step. If this parameter is None then it is assumed
that p_A, p_Q, p_H, p_R do not change over time and indices are not needed.
index[0,:] - correspond to A, index[1,:] - correspond to Q
index[2,:] - correspond to H, index[3,:] - correspond to R.
If index.shape[0] == 1, it is assumed that indides for all matrices
are the same.
m_init: vector or matrix
Initial distribution mean. If None it is assumed to be zero.
For "multiple time series mode" it is matrix, second dimension of
which correspond to different time series. In regular case ("one
time series mode") it is a vector.
P_init: square symmetric matrix or scalar
Initial covariance of the states. If the parameter is scalar
then it is assumed that initial covariance matrix is unit matrix
multiplied by this scalar. If None the unit matrix is used instead.
"multiple time series mode" does not affect it, since it does not
affect anything related to state variaces.
calc_log_likelihood: boolean
Whether to calculate marginal likelihood of the state-space model.
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then "grad_calc_params" parameter must
provide the extra parameters for gradient calculation.
grad_params_no: int
If previous parameter is true, then this parameters gives the
total number of parameters in the gradient.
grad_calc_params: dictionary
Dictionary with derivatives of model matrices with respect
to parameters "dA", "dQ", "dH", "dR", "dm_init", "dP_init".
They can be None, in this case zero matrices (no dependence on parameters)
is assumed. If there is only one parameter then third dimension is
automatically added.
Output:
--------------
M: (no_steps+1,state_dim) matrix or (no_steps+1,state_dim, time_series_no) 3D array
Filter estimates of the state means. In the extra step the initial
value is included. In the "multiple time series mode" third dimension
correspond to different timeseries.
P: (no_steps+1, state_dim, state_dim) 3D array
Filter estimates of the state covariances. In the extra step the initial
value is included.
log_likelihood: double or (1, time_series_no) 3D array.
If the parameter calc_log_likelihood was set to true, return
logarithm of marginal likelihood of the state-space model. If
the parameter was false, return None. In the "multiple time series mode" it is a vector
providing log_likelihood for each time series.
grad_log_likelihood: column vector or (grad_params_no, time_series_no) matrix
If calc_grad_log_likelihood is true, return gradient of log likelihood
with respect to parameters. It returns it column wise, so in
"multiple time series mode" gradients for each time series is in the
corresponding column.
matrs_for_smoother: dict
Dictionary with model functions for smoother. The intrinsic model
functions are computed in this functions and they are returned to
use in smoother for convenience. They are: 'p_a', 'p_f_A', 'p_f_Q'
The dictionary contains the same fields.
"""
#import pdb; pdb.set_trace()
# Parameters checking ->
# index
p_A = np.atleast_1d(p_A)
p_Q = np.atleast_1d(p_Q)
p_H = np.atleast_1d(p_H)
p_R = np.atleast_1d(p_R)
# Reshape and check measurements:
Y.shape, old_Y_shape = cls._reshape_input_data(Y.shape)
measurement_dim = Y.shape[1]
time_series_no = Y.shape[2] # multiple time series mode
if ((len(p_A.shape) == 3) and (len(p_A.shape[2]) != 1)) or\
((len(p_Q.shape) == 3) and (len(p_Q.shape[2]) != 1)) or\
((len(p_H.shape) == 3) and (len(p_H.shape[2]) != 1)) or\
((len(p_R.shape) == 3) and (len(p_R.shape[2]) != 1)):
model_matrices_chage_with_time = True
else:
model_matrices_chage_with_time = False
# Check index
old_index_shape = None
if index is None:
if (len(p_A.shape) == 3) or (len(p_Q.shape) == 3) or\
(len(p_H.shape) == 3) or (len(p_R.shape) == 3):
raise ValueError("Parameter index can not be None for time varying matrices (third dimension is present)")
else: # matrices do not change in time, so form dummy zero indices.
index = np.zeros((1,Y.shape[0]))
else:
if len(index.shape) == 1:
index.shape = (1,index.shape[0])
old_index_shape = (index.shape[0],)
if (index.shape[1] != Y.shape[0]):
raise ValueError("Number of measurements must be equal the number of A_{k}, Q_{k}, H_{k}, R_{k}")
if (index.shape[0] == 1):
A_time_var_index = 0; Q_time_var_index = 0
H_time_var_index = 0; R_time_var_index = 0
elif (index.shape[0] == 4):
A_time_var_index = 0; Q_time_var_index = 1
H_time_var_index = 2; R_time_var_index = 3
else:
raise ValueError("First Dimension of index must be either 1 or 4.")
state_dim = p_A.shape[0]
# Check and make right shape for model matrices. On exit they all are 3 dimensional. Last dimension
# correspond to change in time.
(p_A, old_A_shape) = cls._check_SS_matrix(p_A, state_dim, measurement_dim, which='A')
(p_Q, old_Q_shape) = cls._check_SS_matrix(p_Q, state_dim, measurement_dim, which='Q')
(p_H, old_H_shape) = cls._check_SS_matrix(p_H, state_dim, measurement_dim, which='H')
(p_R, old_R_shape) = cls._check_SS_matrix(p_R, state_dim, measurement_dim, which='R')
# m_init
if m_init is None:
m_init = np.zeros((state_dim, time_series_no))
else:
m_init = np.atleast_2d(m_init).T
# P_init
if P_init is None:
P_init = np.eye(state_dim)
elif not isinstance(P_init, collections.Iterable): #scalar
P_init = P_init*np.eye(state_dim)
if p_kalman_filter_type not in ('regular', 'svd'):
raise ValueError("Kalman filer type neither 'regular nor 'svd'.")
# Functions to pass to the kalman_filter algorithm:
# Parameters:
# k - number of Kalman filter iteration
# m - vector for calculating matrices. Required for EKF. Not used here.
c_p_A = p_A.copy() # create a copy because this object is passed to the smoother
c_p_Q = p_Q.copy() # create a copy because this object is passed to the smoother
c_index = index.copy() # create a copy because this object is passed to the smoother
if calc_grad_log_likelihood:
if model_matrices_chage_with_time:
raise ValueError("When computing likelihood gradient A and Q can not change over time.")
dA = cls._check_grad_state_matrices(grad_calc_params.get('dA'), state_dim, grad_params_no, which = 'dA')
dQ = cls._check_grad_state_matrices(grad_calc_params.get('dQ'), state_dim, grad_params_no, which = 'dQ')
dH = cls._check_grad_measurement_matrices(grad_calc_params.get('dH'), state_dim, grad_params_no, measurement_dim, which = 'dH')
dR = cls._check_grad_measurement_matrices(grad_calc_params.get('dR'), state_dim, grad_params_no, measurement_dim, which = 'dR')
dm_init = grad_calc_params.get('dm_init')
if dm_init is None:
# multiple time series mode. Keep grad_params always as a last dimension
dm_init = np.zeros((state_dim, time_series_no, grad_params_no))
dP_init = grad_calc_params.get('dP_init')
if dP_init is None:
dP_init = np.zeros((state_dim,state_dim,grad_params_no))
else:
dA = None
dQ = None
dH = None
dR = None
dm_init = None
dP_init = None
dynamic_callables = Std_Dynamic_Callables_Class(c_p_A, A_time_var_index, c_p_Q, c_index, Q_time_var_index, 20, dA, dQ)
measurement_callables = Std_Measurement_Callables_Class(p_H, H_time_var_index, p_R, index, R_time_var_index, 20, dH, dR)
(M, P,log_likelihood, grad_log_likelihood, dynamic_callables) = \
cls._kalman_algorithm_raw(state_dim, dynamic_callables,
measurement_callables, Y, m_init,
P_init, p_kalman_filter_type = p_kalman_filter_type,
calc_log_likelihood=calc_log_likelihood,
calc_grad_log_likelihood=calc_grad_log_likelihood,
grad_params_no=grad_params_no,
dm_init=dm_init, dP_init=dP_init)
# restore shapes so that input parameters are unchenged
if old_index_shape is not None:
index.shape = old_index_shape
if old_Y_shape is not None:
Y.shape = old_Y_shape
if old_A_shape is not None:
p_A.shape = old_A_shape
if old_Q_shape is not None:
p_Q.shape = old_Q_shape
if old_H_shape is not None:
p_H.shape = old_H_shape
if old_R_shape is not None:
p_R.shape = old_R_shape
# Return values
return (M, P,log_likelihood, grad_log_likelihood, dynamic_callables)
@classmethod
def extended_kalman_filter(cls,p_state_dim, p_a, p_f_A, p_f_Q, p_h, p_f_H, p_f_R, Y, m_init=None,
P_init=None,calc_log_likelihood=False):
"""
Extended Kalman Filter
Input:
-----------------
p_state_dim: integer
p_a: if None - the function from the linear model is assumed. No non-
linearity in the dynamic is assumed.
function (k, x_{k-1}, A_{k}). Dynamic function.
k: (iteration number),
x_{k-1}: (previous state)
x_{k}: Jacobian matrices of f_a. In the linear case it is exactly A_{k}.
p_f_A: matrix - in this case function which returns this matrix is assumed.
Look at this parameter description in kalman_filter function.
function (k, m, P) return Jacobian of dynamic function, it is
passed into p_a.
k: (iteration number),
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
p_f_Q: matrix. In this case function which returns this matrix is asumed.
Look at this parameter description in kalman_filter function.
function (k). Returns noise matrix of dynamic model on iteration k.
k: (iteration number).
p_h: if None - the function from the linear measurement model is assumed.
No nonlinearity in the measurement is assumed.
function (k, x_{k}, H_{k}). Measurement function.
k: (iteration number),
x_{k}: (current state)
H_{k}: Jacobian matrices of f_h. In the linear case it is exactly H_{k}.
p_f_H: matrix - in this case function which returns this matrix is assumed.
function (k, m, P) return Jacobian of dynamic function, it is
passed into p_h.
k: (iteration number),
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
p_f_R: matrix. In this case function which returns this matrix is asumed.
function (k). Returns noise matrix of measurement equation
on iteration k.
k: (iteration number).
Y: matrix or vector
Data. If Y is matrix then samples are along 0-th dimension and
features along the 1-st. May have missing values.
p_mean: vector
Initial distribution mean. If None it is assumed to be zero
P_init: square symmetric matrix or scalar
Initial covariance of the states. If the parameter is scalar
then it is assumed that initial covariance matrix is unit matrix
multiplied by this scalar. If None the unit matrix is used instead.
calc_log_likelihood: boolean
Whether to calculate marginal likelihood of the state-space model.
"""
# Y
Y.shape, old_Y_shape = cls._reshape_input_data(Y.shape)
# m_init
if m_init is None:
m_init = np.zeros((p_state_dim,1))
else:
m_init = np.atleast_2d(m_init).T
# P_init
if P_init is None:
P_init = np.eye(p_state_dim)
elif not isinstance(P_init, collections.Iterable): #scalar
P_init = P_init*np.eye(p_state_dim)
if p_a is None:
p_a = lambda k,m,A: np.dot(A, m)
old_A_shape = None
if not isinstance(p_f_A, types.FunctionType): # not a function but array
p_f_A = np.atleast_1d(p_f_A)
(p_A, old_A_shape) = cls._check_A_matrix(p_f_A)
p_f_A = lambda k, m, P: p_A[:,:, 0] # make function
else:
if p_f_A(1, m_init, P_init).shape[0] != m_init.shape[0]:
raise ValueError("p_f_A function returns matrix of wrong size")
old_Q_shape = None
if not isinstance(p_f_Q, types.FunctionType): # not a function but array
p_f_Q = np.atleast_1d(p_f_Q)
(p_Q, old_Q_shape) = cls._check_Q_matrix(p_f_Q)
p_f_Q = lambda k: p_Q[:,:, 0] # make function
else:
if p_f_Q(1).shape[0] != m_init.shape[0]:
raise ValueError("p_f_Q function returns matrix of wrong size")
if p_h is None:
lambda k,m,H: np.dot(H, m)
old_H_shape = None
if not isinstance(p_f_H, types.FunctionType): # not a function but array
p_f_H = np.atleast_1d(p_f_H)
(p_H, old_H_shape) = cls._check_H_matrix(p_f_H)
p_f_H = lambda k, m, P: p_H # make function
else:
if p_f_H(1, m_init, P_init).shape[0] != Y.shape[1]:
raise ValueError("p_f_H function returns matrix of wrong size")
old_R_shape = None
if not isinstance(p_f_R, types.FunctionType): # not a function but array
p_f_R = np.atleast_1d(p_f_R)
(p_R, old_R_shape) = cls._check_H_matrix(p_f_R)
p_f_R = lambda k: p_R # make function
else:
if p_f_R(1).shape[0] != m_init.shape[0]:
raise ValueError("p_f_R function returns matrix of wrong size")
# class dynamic_callables_class(Dynamic_Model_Callables):
#
# Ak =
# Qk =
class measurement_callables_class(R_handling_Class):
def __init__(self,R, index, R_time_var_index, unique_R_number):
super(measurement_callables_class,self).__init__(R, index, R_time_var_index, unique_R_number)
Hk = AddMethodToClass(f_H)
f_h = AddMethodToClass(f_hl)
(M, P,log_likelihood, grad_log_likelihood) = cls._kalman_algorithm_raw(p_state_dim, p_a, p_f_A, p_f_Q, p_h, p_f_H, p_f_R, Y, m_init,
P_init, calc_log_likelihood,
calc_grad_log_likelihood=False, grad_calc_params=None)
if old_Y_shape is not None:
Y.shape = old_Y_shape
if old_A_shape is not None:
p_A.shape = old_A_shape
if old_Q_shape is not None:
p_Q.shape = old_Q_shape
if old_H_shape is not None:
p_H.shape = old_H_shape
if old_R_shape is not None:
p_R.shape = old_R_shape
return (M, P)
@classmethod
def _kalman_algorithm_raw(cls,state_dim, p_dynamic_callables, p_measurement_callables, Y, m_init,
P_init, p_kalman_filter_type='regular',
calc_log_likelihood=False,
calc_grad_log_likelihood=False, grad_params_no=None,
dm_init=None, dP_init=None):
"""
General nonlinear filtering algorithm for inference in the state-space
model:
x_{k} = f_a(k, x_{k-1}, A_{k}) + q_{k-1}; q_{k-1} ~ N(0, Q_{k-1})
y_{k} = f_h(k, x_{k}, H_{k}) + r_{k}; r_{k-1} ~ N(0, R_{k})
Returns estimated filter distributions x_{k} ~ N(m_{k}, P(k))
Current Features:
----------------------------------------
1) Function support "multiple time series mode" which means that exactly
the same State-Space model is used to filter several sets of measurements.
In this case third dimension of Y should include these state-space measurements
Log_likelihood and Grad_log_likelihood have the corresponding dimensions then.
2) Measurement may include missing values. In this case update step is
not done for this measurement. (later may be changed)
Input:
-----------------
state_dim: int
Demensionality of the states
p_a: function (k, x_{k-1}, A_{k}). Dynamic function.
k (iteration number),
x_{k-1}
A_{k} Jacobian matrices of f_a. In the linear case it is exactly A_{k}.
p_f_A: function (k, m, P) return Jacobian of dynamic function, it is
passed into p_a.
k (iteration number),
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
p_f_Q: function (k). Returns noise matrix of dynamic model on iteration k.
k (iteration number).
p_h: function (k, x_{k}, H_{k}). Measurement function.
k (iteration number),
x_{k}
H_{k} Jacobian matrices of f_h. In the linear case it is exactly H_{k}.
p_f_H: function (k, m, P) return Jacobian of dynamic function, it is
passed into p_h.
k (iteration number),
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
p_f_R: function (k). Returns noise matrix of measurement equation
on iteration k.
k (iteration number).
Y: matrix or vector or 3D array
Data. If Y is matrix then samples are along 0-th dimension and
features along the 1-st. If 3D array then third dimension
correspond to "multiple time series mode".
m_init: vector or matrix
Initial distribution mean. For "multiple time series mode"
it is matrix, second dimension of which correspond to different
time series. In regular case ("one time series mode") it is a
vector.
P_init: matrix or scalar
Initial covariance of the states. Must be not None
"multiple time series mode" does not affect it, since it does not
affect anything related to state variaces.
p_kalman_filter_type: string
calc_log_likelihood: boolean
Whether to calculate marginal likelihood of the state-space model.
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then the next parameter must
provide the extra parameters for gradient calculation.
grad_calc_params: dictionary
Dictionary with derivatives of model matrices with respect
to parameters "dA", "dQ", "dH", "dR", "dm_init", "dP_init".
Output:
--------------
M: (no_steps+1,state_dim) matrix or (no_steps+1,state_dim, time_series_no) 3D array
Filter estimates of the state means. In the extra step the initial
value is included. In the "multiple time series mode" third dimension
correspond to different timeseries.
P: (no_steps+1, state_dim, state_dim) 3D array
Filter estimates of the state covariances. In the extra step the initial
value is included.
log_likelihood: double or (1, time_series_no) 3D array.
If the parameter calc_log_likelihood was set to true, return
logarithm of marginal likelihood of the state-space model. If
the parameter was false, return None. In the "multiple time series mode" it is a vector
providing log_likelihood for each time series.
grad_log_likelihood: column vector or (grad_params_no, time_series_no) matrix
If calc_grad_log_likelihood is true, return gradient of log likelihood
with respect to parameters. It returns it column wise, so in
"multiple time series mode" gradients for each time series is in the
corresponding column.
"""
steps_no = Y.shape[0] # number of steps in the Kalman Filter
time_series_no = Y.shape[2] # multiple time series mode
# Allocate space for results
# Mean estimations. Initial values will be included
M = np.empty(((steps_no+1),state_dim,time_series_no))
M[0,:,:] = m_init # Initialize mean values
# Variance estimations. Initial values will be included
P = np.empty(((steps_no+1),state_dim,state_dim))
P_init = 0.5*( P_init + P_init.T) # symmetrize initial covariance. In some ustable cases this is uiseful
P[0,:,:] = P_init # Initialize initial covariance matrix
if p_kalman_filter_type == 'svd':
(U,S,Vh) = sp.linalg.svd( P_init,full_matrices=False, compute_uv=True,
overwrite_a=False,check_finite=True)
S[ (S==0) ] = 1e-17 # allows to run algorithm for singular initial variance
P_upd = (P_init, S,U)
log_likelihood = 0 if calc_log_likelihood else None
grad_log_likelihood = 0 if calc_grad_log_likelihood else None
#setting initial values for derivatives update
dm_upd = dm_init
dP_upd = dP_init
# Main loop of the Kalman filter
for k in range(0,steps_no):
# In this loop index for new estimations is (k+1), old - (k)
# This happened because initial values are stored at 0-th index.
prev_mean = M[k,:,:] # mean from the previous step
if p_kalman_filter_type == 'svd':
m_pred, P_pred, dm_pred, dP_pred = \
cls._kalman_prediction_step_SVD(k, prev_mean ,P_upd, p_dynamic_callables,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_upd, p_dP = dP_upd)
else:
m_pred, P_pred, dm_pred, dP_pred = \
cls._kalman_prediction_step(k, prev_mean ,P[k,:,:], p_dynamic_callables,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_upd, p_dP = dP_upd )
k_measurment = Y[k,:,:]
if (np.any(np.isnan(k_measurment)) == False):
if p_kalman_filter_type == 'svd':
m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update = \
cls._kalman_update_step_SVD(k, m_pred , P_pred, p_measurement_callables,
k_measurment, calc_log_likelihood=calc_log_likelihood,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_pred, p_dP = dP_pred )
# m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update = \
# cls._kalman_update_step(k, m_pred , P_pred[0], f_h, f_H, p_R.f_R, k_measurment,
# calc_log_likelihood=calc_log_likelihood,
# calc_grad_log_likelihood=calc_grad_log_likelihood,
# p_dm = dm_pred, p_dP = dP_pred, grad_calc_params_2 = (dH, dR))
#
# (U,S,Vh) = sp.linalg.svd( P_upd,full_matrices=False, compute_uv=True,
# overwrite_a=False,check_finite=True)
# P_upd = (P_upd, S,U)
else:
m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update = \
cls._kalman_update_step(k, m_pred , P_pred, p_measurement_callables, k_measurment,
calc_log_likelihood=calc_log_likelihood,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_pred, p_dP = dP_pred )
else:
# if k_measurment.shape != (1,1):
# raise ValueError("Nan measurements are currently not supported for \
# multidimensional output and multiple time series.")
# else:
# m_upd = m_pred; P_upd = P_pred; dm_upd = dm_pred; dP_upd = dP_pred
# log_likelihood_update = 0.0;
# d_log_likelihood_update = 0.0;
if not np.all(np.isnan(k_measurment)):
raise ValueError("""Nan measurements are currently not supported if
they are intermixed with not NaN measurements""")
else:
m_upd = m_pred; P_upd = P_pred; dm_upd = dm_pred; dP_upd = dP_pred
if calc_log_likelihood:
log_likelihood_update = np.zeros((time_series_no,))
if calc_grad_log_likelihood:
d_log_likelihood_update = np.zeros((grad_params_no,time_series_no))
if calc_log_likelihood:
log_likelihood += log_likelihood_update
if calc_grad_log_likelihood:
grad_log_likelihood += d_log_likelihood_update
M[k+1,:,:] = m_upd # separate mean value for each time series
if p_kalman_filter_type == 'svd':
P[k+1,:,:] = P_upd[0]
else:
P[k+1,:,:] = P_upd
# !!!Print statistics! Print sizes of matrices
# !!!Print statistics! Print iteration time base on another boolean variable
return (M, P, log_likelihood, grad_log_likelihood, p_dynamic_callables.reset(False))
@staticmethod
def _kalman_prediction_step(k, p_m , p_P, p_dyn_model_callable, calc_grad_log_likelihood=False,
p_dm = None, p_dP = None):
"""
Desctrete prediction function
Input:
k:int
Iteration No. Starts at 0. Total number of iterations equal to the
number of measurements.
p_m: matrix of size (state_dim, time_series_no)
Mean value from the previous step. For "multiple time series mode"
it is matrix, second dimension of which correspond to different
time series.
p_P:
Covariance matrix from the previous step.
p_dyn_model_callable: class
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then the next parameter must
provide the extra parameters for gradient calculation.
p_dm: 3D array (state_dim, time_series_no, parameters_no)
Mean derivatives from the previous step. For "multiple time series mode"
it is 3D array, second dimension of which correspond to different
time series.
p_dP: 3D array (state_dim, state_dim, parameters_no)
Mean derivatives from the previous step
Output:
----------------------------
m_pred, P_pred, dm_pred, dP_pred: metrices, 3D objects
Results of the prediction steps.
"""
# index correspond to values from previous iteration.
A = p_dyn_model_callable.Ak(k,p_m,p_P) # state transition matrix (or Jacobian)
Q = p_dyn_model_callable.Qk(k) # state noise matrix
# Prediction step ->
m_pred = p_dyn_model_callable.f_a(k, p_m, A) # predicted mean
P_pred = A.dot(p_P).dot(A.T) + Q # predicted variance
# Prediction step <-
if calc_grad_log_likelihood:
dA_all_params = p_dyn_model_callable.dAk(k) # derivatives of A wrt parameters
dQ_all_params = p_dyn_model_callable.dQk(k) # derivatives of Q wrt parameters
param_number = p_dP.shape[2]
# p_dm, p_dP - derivatives form the previoius step
dm_pred = np.empty(p_dm.shape)
dP_pred = np.empty(p_dP.shape)
for j in range(param_number):
dA = dA_all_params[:,:,j]
dQ = dQ_all_params[:,:,j]
dP = p_dP[:,:,j]
dm = p_dm[:,:,j]
dm_pred[:,:,j] = np.dot(dA, p_m) + np.dot(A, dm)
# prediction step derivatives for current parameter:
dP_pred[:,:,j] = np.dot( dA ,np.dot(p_P, A.T))
dP_pred[:,:,j] += dP_pred[:,:,j].T
dP_pred[:,:,j] += np.dot( A ,np.dot(dP, A.T)) + dQ
dP_pred[:,:,j] = 0.5*(dP_pred[:,:,j] + dP_pred[:,:,j].T) #symmetrize
else:
dm_pred = None
dP_pred = None
return m_pred, P_pred, dm_pred, dP_pred
@staticmethod
def _kalman_prediction_step_SVD(k, p_m , p_P, p_dyn_model_callable, calc_grad_log_likelihood=False,
p_dm = None, p_dP = None):
"""
Desctrete prediction function
Input:
k:int
Iteration No. Starts at 0. Total number of iterations equal to the
number of measurements.
p_m: matrix of size (state_dim, time_series_no)
Mean value from the previous step. For "multiple time series mode"
it is matrix, second dimension of which correspond to different
time series.
p_P: tuple (Prev_cov, S, V)
Covariance matrix from the previous step and its SVD decomposition.
Prev_cov = V * S * V.T The tuple is (Prev_cov, S, V)
p_dyn_model_callable: object
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then the next parameter must
provide the extra parameters for gradient calculation.
p_dm: 3D array (state_dim, time_series_no, parameters_no)
Mean derivatives from the previous step. For "multiple time series mode"
it is 3D array, second dimension of which correspond to different
time series.
p_dP: 3D array (state_dim, state_dim, parameters_no)
Mean derivatives from the previous step
Output:
----------------------------
m_pred, P_pred, dm_pred, dP_pred: metrices, 3D objects
Results of the prediction steps.
"""
# covariance from the previous step and its SVD decomposition
# p_prev_cov = v * S * V.T
Prev_cov, S_old, V_old = p_P
#p_prev_cov_tst = np.dot(p_V, (p_S * p_V).T) # reconstructed covariance from the previous step
# index correspond to values from previous iteration.
A = p_dyn_model_callable.Ak(k,p_m,Prev_cov) # state transition matrix (or Jacobian)
Q = p_dyn_model_callable.Qk(k) # state noise matrx. This is necessary for the square root calculation (next step)
Q_sr = p_dyn_model_callable.Q_srk(k)
# Prediction step ->
m_pred = p_dyn_model_callable.f_a(k, p_m, A) # predicted mean
# coavariance prediction have changed:
svd_1_matr = np.vstack( ( (np.sqrt(S_old)* np.dot(A,V_old)).T , Q_sr.T) )
(U,S,Vh) = sp.linalg.svd( svd_1_matr,full_matrices=False, compute_uv=True,
overwrite_a=False,check_finite=True)
# predicted variance computed by the regular method. For testing
#P_pred_tst = A.dot(Prev_cov).dot(A.T) + Q
V_new = Vh.T
S_new = S**2
P_pred = np.dot(V_new * S_new, V_new.T) # prediction covariance
P_pred = (P_pred, S_new, Vh.T)
# Prediction step <-
# derivatives
if calc_grad_log_likelihood:
dA_all_params = p_dyn_model_callable.dAk(k) # derivatives of A wrt parameters
dQ_all_params = p_dyn_model_callable.dQk(k) # derivatives of Q wrt parameters
param_number = p_dP.shape[2]
# p_dm, p_dP - derivatives form the previoius step
dm_pred = np.empty(p_dm.shape)
dP_pred = np.empty(p_dP.shape)
for j in range(param_number):
dA = dA_all_params[:,:,j]
dQ = dQ_all_params[:,:,j]
#dP = p_dP[:,:,j]
#dm = p_dm[:,:,j]
dm_pred[:,:,j] = np.dot(dA, p_m) + np.dot(A, p_dm[:,:,j])
# prediction step derivatives for current parameter:
dP_pred[:,:,j] = np.dot( dA ,np.dot(Prev_cov, A.T))
dP_pred[:,:,j] += dP_pred[:,:,j].T
dP_pred[:,:,j] += np.dot( A ,np.dot(p_dP[:,:,j], A.T)) + dQ
dP_pred[:,:,j] = 0.5*(dP_pred[:,:,j] + dP_pred[:,:,j].T) #symmetrize
else:
dm_pred = None
dP_pred = None
return m_pred, P_pred, dm_pred, dP_pred
@staticmethod
def _kalman_update_step(k, p_m , p_P, p_meas_model_callable, measurement, calc_log_likelihood= False,
calc_grad_log_likelihood=False, p_dm = None, p_dP = None):
"""
Input:
k: int
Iteration No. Starts at 0. Total number of iterations equal to the
number of measurements.
m_P: matrix of size (state_dim, time_series_no)
Mean value from the previous step. For "multiple time series mode"
it is matrix, second dimension of which correspond to different
time series.
p_P:
Covariance matrix from the prediction step.
p_meas_model_callable: object
measurement: (measurement_dim, time_series_no) matrix
One measurement used on the current update step. For
"multiple time series mode" it is matrix, second dimension of
which correspond to different time series.
calc_log_likelihood: boolean
Whether to calculate marginal likelihood of the state-space model.
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then the next parameter must
provide the extra parameters for gradient calculation.
p_dm: 3D array (state_dim, time_series_no, parameters_no)
Mean derivatives from the prediction step. For "multiple time series mode"
it is 3D array, second dimension of which correspond to different
time series.
p_dP: array
Covariance derivatives from the prediction step.
Output:
----------------------------
m_upd, P_upd, dm_upd, dP_upd: metrices, 3D objects
Results of the prediction steps.
log_likelihood_update: double or 1D array
Update to the log_likelihood from this step
d_log_likelihood_update: (grad_params_no, time_series_no) matrix
Update to the gradient of log_likelihood, "multiple time series mode"
adds extra columns to the gradient.
"""
#import pdb; pdb.set_trace()
m_pred = p_m # from prediction step
P_pred = p_P # from prediction step
H = p_meas_model_callable.Hk(k, m_pred, P_pred)
R = p_meas_model_callable.Rk(k)
time_series_no = p_m.shape[1] # number of time serieses
log_likelihood_update=None; dm_upd=None; dP_upd=None; d_log_likelihood_update=None
# Update step (only if there is data)
#if not np.any(np.isnan(measurement)): # TODO: if some dimensions are missing, do properly computations for other.
v = measurement-p_meas_model_callable.f_h(k, m_pred, H)
S = H.dot(P_pred).dot(H.T) + R
if measurement.shape[0]==1: # measurements are one dimensional
if (S < 0):
raise ValueError("Kalman Filter Update: S is negative step %i" % k )
#import pdb; pdb.set_trace()
K = P_pred.dot(H.T) / S
if calc_log_likelihood:
log_likelihood_update = -0.5 * ( np.log(2*np.pi) + np.log(S) +
v*v / S)
#log_likelihood_update = log_likelihood_update[0,0] # to make int
if np.any(np.isnan(log_likelihood_update)): # some member in P_pred is None.
raise ValueError("Nan values in likelihood update!")
LL = None; islower = None
else:
LL,islower = linalg.cho_factor(S)
K = linalg.cho_solve((LL,islower), H.dot(P_pred.T)).T
if calc_log_likelihood:
log_likelihood_update = -0.5 * ( v.shape[0]*np.log(2*np.pi) +
2*np.sum( np.log(np.diag(LL)) ) +\
np.sum((linalg.cho_solve((LL,islower),v)) * v, axis = 0) ) # diagonal of v.T*S^{-1}*v
if calc_grad_log_likelihood:
dm_pred_all_params = p_dm # derivativas of the prediction phase
dP_pred_all_params = p_dP
param_number = p_dP.shape[2]
dH_all_params = p_meas_model_callable.dHk(k)
dR_all_params = p_meas_model_callable.dRk(k)
dm_upd = np.empty(dm_pred_all_params.shape)
dP_upd = np.empty(dP_pred_all_params.shape)
# firts dimension parameter_no, second - time series number
d_log_likelihood_update = np.empty((param_number,time_series_no))
for param in range(param_number):
dH = dH_all_params[:,:,param]
dR = dR_all_params[:,:,param]
dm_pred = dm_pred_all_params[:,:,param]
dP_pred = dP_pred_all_params[:,:,param]
# Terms in the likelihood derivatives
dv = - np.dot( dH, m_pred) - np.dot( H, dm_pred)
dS = np.dot(dH, np.dot( P_pred, H.T))
dS += dS.T
dS += np.dot(H, np.dot( dP_pred, H.T)) + dR
# TODO: maybe symmetrize dS
#dm and dP for the next stem
if LL is not None: # the state vector is not a scalar
tmp1 = linalg.cho_solve((LL,islower), H).T
tmp2 = linalg.cho_solve((LL,islower), dH).T
tmp3 = linalg.cho_solve((LL,islower), dS).T
else: # the state vector is a scalar
tmp1 = H.T / S
tmp2 = dH.T / S
tmp3 = dS.T / S
dK = np.dot( dP_pred, tmp1) + np.dot( P_pred, tmp2) - \
np.dot( P_pred, np.dot( tmp1, tmp3 ) )
# terms required for the next step, save this for each parameter
dm_upd[:,:,param] = dm_pred + np.dot(dK, v) + np.dot(K, dv)
dP_upd[:,:,param] = -np.dot(dK, np.dot(S, K.T))
dP_upd[:,:,param] += dP_upd[:,:,param].T
dP_upd[:,:,param] += dP_pred - np.dot(K , np.dot( dS, K.T))
dP_upd[:,:,param] = 0.5*(dP_upd[:,:,param] + dP_upd[:,:,param].T) #symmetrize
# computing the likelihood change for each parameter:
if LL is not None: # the state vector is not 1D
#tmp4 = linalg.cho_solve((LL,islower), dv)
tmp5 = linalg.cho_solve((LL,islower), v)
else: # the state vector is a scalar
#tmp4 = dv / S
tmp5 = v / S
d_log_likelihood_update[param,:] = -(0.5*np.sum(np.diag(tmp3)) + \
np.sum(tmp5*dv, axis=0) - 0.5 * np.sum(tmp5 * np.dot(dS, tmp5), axis=0) )
# Before
#d_log_likelihood_update[param,0] = -(0.5*np.sum(np.diag(tmp3)) + \
#np.dot(tmp5.T, dv) - 0.5 * np.dot(tmp5.T ,np.dot(dS, tmp5)) )
# Compute the actual updates for mean and variance of the states.
m_upd = m_pred + K.dot( v )
# Covariance update and ensure it is symmetric
P_upd = K.dot(S).dot(K.T)
P_upd = 0.5*(P_upd + P_upd.T)
P_upd = P_pred - P_upd# this update matrix is symmetric
return m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update
@staticmethod
def _kalman_update_step_SVD(k, p_m , p_P, p_meas_model_callable, measurement, calc_log_likelihood= False,
calc_grad_log_likelihood=False, p_dm = None, p_dP = None):
"""
Input:
k: int
Iteration No. Starts at 0. Total number of iterations equal to the
number of measurements.
m_P: matrix of size (state_dim, time_series_no)
Mean value from the previous step. For "multiple time series mode"
it is matrix, second dimension of which correspond to different
time series.
p_P: tuple (P_pred, S, V)
Covariance matrix from the prediction step and its SVD decomposition.
P_pred = V * S * V.T The tuple is (P_pred, S, V)
p_h: function (k, x_{k}, H_{k}). Measurement function.
k (iteration number), starts at 0
x_{k} state
H_{k} Jacobian matrices of f_h. In the linear case it is exactly H_{k}.
p_f_H: function (k, m, P) return Jacobian of measurement function, it is
passed into p_h.
k (iteration number), starts at 0
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
p_f_R: function (k). Returns noise matrix of measurement equation
on iteration k.
k (iteration number). starts at 0
p_f_iRsr: function (k). Returns the square root of the noise matrix of
measurement equation on iteration k.
k (iteration number). starts at 0
measurement: (measurement_dim, time_series_no) matrix
One measurement used on the current update step. For
"multiple time series mode" it is matrix, second dimension of
which correspond to different time series.
calc_log_likelihood: boolean
Whether to calculate marginal likelihood of the state-space model.
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then the next parameter must
provide the extra parameters for gradient calculation.
p_dm: 3D array (state_dim, time_series_no, parameters_no)
Mean derivatives from the prediction step. For "multiple time series mode"
it is 3D array, second dimension of which correspond to different
time series.
p_dP: array
Covariance derivatives from the prediction step.
grad_calc_params_2: List or None
List with derivatives. The first component is 'f_dH' - function(k)
which returns the derivative of H. The second element is 'f_dR'
- function(k). Function which returns the derivative of R.
Output:
----------------------------
m_upd, P_upd, dm_upd, dP_upd: metrices, 3D objects
Results of the prediction steps.
log_likelihood_update: double or 1D array
Update to the log_likelihood from this step
d_log_likelihood_update: (grad_params_no, time_series_no) matrix
Update to the gradient of log_likelihood, "multiple time series mode"
adds extra columns to the gradient.
"""
#import pdb; pdb.set_trace()
m_pred = p_m # from prediction step
P_pred,S_pred,V_pred = p_P # from prediction step
H = p_meas_model_callable.Hk(k, m_pred, P_pred)
R = p_meas_model_callable.Rk(k)
R_isr = p_meas_model_callable.R_isrk(k) # square root of the inverse of R matrix
time_series_no = p_m.shape[1] # number of time serieses
log_likelihood_update=None; dm_upd=None; dP_upd=None; d_log_likelihood_update=None
# Update step (only if there is data)
#if not np.any(np.isnan(measurement)): # TODO: if some dimensions are missing, do properly computations for other.
v = measurement-p_meas_model_callable.f_h(k, m_pred, H)
svd_2_matr = np.vstack( ( np.dot( R_isr.T, np.dot(H, V_pred)) , np.diag( 1.0/np.sqrt(S_pred) ) ) )
(U,S,Vh) = sp.linalg.svd( svd_2_matr,full_matrices=False, compute_uv=True,
overwrite_a=False,check_finite=True)
# P_upd = U_upd S_upd**2 U_upd.T
U_upd = np.dot(V_pred, Vh.T)
S_upd = (1.0/S)**2
P_upd = np.dot(U_upd * S_upd, U_upd.T) # update covariance
P_upd = (P_upd,S_upd,U_upd) # tuple to pass to the next step
# stil need to compute S and K for derivative computation
S = H.dot(P_pred).dot(H.T) + R
if measurement.shape[0]==1: # measurements are one dimensional
if (S < 0):
raise ValueError("Kalman Filter Update: S is negative step %i" % k )
#import pdb; pdb.set_trace()
K = P_pred.dot(H.T) / S
if calc_log_likelihood:
log_likelihood_update = -0.5 * ( np.log(2*np.pi) + np.log(S) +
v*v / S)
#log_likelihood_update = log_likelihood_update[0,0] # to make int
if np.any(np.isnan(log_likelihood_update)): # some member in P_pred is None.
raise ValueError("Nan values in likelihood update!")
LL = None; islower = None
else:
LL,islower = linalg.cho_factor(S)
K = linalg.cho_solve((LL,islower), H.dot(P_pred.T)).T
if calc_log_likelihood:
log_likelihood_update = -0.5 * ( v.shape[0]*np.log(2*np.pi) +
2*np.sum( np.log(np.diag(LL)) ) +\
np.sum((linalg.cho_solve((LL,islower),v)) * v, axis = 0) ) # diagonal of v.T*S^{-1}*v
# Old method of computing updated covariance (for testing) ->
#P_upd_tst = K.dot(S).dot(K.T)
#P_upd_tst = 0.5*(P_upd_tst + P_upd_tst.T)
#P_upd_tst = P_pred - P_upd_tst# this update matrix is symmetric
# Old method of computing updated covariance (for testing) <-
if calc_grad_log_likelihood:
dm_pred_all_params = p_dm # derivativas of the prediction phase
dP_pred_all_params = p_dP
param_number = p_dP.shape[2]
dH_all_params = p_meas_model_callable.dHk(k)
dR_all_params = p_meas_model_callable.dRk(k)
dm_upd = np.empty(dm_pred_all_params.shape)
dP_upd = np.empty(dP_pred_all_params.shape)
# firts dimension parameter_no, second - time series number
d_log_likelihood_update = np.empty((param_number,time_series_no))
for param in range(param_number):
dH = dH_all_params[:,:,param]
dR = dR_all_params[:,:,param]
dm_pred = dm_pred_all_params[:,:,param]
dP_pred = dP_pred_all_params[:,:,param]
# Terms in the likelihood derivatives
dv = - np.dot( dH, m_pred) - np.dot( H, dm_pred)
dS = np.dot(dH, np.dot( P_pred, H.T))
dS += dS.T
dS += np.dot(H, np.dot( dP_pred, H.T)) + dR
# TODO: maybe symmetrize dS
#dm and dP for the next stem
if LL is not None: # the state vector is not a scalar
tmp1 = linalg.cho_solve((LL,islower), H).T
tmp2 = linalg.cho_solve((LL,islower), dH).T
tmp3 = linalg.cho_solve((LL,islower), dS).T
else: # the state vector is a scalar
tmp1 = H.T / S
tmp2 = dH.T / S
tmp3 = dS.T / S
dK = np.dot( dP_pred, tmp1) + np.dot( P_pred, tmp2) - \
np.dot( P_pred, np.dot( tmp1, tmp3 ) )
# terms required for the next step, save this for each parameter
dm_upd[:,:,param] = dm_pred + np.dot(dK, v) + np.dot(K, dv)
dP_upd[:,:,param] = -np.dot(dK, np.dot(S, K.T))
dP_upd[:,:,param] += dP_upd[:,:,param].T
dP_upd[:,:,param] += dP_pred - np.dot(K , np.dot( dS, K.T))
dP_upd[:,:,param] = 0.5*(dP_upd[:,:,param] + dP_upd[:,:,param].T) #symmetrize
# computing the likelihood change for each parameter:
if LL is not None: # the state vector is not 1D
tmp5 = linalg.cho_solve((LL,islower), v)
else: # the state vector is a scalar
tmp5 = v / S
d_log_likelihood_update[param,:] = -(0.5*np.sum(np.diag(tmp3)) + \
np.sum(tmp5*dv, axis=0) - 0.5 * np.sum(tmp5 * np.dot(dS, tmp5), axis=0) )
# Before
#d_log_likelihood_update[param,0] = -(0.5*np.sum(np.diag(tmp3)) + \
#np.dot(tmp5.T, dv) - 0.5 * np.dot(tmp5.T ,np.dot(dS, tmp5)) )
# Compute the actual updates for mean of the states. Variance update
# is computed earlier.
m_upd = m_pred + K.dot( v )
return m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update
@staticmethod
def _rts_smoother_update_step(k, p_m , p_P, p_m_pred, p_P_pred, p_m_prev_step,
p_P_prev_step, p_dynamic_callables):
"""
Rauch–Tung–Striebel(RTS) update step
Input:
-----------------------------
k: int
Iteration No. Starts at 0. Total number of iterations equal to the
number of measurements.
p_m: matrix of size (state_dim, time_series_no)
Filter mean on step k
p_P: matrix of size (state_dim,state_dim)
Filter Covariance on step k
p_m_pred: matrix of size (state_dim, time_series_no)
Means from the smoother prediction step.
p_P_pred:
Covariance from the smoother prediction step.
p_m_prev_step
Smoother mean from the previous step.
p_P_prev_step:
Smoother covariance from the previous step.
p_f_A: function (k, m, P) return Jacobian of dynamic function, it is
passed into p_a.
k (iteration number), starts at 0
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
"""
A = p_dynamic_callables.Ak(k,p_m,p_P) # state transition matrix (or Jacobian)
tmp = np.dot( A, p_P.T)
if A.shape[0] == 1: # 1D states
G = tmp.T / p_P_pred # P[:,:,k] is symmetric
else:
try:
LL,islower = linalg.cho_factor(p_P_pred)
G = linalg.cho_solve((LL,islower),tmp).T
except:
# It happende that p_P_pred has several near zero eigenvalues
# hence the Cholesky method does not work.
res = sp.linalg.lstsq(p_P_pred, tmp)
G = res[0].T
m_upd = p_m + G.dot( p_m_prev_step-p_m_pred )
P_upd = p_P + G.dot( p_P_prev_step-p_P_pred).dot(G.T)
P_upd = 0.5*(P_upd + P_upd.T)
return m_upd, P_upd, G
@classmethod
def rts_smoother(cls,state_dim, p_dynamic_callables, filter_means,
filter_covars):
"""
This function implements Rauch–Tung–Striebel(RTS) smoother algorithm
based on the results of kalman_filter_raw.
These notations are the same:
x_{k} = A_{k} * x_{k-1} + q_{k-1}; q_{k-1} ~ N(0, Q_{k-1})
y_{k} = H_{k} * x_{k} + r_{k}; r_{k-1} ~ N(0, R_{k})
Returns estimated smoother distributions x_{k} ~ N(m_{k}, P(k))
Input:
--------------
p_a: function (k, x_{k-1}, A_{k}). Dynamic function.
k (iteration number), starts at 0
x_{k-1} State from the previous step
A_{k} Jacobian matrices of f_a. In the linear case it is exactly A_{k}.
p_f_A: function (k, m, P) return Jacobian of dynamic function, it is
passed into p_a.
k (iteration number), starts at 0
m: point where Jacobian is evaluated
P: parameter for Jacobian, usually covariance matrix.
p_f_Q: function (k). Returns noise matrix of dynamic model on iteration k.
k (iteration number). starts at 0
filter_means: (no_steps+1,state_dim) matrix or (no_steps+1,state_dim, time_series_no) 3D array
Results of the Kalman Filter means estimation.
filter_covars: (no_steps+1, state_dim, state_dim) 3D array
Results of the Kalman Filter covariance estimation.
Output:
-------------
M: (no_steps+1, state_dim) matrix
Smoothed estimates of the state means
P: (no_steps+1, state_dim, state_dim) 3D array
Smoothed estimates of the state covariances
"""
no_steps = filter_covars.shape[0]-1# number of steps (minus initial covariance)
M = np.empty(filter_means.shape) # smoothed means
P = np.empty(filter_covars.shape) # smoothed covars
#G = np.empty( (no_steps,state_dim,state_dim) ) # G from the update step of the smoother
M[-1,:] = filter_means[-1,:]
P[-1,:,:] = filter_covars[-1,:,:]
for k in range(no_steps-1,-1,-1):
m_pred, P_pred, tmp1, tmp2 = \
cls._kalman_prediction_step(k, filter_means[k,:],
filter_covars[k,:,:], p_dynamic_callables,
calc_grad_log_likelihood=False)
p_m = filter_means[k,:]
if len(p_m.shape)<2:
p_m.shape = (p_m.shape[0],1)
p_m_prev_step = M[k+1,:]
if len(p_m_prev_step.shape)<2:
p_m_prev_step.shape = (p_m_prev_step.shape[0],1)
m_upd, P_upd, G_tmp = cls._rts_smoother_update_step(k,
p_m ,filter_covars[k,:,:],
m_pred, P_pred, p_m_prev_step ,P[k+1,:,:], p_dynamic_callables)
M[k,:] = m_upd#np.squeeze(m_upd)
P[k,:,:] = P_upd
#G[k,:,:] = G_upd.T # store transposed G.
# Return values
return (M, P) #, G)
@staticmethod
def _EM_gradient(A,Q,H,R,m_init,P_init,measurements, M, P, G, dA, dQ, dH, dR, dm_init, dP_init):
"""
Gradient computation with the EM algorithm.
Input:
-----------------
M: Means from the smoother
P: Variances from the smoother
G: Gains? from the smoother
"""
import pdb; pdb.set_trace();
param_number = dA.shape[-1]
d_log_likelihood_update = np.empty((param_number,1))
sample_no = measurements.shape[0]
P_1 = P[1:,:,:] # remove 0-th step
P_2 = P[0:-1,:,:] # remove 0-th step
M_1 = M[1:,:] # remove 0-th step
M_2 = M[0:-1,:] # remove the last step
Sigma = np.mean(P_1,axis=0) + np.dot(M_1.T, M_1) / sample_no #
Phi = np.mean(P_2,axis=0) + np.dot(M_2.T, M_2) / sample_no #
B = np.dot( measurements.T, M_1 )/ sample_no
C = (sp.einsum( 'ijk,ikl', P_1, G) + np.dot(M_1.T, M_2)) / sample_no #
# C1 = np.zeros( (P_1.shape[1],P_1.shape[1]) )
# for k in range(P_1.shape[0]):
# C1 += np.dot(P_1[k,:,:],G[k,:,:]) + sp.outer( M_1[k,:], M_2[k,:] )
# C1 = C1 / sample_no
D = np.dot( measurements.T, measurements ) / sample_no
try:
P_init_inv = sp.linalg.inv(P_init)
if np.max( np.abs(P_init_inv)) > 10e13:
compute_P_init_terms = False
else:
compute_P_init_terms = True
except np.linalg.LinAlgError:
compute_P_init_terms = False
try:
Q_inv = sp.linalg.inv(Q)
if np.max( np.abs(Q_inv)) > 10e13:
compute_Q_terms = False
else:
compute_Q_terms = True
except np.linalg.LinAlgError:
compute_Q_terms = False
try:
R_inv = sp.linalg.inv(R)
if np.max( np.abs(R_inv)) > 10e13:
compute_R_terms = False
else:
compute_R_terms = True
except np.linalg.LinAlgError:
compute_R_terms = False
d_log_likelihood_update = np.zeros((param_number,1))
for j in range(param_number):
if compute_P_init_terms:
d_log_likelihood_update[j,:] -= 0.5 * np.sum(P_init_inv* dP_init[:,:,j].T ) #p #m
M0_smoothed = M[0]; M0_smoothed.shape = (M0_smoothed.shape[0],1)
tmp1 = np.dot( dP_init[:,:,j], np.dot( P_init_inv, (P[0,:,:] + sp.outer( (M0_smoothed - m_init), (M0_smoothed - m_init) )) ) ) #p #m
d_log_likelihood_update[j,:] += 0.5 * np.sum(P_init_inv* tmp1.T )
tmp2 = sp.outer( dm_init[:,j], M0_smoothed )
tmp2 += tmp2.T
d_log_likelihood_update[j,:] += 0.5 * np.sum(P_init_inv* tmp2.T )
if compute_Q_terms:
d_log_likelihood_update[j,:] -= sample_no/2.0 * np.sum(Q_inv* dQ[:,:,j].T ) #m
tmp1 = np.dot(C,A.T); tmp1 += tmp1.T; tmp1 = Sigma - tmp1 + np.dot(A, np.dot(Phi,A.T)) #m
tmp1 = np.dot( dQ[:,:,j], np.dot( Q_inv, tmp1) )
d_log_likelihood_update[j,:] += sample_no/2.0 * np.sum(Q_inv * tmp1.T)
tmp2 = np.dot( dA[:,:,j], C.T); tmp2 += tmp2.T;
tmp3 = np.dot(dA[:,:,j], np.dot(Phi,A.T)); tmp3 += tmp3.T
d_log_likelihood_update[j,:] -= sample_no/2.0 * np.sum(Q_inv.T * (tmp3 - tmp2) )
if compute_R_terms:
d_log_likelihood_update[j,:] -= sample_no/2.0 * np.sum(R_inv* dR[:,:,j].T )
tmp1 = np.dot(B,H.T); tmp1 += tmp1.T; tmp1 = D - tmp1 + np.dot(H, np.dot(Sigma,H.T))
tmp1 = np.dot( dR[:,:,j], np.dot( R_inv, tmp1) )
d_log_likelihood_update[j,:] += sample_no/2.0 * np.sum(R_inv * tmp1.T)
tmp2 = np.dot( dH[:,:,j], B.T); tmp2 += tmp2.T;
tmp3 = np.dot(dH[:,:,j], np.dot(Sigma,H.T)); tmp3 += tmp3.T
d_log_likelihood_update[j,:] -= sample_no/2.0 * np.sum(R_inv.T * (tmp3 - tmp2) )
return d_log_likelihood_update
@staticmethod
def _check_SS_matrix(p_M, state_dim, measurement_dim, which='A'):
"""
Veryfy that on exit the matrix has appropriate shape for the KF algorithm.
Input:
p_M: matrix
As it is given for the user
state_dim: int
State dimensioanlity
measurement_dim: int
Measurement dimensionality
which: string
One of: 'A', 'Q', 'H', 'R'
Output:
---------------
p_M: matrix of the right shape
old_M_shape: tuple
Old Shape
"""
old_M_shape = None
if len(p_M.shape) < 3: # new shape is 3 dimensional
old_M_shape = p_M.shape # save shape to restore it on exit
if len(p_M.shape) == 2: # matrix
p_M.shape = (p_M.shape[0],p_M.shape[1],1)
elif len(p_M.shape) == 1: # scalar but in array already
if (p_M.shape[0] != 1):
raise ValueError("Matrix %s is an 1D array, while it must be a matrix or scalar", which)
else:
p_M.shape = (1,1,1)
if (which == 'A') or (which == 'Q'):
if (p_M.shape[0] != state_dim) or (p_M.shape[1] != state_dim):
raise ValueError("%s must be a square matrix of size (%i,%i)" % (which, state_dim, state_dim))
if (which == 'H'):
if (p_M.shape[0] != measurement_dim) or (p_M.shape[1] != state_dim):
raise ValueError("H must be of shape (measurement_dim, state_dim) (%i,%i)" % (measurement_dim, state_dim))
if (which == 'R'):
if (p_M.shape[0] != measurement_dim) or (p_M.shape[1] != measurement_dim):
raise ValueError("R must be of shape (measurement_dim, measurement_dim) (%i,%i)" % (measurement_dim, measurement_dim))
return (p_M,old_M_shape)
@staticmethod
def _check_grad_state_matrices(dM, state_dim, grad_params_no, which = 'dA'):
"""
Function checks (mostly check dimensions) matrices for marginal likelihood
gradient parameters calculation. It check dA, dQ matrices.
Input:
-------------
dM: None, scaler or 3D matrix
It is supposed to be (state_dim,state_dim,grad_params_no) matrix.
If None then zero matrix is assumed. If scalar then the function
checks consistency with "state_dim" and "grad_params_no".
state_dim: int
State dimensionality
grad_params_no: int
How many parrameters of likelihood gradient in total.
which: string
'dA' or 'dQ'
Output:
--------------
function of (k) which returns the parameters matrix.
"""
if dM is None:
dM=np.zeros((state_dim,state_dim,grad_params_no))
elif isinstance(dM, np.ndarray):
if state_dim == 1:
if len(dM.shape) < 3:
dM.shape = (1,1,1)
else:
if len(dM.shape) < 3:
dM.shape = (state_dim,state_dim,1)
elif isinstance(dM, np.int):
if state_dim > 1:
raise ValueError("When computing likelihood gradient wrong %s dimension." % which)
else:
dM = np.ones((1,1,1)) * dM
# if not isinstance(dM, types.FunctionType):
# f_dM = lambda k: dM
# else:
# f_dM = dM
return dM
@staticmethod
def _check_grad_measurement_matrices(dM, state_dim, grad_params_no, measurement_dim, which = 'dH'):
"""
Function checks (mostly check dimensions) matrices for marginal likelihood
gradient parameters calculation. It check dH, dR matrices.
Input:
-------------
dM: None, scaler or 3D matrix
It is supposed to be
(measurement_dim ,state_dim,grad_params_no) for "dH" matrix.
(measurement_dim,measurement_dim,grad_params_no) for "dR"
If None then zero matrix is assumed. If scalar then the function
checks consistency with "state_dim" and "grad_params_no".
state_dim: int
State dimensionality
grad_params_no: int
How many parrameters of likelihood gradient in total.
measurement_dim: int
Dimensionality of measurements.
which: string
'dH' or 'dR'
Output:
--------------
function of (k) which returns the parameters matrix.
"""
if dM is None:
if which == 'dH':
dM=np.zeros((measurement_dim ,state_dim,grad_params_no))
elif which == 'dR':
dM=np.zeros((measurement_dim,measurement_dim,grad_params_no))
elif isinstance(dM, np.ndarray):
if state_dim == 1:
if len(dM.shape) < 3:
dM.shape = (1,1,1)
else:
if len(dM.shape) < 3:
if which == 'dH':
dM.shape = (measurement_dim,state_dim,1)
elif which == 'dR':
dM.shape = (measurement_dim,measurement_dim,1)
elif isinstance(dM, np.int):
if state_dim > 1:
raise ValueError("When computing likelihood gradient wrong dH dimension.")
else:
dM = np.ones((1,1,1)) * dM
# if not isinstance(dM, types.FunctionType):
# f_dM = lambda k: dM
# else:
# f_dM = dM
return dM
class Struct(object):
pass
class ContDescrStateSpace(DescreteStateSpace):
"""
Class for continuous-discrete Kalman filter. State equation is
continuous while measurement equation is discrete.
d x(t)/ dt = F x(t) + L q; where q~ N(0, Qc)
y_{t_k} = H_{k} x_{t_k} + r_{k}; r_{k-1} ~ N(0, R_{k})
"""
class AQcompute_once(Q_handling_Class):
"""
Class for calculating matrices A, Q, dA, dQ of the discrete Kalman Filter
from the matrices F, L, Qc, P_ing, dF, dQc, dP_inf of the continuos state
equation. dt - time steps.
It has the same interface as AQcompute_batch.
It computes matrices for only one time step. This object is used when
there are many different time steps and storing matrices for each of them
would take too much memory.
"""
def __init__(self, F,L,Qc,dt,compute_derivatives=False, grad_params_no=None, P_inf=None, dP_inf=None, dF = None, dQc=None):
"""
Constructor. All necessary parameters are passed here and stored
in the opject.
Input:
-------------------
F, L, Qc, P_inf : matrices
Parameters of corresponding continuous state model
dt: array
All time steps
compute_derivatives: bool
Whether to calculate derivatives
dP_inf, dF, dQc: 3D array
Derivatives if they are required
Output:
-------------------
Nothing
"""
# Copies are done because this object is used later in smoother
# and these parameters must not change.
self.F = F.copy()
self.L = L.copy()
self.Qc = Qc.copy()
self.dt = dt # copy is not taken because dt is internal parameter
# Parameters are used to calculate derivatives but derivatives
# are not used in the smoother. Therefore copies are not taken.
self.P_inf = P_inf
self.dP_inf = dP_inf
self.dF = dF
self.dQc = dQc
self.compute_derivatives = compute_derivatives
self.grad_params_no = grad_params_no
self.last_k = 0
self.last_k_computed = False
self.v_Ak = None
self.v_Qk = None
self.v_dAk = None
self.v_dQk = None
self.square_root_computed = False
# !!!Print statistics! Which object is created
def f_a(self, k,m,A):
"""
Dynamic model
"""
return np.dot(A, m) # default dynamic model
def _recompute_for_new_k(self,k):
"""
Computes the necessary matrices for an index k and store the results.
Input:
----------------------
k: int
Index in the time differences array dt where to compute matrices
Output:
----------------------
Ak,Qk, dAk, dQk: matrices and/or 3D arrays
A, Q, dA dQ on step k
"""
if (self.last_k != k) or (self.last_k_computed == False):
v_Ak,v_Qk, tmp, v_dAk, v_dQk = ContDescrStateSpace.lti_sde_to_descrete(self.F,
self.L,self.Qc,self.dt[k],self.compute_derivatives,
grad_params_no=self.grad_params_no, P_inf=self.P_inf, dP_inf=self.dP_inf, dF=self.dF, dQc=self.dQc)
self.last_k = k
self.last_k_computed = True
self.v_Ak = v_Ak
self.v_Qk = v_Qk
self.v_dAk = v_dAk
self.v_dQk = v_dQk
self.Q_square_root_computed = False
else:
v_Ak = self.v_Ak
v_Qk = self.v_Qk
v_dAk = self.v_dAk
v_dQk = self.v_dQk
# !!!Print statistics! Print sizes of matrices
return v_Ak,v_Qk, v_dAk, v_dQk
def reset(self, compute_derivatives):
"""
For reusing this object e.g. in smoother computation. Actually,
this object can not be reused because it computes the matrices on
every iteration. But this method is written for keeping the same
interface with the class AQcompute_batch.
"""
self.last_k = 0
self.last_k_computed = False
self.compute_derivatives = compute_derivatives
self.Q_square_root_computed = False
return self
def Ak(self,k,m,P):
v_Ak,v_Qk, v_dAk, v_dQk = self._recompute_for_new_k(k)
return v_Ak
def Qk(self,k):
v_Ak,v_Qk, v_dAk, v_dQk = self._recompute_for_new_k(k)
return v_Qk
def dAk(self, k):
v_Ak,v_Qk, v_dAk, v_dQk = self._recompute_for_new_k(k)
return v_dAk
def dQk(self, k):
v_Ak,v_Qk, v_dAk, v_dQk = self._recompute_for_new_k(k)
return v_dQk
def Q_srk(self,k):
"""
Square root of the noise matrix Q
"""
if ((self.last_k == k) and (self.last_k_computed == True)):
if not self.Q_square_root_computed:
(U, S, Vh) = sp.linalg.svd( self.v_Qk, full_matrices=False, compute_uv=True, overwrite_a=False, check_finite=False)
square_root = U * np.sqrt(S)
self.square_root_computed = True
self.Q_square_root = square_root
else:
square_root = self.Q_square_root
else:
raise ValueError("Square root of Q can not be computed")
return square_root
def return_last(self):
"""
Function returns last computed matrices.
"""
if not self.last_k_computed:
raise ValueError("Matrices are not computed.")
else:
k = self.last_k
A = self.v_Ak
Q = self.v_Qk
dA = self.v_dAk
dQ = self.v_dQk
return k, A, Q, dA, dQ
class AQcompute_batch_Python(Q_handling_Class):
"""
Class for calculating matrices A, Q, dA, dQ of the discrete Kalman Filter
from the matrices F, L, Qc, P_ing, dF, dQc, dP_inf of the continuos state
equation. dt - time steps.
It has the same interface as AQcompute_once.
It computes matrices for all time steps. This object is used when
there are not so many (controlled by internal variable)
different time steps and storing all the matrices do not take too much memory.
Since all the matrices are computed all together, this object can be used
in smoother without repeating the computations.
"""
def __init__(self, F,L,Qc,dt,compute_derivatives=False, grad_params_no=None, P_inf=None, dP_inf=None, dF = None, dQc=None):
"""
Constructor. All necessary parameters are passed here and stored
in the opject.
Input:
-------------------
F, L, Qc, P_inf : matrices
Parameters of corresponding continuous state model
dt: array
All time steps
compute_derivatives: bool
Whether to calculate derivatives
dP_inf, dF, dQc: 3D array
Derivatives if they are required
Output:
-------------------
Nothing
"""
As, Qs, reconstruct_indices, dAs, dQs = ContDescrStateSpace.lti_sde_to_descrete(F,
L,Qc,dt,compute_derivatives,
grad_params_no=grad_params_no, P_inf=P_inf, dP_inf=dP_inf, dF=dF, dQc=dQc)
self.As = As
self.Qs = Qs
self.dAs = dAs
self.dQs = dQs
self.reconstruct_indices = reconstruct_indices
self.total_size_of_data = self.As.nbytes + self.Qs.nbytes +\
(self.dAs.nbytes if (self.dAs is not None) else 0) +\
(self.dQs.nbytes if (self.dQs is not None) else 0) +\
(self.reconstruct_indices.nbytes if (self.reconstruct_indices is not None) else 0)
self.Q_svd_dict = {}
self.last_k = None
# !!!Print statistics! Which object is created
# !!!Print statistics! Print sizes of matrices
def f_a(self, k,m,A):
"""
Dynamic model
"""
return np.dot(A, m) # default dynamic model
def reset(self, compute_derivatives=False):
"""
For reusing this object e.g. in smoother computation. It makes sence
because necessary matrices have been already computed for all
time steps.
"""
return self
def Ak(self,k,m,P):
self.last_k = k
return self.As[:,:, self.reconstruct_indices[k]]
def Qk(self,k):
self.last_k = k
return self.Qs[:,:, self.reconstruct_indices[k]]
def dAk(self,k):
self.last_k = k
return self.dAs[:,:, :, self.reconstruct_indices[k]]
def dQk(self,k):
self.last_k = k
return self.dQs[:,:, :, self.reconstruct_indices[k]]
def Q_srk(self,k):
"""
Square root of the noise matrix Q
"""
matrix_index = self.reconstruct_indices[k]
if matrix_index in self.Q_svd_dict:
square_root = self.Q_svd_dict[matrix_index]
else:
(U, S, Vh) = sp.linalg.svd( self.Qs[:,:, matrix_index],
full_matrices=False, compute_uv=True,
overwrite_a=False, check_finite=False)
square_root = U * np.sqrt(S)
self.Q_svd_dict[matrix_index] = square_root
return square_root
def return_last(self):
"""
Function returns last available matrices.
"""
if (self.last_k is None):
raise ValueError("Matrices are not computed.")
else:
ind = self.reconstruct_indices[self.last_k]
A = self.As[:,:, ind]
Q = self.Qs[:,:, ind]
dA = self.dAs[:,:, :, ind]
dQ = self.dQs[:,:, :, ind]
return self.last_k, A, Q, dA, dQ
@classmethod
def cont_discr_kalman_filter(cls, F, L, Qc, p_H, p_R, P_inf, X, Y, index = None,
m_init=None, P_init=None,
p_kalman_filter_type='regular',
calc_log_likelihood=False,
calc_grad_log_likelihood=False,
grad_params_no=0, grad_calc_params=None):
"""
This function implements the continuous-discrete Kalman Filter algorithm
These notations for the State-Space model are assumed:
d/dt x(t) = F * x(t) + L * w(t); w(t) ~ N(0, Qc)
y_{k} = H_{k} * x_{k} + r_{k}; r_{k-1} ~ N(0, R_{k})
Returns estimated filter distributions x_{k} ~ N(m_{k}, P(k))
Current Features:
----------------------------------------
1) The function generaly do not modify the passed parameters. If
it happens then it is an error. There are several exeprions: scalars
can be modified into a matrix, in some rare cases shapes of
the derivatives matrices may be changed, it is ignored for now.
2) Copies of F,L,Qc are created in memory because they may be used later
in smoother. References to copies are kept in "AQcomp" object
return parameter.
3) Function support "multiple time series mode" which means that exactly
the same State-Space model is used to filter several sets of measurements.
In this case third dimension of Y should include these state-space measurements
Log_likelihood and Grad_log_likelihood have the corresponding dimensions then.
4) Calculation of Grad_log_likelihood is not supported if matrices
H, or R changes overf time (with index k). (later may be changed)
5) Measurement may include missing values. In this case update step is
not done for this measurement. (later may be changed)
Input:
-----------------
F: (state_dim, state_dim) matrix
F in the model.
L: (state_dim, noise_dim) matrix
L in the model.
Qc: (noise_dim, noise_dim) matrix
Q_c in the model.
p_H: scalar, matrix (measurement_dim, state_dim) , 3D array
H_{k} in the model. If matrix then H_{k} = H - constant.
If it is 3D array then H_{k} = p_Q[:,:, index[2,k]]
p_R: scalar, square symmetric matrix, 3D array
R_{k} in the model. If matrix then R_{k} = R - constant.
If it is 3D array then R_{k} = p_R[:,:, index[3,k]]
P_inf: (state_dim, state_dim) matrix
State varince matrix on infinity.
X: 1D array
Time points of measurements. Needed for converting continuos
problem to the discrete one.
Y: matrix or vector or 3D array
Data. If Y is matrix then samples are along 0-th dimension and
features along the 1-st. If 3D array then third dimension
correspond to "multiple time series mode".
index: vector
Which indices (on 3-rd dimension) from arrays p_H, p_R to use
on every time step. If this parameter is None then it is assumed
that p_H, p_R do not change over time and indices are not needed.
index[0,:] - correspond to H, index[1,:] - correspond to R
If index.shape[0] == 1, it is assumed that indides for all matrices
are the same.
m_init: vector or matrix
Initial distribution mean. If None it is assumed to be zero.
For "multiple time series mode" it is matrix, second dimension of
which correspond to different time series. In regular case ("one
time series mode") it is a vector.
P_init: square symmetric matrix or scalar
Initial covariance of the states. If the parameter is scalar
then it is assumed that initial covariance matrix is unit matrix
multiplied by this scalar. If None the unit matrix is used instead.
"multiple time series mode" does not affect it, since it does not
affect anything related to state variaces.
p_kalman_filter_type: string, one of ('regular', 'svd')
Which Kalman Filter is used. Regular or SVD. SVD is more numerically
stable, in particular, Covariace matrices are guarantied to be
positive semi-definite. However, 'svd' works slower, especially for
small data due to SVD call overhead.
calc_log_likelihood: boolean
Whether to calculate marginal likelihood of the state-space model.
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then "grad_calc_params" parameter must
provide the extra parameters for gradient calculation.
grad_params_no: int
If previous parameter is true, then this parameters gives the
total number of parameters in the gradient.
grad_calc_params: dictionary
Dictionary with derivatives of model matrices with respect
to parameters "dF", "dL", "dQc", "dH", "dR", "dm_init", "dP_init".
They can be None, in this case zero matrices (no dependence on parameters)
is assumed. If there is only one parameter then third dimension is
automatically added.
Output:
--------------
M: (no_steps+1,state_dim) matrix or (no_steps+1,state_dim, time_series_no) 3D array
Filter estimates of the state means. In the extra step the initial
value is included. In the "multiple time series mode" third dimension
correspond to different timeseries.
P: (no_steps+1, state_dim, state_dim) 3D array
Filter estimates of the state covariances. In the extra step the initial
value is included.
log_likelihood: double or (1, time_series_no) 3D array.
If the parameter calc_log_likelihood was set to true, return
logarithm of marginal likelihood of the state-space model. If
the parameter was false, return None. In the "multiple time series mode" it is a vector
providing log_likelihood for each time series.
grad_log_likelihood: column vector or (grad_params_no, time_series_no) matrix
If calc_grad_log_likelihood is true, return gradient of log likelihood
with respect to parameters. It returns it column wise, so in
"multiple time series mode" gradients for each time series is in the
corresponding column.
AQcomp: object
Contains some pre-computed values for converting continuos model into
discrete one. It can be used later in the smoothing pahse.
"""
p_H = np.atleast_1d(p_H)
p_R = np.atleast_1d(p_R)
X.shape, old_X_shape = cls._reshape_input_data(X.shape, 2) # represent as column
if (X.shape[1] != 1):
raise ValueError("Only one dimensional X data is supported.")
Y.shape, old_Y_shape = cls._reshape_input_data(Y.shape) # represent as column
state_dim = F.shape[0]
measurement_dim = Y.shape[1]
time_series_no = Y.shape[2] # multiple time series mode
if ((len(p_H.shape) == 3) and (len(p_H.shape[2]) != 1)) or\
((len(p_R.shape) == 3) and (len(p_R.shape[2]) != 1)):
model_matrices_chage_with_time = True
else:
model_matrices_chage_with_time = False
# Check index
old_index_shape = None
if index is None:
if (len(p_H.shape) == 3) or (len(p_R.shape) == 3):
raise ValueError("Parameter index can not be None for time varying matrices (third dimension is present)")
else: # matrices do not change in time, so form dummy zero indices.
index = np.zeros((1,Y.shape[0]))
else:
if len(index.shape) == 1:
index.shape = (1,index.shape[0])
old_index_shape = (index.shape[0],)
if (index.shape[1] != Y.shape[0]):
raise ValueError("Number of measurements must be equal the number of H_{k}, R_{k}")
if (index.shape[0] == 1):
H_time_var_index = 0; R_time_var_index = 0
elif (index.shape[0] == 4):
H_time_var_index = 0; R_time_var_index = 1
else:
raise ValueError("First Dimension of index must be either 1 or 2.")
(p_H, old_H_shape) = cls._check_SS_matrix(p_H, state_dim, measurement_dim, which='H')
(p_R, old_R_shape) = cls._check_SS_matrix(p_R, state_dim, measurement_dim, which='R')
if m_init is None:
m_init = np.zeros((state_dim, time_series_no))
else:
m_init = np.atleast_2d(m_init).T
if P_init is None:
P_init = P_inf.copy()
if p_kalman_filter_type not in ('regular', 'svd'):
raise ValueError("Kalman filer type neither 'regular nor 'svd'.")
# Functions to pass to the kalman_filter algorithm:
# Parameters:
# k - number of Kalman filter iteration
# m - vector for calculating matrices. Required for EKF. Not used here.
# f_hl = lambda k,m,H: np.dot(H, m)
# f_H = lambda k,m,P: p_H[:,:, index[H_time_var_index, k]]
#f_R = lambda k: p_R[:,:, index[R_time_var_index, k]]
#o_R = R_handling( p_R, index, R_time_var_index, 20)
if calc_grad_log_likelihood:
dF = cls._check_grad_state_matrices(grad_calc_params.get('dF'), state_dim, grad_params_no, which = 'dA')
dQc = cls._check_grad_state_matrices(grad_calc_params.get('dQc'), state_dim, grad_params_no, which = 'dQ')
dP_inf = cls._check_grad_state_matrices(grad_calc_params.get('dP_inf'), state_dim, grad_params_no, which = 'dA')
dH = cls._check_grad_measurement_matrices(grad_calc_params.get('dH'), state_dim, grad_params_no, measurement_dim, which = 'dH')
dR = cls._check_grad_measurement_matrices(grad_calc_params.get('dR'), state_dim, grad_params_no, measurement_dim, which = 'dR')
dm_init = grad_calc_params.get('dm_init') # Initial values for the Kalman Filter
if dm_init is None:
# multiple time series mode. Keep grad_params always as a last dimension
dm_init = np.zeros( (state_dim, time_series_no, grad_params_no) )
dP_init = grad_calc_params.get('dP_init') # Initial values for the Kalman Filter
if dP_init is None:
dP_init = dP_inf(0).copy() # get the dP_init matrix, because now it is a function
else:
dP_inf = None
dF = None
dQc = None
dH = None
dR = None
dm_init = None
dP_init = None
measurement_callables = Std_Measurement_Callables_Class(p_H, H_time_var_index, p_R, index, R_time_var_index, 20, dH, dR)
#import pdb; pdb.set_trace()
dynamic_callables = cls._cont_to_discrete_object(X, F, L, Qc, compute_derivatives=calc_grad_log_likelihood,
grad_params_no=grad_params_no,
P_inf=P_inf, dP_inf=dP_inf, dF = dF, dQc=dQc)
if print_verbose:
print("General: run Continuos-Discrete Kalman Filter")
# Also for dH, dR and probably for all derivatives
(M, P, log_likelihood, grad_log_likelihood, AQcomp) = cls._cont_discr_kalman_filter_raw(state_dim,
dynamic_callables, measurement_callables,
X, Y, m_init=m_init, P_init=P_init,
p_kalman_filter_type=p_kalman_filter_type,
calc_log_likelihood=calc_log_likelihood,
calc_grad_log_likelihood=calc_grad_log_likelihood, grad_params_no=grad_params_no,
dm_init=dm_init, dP_init=dP_init)
if old_index_shape is not None:
index.shape = old_index_shape
if old_X_shape is not None:
X.shape = old_X_shape
if old_Y_shape is not None:
Y.shape = old_Y_shape
if old_H_shape is not None:
p_H.shape = old_H_shape
if old_R_shape is not None:
p_R.shape = old_R_shape
return (M, P, log_likelihood, grad_log_likelihood, AQcomp)
@classmethod
def _cont_discr_kalman_filter_raw(cls,state_dim, p_dynamic_callables, p_measurement_callables, X, Y,
m_init, P_init,
p_kalman_filter_type='regular',
calc_log_likelihood=False,
calc_grad_log_likelihood=False, grad_params_no=None,
dm_init=None, dP_init=None):
"""
General filtering algorithm for inference in the continuos-discrete
state-space model:
d/dt x(t) = F * x(t) + L * w(t); w(t) ~ N(0, Qc)
y_{k} = H_{k} * x_{k} + r_{k}; r_{k-1} ~ N(0, R_{k})
Returns estimated filter distributions x_{k} ~ N(m_{k}, P(k))
Current Features:
----------------------------------------
1) Function support "multiple time series mode" which means that exactly
the same State-Space model is used to filter several sets of measurements.
In this case third dimension of Y should include these state-space measurements
Log_likelihood and Grad_log_likelihood have the corresponding dimensions then.
2) Measurement may include missing values. In this case update step is
not done for this measurement. (later may be changed)
Input:
-----------------
state_dim: int
Demensionality of the states
F: (state_dim, state_dim) matrix
F in the model.
L: (state_dim, noise_dim) matrix
L in the model.
Qc: (noise_dim, noise_dim) matrix
Q_c in the model.
P_inf: (state_dim, state_dim) matrix
State varince matrix on infinity.
p_h: function (k, x_{k}, H_{k}). Measurement function.
k (iteration number),
x_{k}
H_{k} Jacobian matrices of f_h. In the linear case it is exactly H_{k}.
f_H: function (k, m, P) return Jacobian of dynamic function, it is
passed into p_h.
k (iteration number),
m: point where Jacobian is evaluated,
P: parameter for Jacobian, usually covariance matrix.
p_f_R: function (k). Returns noise matrix of measurement equation
on iteration k.
k (iteration number).
m_init: vector or matrix
Initial distribution mean. For "multiple time series mode"
it is matrix, second dimension of which correspond to different
time series. In regular case ("one time series mode") it is a
vector.
P_init: matrix or scalar
Initial covariance of the states. Must be not None
"multiple time series mode" does not affect it, since it does not
affect anything related to state variaces.
p_kalman_filter_type: string, one of ('regular', 'svd')
Which Kalman Filter is used. Regular or SVD. SVD is more numerically
stable, in particular, Covariace matrices are guarantied to be
positive semi-definite. However, 'svd' works slower, especially for
small data due to SVD call overhead.
calc_log_likelihood: boolean
Whether to calculate marginal likelihood of the state-space model.
calc_grad_log_likelihood: boolean
Whether to calculate gradient of the marginal likelihood
of the state-space model. If true then the next parameter must
provide the extra parameters for gradient calculation.
grad_params_no: int
Number of gradient parameters
dP_inf, dF, dQc, dH, dR, dm_init, dP_init: matrices or 3D arrays.
Necessary parameters for derivatives calculation.
"""
#import pdb; pdb.set_trace()
steps_no = Y.shape[0] # number of steps in the Kalman Filter
time_series_no = Y.shape[2] # multiple time series mode
# Allocate space for results
# Mean estimations. Initial values will be included
M = np.empty(((steps_no+1),state_dim,time_series_no))
M[0,:,:] = m_init # Initialize mean values
# Variance estimations. Initial values will be included
P = np.empty(((steps_no+1),state_dim,state_dim))
P_init = 0.5*( P_init + P_init.T) # symmetrize initial covariance. In some ustable cases this is uiseful
P[0,:,:] = P_init # Initialize initial covariance matrix
#import pdb;pdb.set_trace()
if p_kalman_filter_type == 'svd':
(U,S,Vh) = sp.linalg.svd( P_init,full_matrices=False, compute_uv=True,
overwrite_a=False,check_finite=True)
S[ (S==0) ] = 1e-17 # allows to run algorithm for singular initial variance
P_upd = (P_init, S,U)
#log_likelihood = 0
#grad_log_likelihood = np.zeros((grad_params_no,1))
log_likelihood = 0 if calc_log_likelihood else None
grad_log_likelihood = 0 if calc_grad_log_likelihood else None
#setting initial values for derivatives update
dm_upd = dm_init
dP_upd = dP_init
# Main loop of the Kalman filter
for k in range(0,steps_no):
# In this loop index for new estimations is (k+1), old - (k)
# This happened because initial values are stored at 0-th index.
#import pdb; pdb.set_trace()
prev_mean = M[k,:,:] # mean from the previous step
if p_kalman_filter_type == 'svd':
m_pred, P_pred, dm_pred, dP_pred = \
cls._kalman_prediction_step_SVD(k, prev_mean ,P_upd, p_dynamic_callables,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_upd, p_dP = dP_upd)
else:
m_pred, P_pred, dm_pred, dP_pred = \
cls._kalman_prediction_step(k, prev_mean ,P[k,:,:], p_dynamic_callables,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_upd, p_dP = dP_upd )
#import pdb; pdb.set_trace()
k_measurment = Y[k,:,:]
if (np.any(np.isnan(k_measurment)) == False):
if p_kalman_filter_type == 'svd':
m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update = \
cls._kalman_update_step_SVD(k, m_pred , P_pred, p_measurement_callables,
k_measurment, calc_log_likelihood=calc_log_likelihood,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_pred, p_dP = dP_pred )
# m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update = \
# cls._kalman_update_step(k, m_pred , P_pred[0], f_h, f_H, p_R.f_R, k_measurment,
# calc_log_likelihood=calc_log_likelihood,
# calc_grad_log_likelihood=calc_grad_log_likelihood,
# p_dm = dm_pred, p_dP = dP_pred, grad_calc_params_2 = (dH, dR))
#
# (U,S,Vh) = sp.linalg.svd( P_upd,full_matrices=False, compute_uv=True,
# overwrite_a=False,check_finite=True)
# P_upd = (P_upd, S,U)
else:
m_upd, P_upd, log_likelihood_update, dm_upd, dP_upd, d_log_likelihood_update = \
cls._kalman_update_step(k, m_pred , P_pred, p_measurement_callables, k_measurment,
calc_log_likelihood=calc_log_likelihood,
calc_grad_log_likelihood=calc_grad_log_likelihood,
p_dm = dm_pred, p_dP = dP_pred )
else:
if k_measurment.shape != (1,1):
raise ValueError("Nan measurements are currently not supported for \
multidimensional output and multiple tiem series.")
else:
m_upd = m_pred; P_upd = P_pred; dm_upd = dm_pred; dP_upd = dP_pred
log_likelihood_update = 0.0;
d_log_likelihood_update = 0.0;
if calc_log_likelihood:
log_likelihood += log_likelihood_update
if calc_grad_log_likelihood:
grad_log_likelihood += d_log_likelihood_update
M[k+1,:,:] = m_upd # separate mean value for each time series
if p_kalman_filter_type == 'svd':
P[k+1,:,:] = P_upd[0]
else:
P[k+1,:,:] = P_upd
#print("kf it: %i" % k)
# !!!Print statistics! Print sizes of matrices
# !!!Print statistics! Print iteration time base on another boolean variable
return (M, P, log_likelihood, grad_log_likelihood, p_dynamic_callables.reset(False))
@classmethod
def cont_discr_rts_smoother(cls,state_dim, filter_means, filter_covars,
p_dynamic_callables=None, X=None, F=None,L=None,Qc=None):
"""
Continuos-discrete Rauch–Tung–Striebel(RTS) smoother.
This function implements Rauch–Tung–Striebel(RTS) smoother algorithm
based on the results of _cont_discr_kalman_filter_raw.
Model:
d/dt x(t) = F * x(t) + L * w(t); w(t) ~ N(0, Qc)
y_{k} = H_{k} * x_{k} + r_{k}; r_{k-1} ~ N(0, R_{k})
Returns estimated smoother distributions x_{k} ~ N(m_{k}, P(k))
Input:
--------------
filter_means: (no_steps+1,state_dim) matrix or (no_steps+1,state_dim, time_series_no) 3D array
Results of the Kalman Filter means estimation.
filter_covars: (no_steps+1, state_dim, state_dim) 3D array
Results of the Kalman Filter covariance estimation.
Dynamic_callables: object or None
Object form the filter phase which provides functions for computing
A, Q, dA, dQ fro discrete model from the continuos model.
X, F, L, Qc: matrices
If AQcomp is None, these matrices are used to create this object from scratch.
Output:
-------------
M: (no_steps+1,state_dim) matrix
Smoothed estimates of the state means
P: (no_steps+1,state_dim, state_dim) 3D array
Smoothed estimates of the state covariances
"""
f_a = lambda k,m,A: np.dot(A, m) # state dynamic model
if p_dynamic_callables is None: # make this object from scratch
p_dynamic_callables = cls._cont_to_discrete_object(cls, X, F,L,Qc,f_a,compute_derivatives=False,
grad_params_no=None, P_inf=None, dP_inf=None, dF = None, dQc=None)
no_steps = filter_covars.shape[0]-1# number of steps (minus initial covariance)
M = np.empty(filter_means.shape) # smoothed means
P = np.empty(filter_covars.shape) # smoothed covars
if print_verbose:
print("General: run Continuos-Discrete Kalman Smoother")
M[-1,:,:] = filter_means[-1,:,:]
P[-1,:,:] = filter_covars[-1,:,:]
for k in range(no_steps-1,-1,-1):
prev_mean = filter_means[k,:] # mean from the previous step
m_pred, P_pred, tmp1, tmp2 = \
cls._kalman_prediction_step(k, prev_mean,
filter_covars[k,:,:], p_dynamic_callables,
calc_grad_log_likelihood=False)
p_m = filter_means[k,:]
p_m_prev_step = M[(k+1),:]
m_upd, P_upd, tmp_G = cls._rts_smoother_update_step(k,
p_m ,filter_covars[k,:,:],
m_pred, P_pred, p_m_prev_step ,P[(k+1),:,:], p_dynamic_callables)
M[k,:,:] = m_upd
P[k,:,:] = P_upd
# Return values
return (M, P)
@classmethod
def _cont_to_discrete_object(cls, X, F, L, Qc, compute_derivatives=False,
grad_params_no=None,
P_inf=None, dP_inf=None, dF = None, dQc=None):
"""
Function return the object which is used in Kalman filter and/or
smoother to obtain matrices A, Q and their derivatives for discrete model
from the continuous model.
There are 2 objects AQcompute_once and AQcompute_batch and the function
returs the appropriate one based on the number of different time steps.
Input:
----------------------
X, F, L, Qc: matrices
Continuous model matrices
f_a: function
Dynamic Function is attached to the Dynamic_Model_Callables class
compute_derivatives: boolean
Whether to compute derivatives
grad_params_no: int
Number of parameters in the gradient
P_inf, dP_inf, dF, dQ: matrices and 3D objects
Data necessary to compute derivatives.
Output:
--------------------------
AQcomp: object
Its methods return matrices (and optionally derivatives) for the
discrete state-space model.
"""
unique_round_decimals = 10
threshold_number_of_unique_time_steps = 20 # above which matrices are separately each time
dt = np.empty((X.shape[0],))
dt[1:] = np.diff(X[:,0],axis=0)
dt[0] = 0#dt[1]
unique_indices = np.unique(np.round(dt, decimals=unique_round_decimals))
number_unique_indices = len(unique_indices)
#import pdb; pdb.set_trace()
if use_cython:
class AQcompute_batch(state_space_cython.AQcompute_batch_Cython):
def __init__(self, F,L,Qc,dt,compute_derivatives=False, grad_params_no=None, P_inf=None, dP_inf=None, dF = None, dQc=None):
As, Qs, reconstruct_indices, dAs, dQs = ContDescrStateSpace.lti_sde_to_descrete(F,
L,Qc,dt,compute_derivatives,
grad_params_no=grad_params_no, P_inf=P_inf, dP_inf=dP_inf, dF=dF, dQc=dQc)
super(AQcompute_batch,self).__init__(As, Qs, reconstruct_indices, dAs,dQs)
else:
AQcompute_batch = cls.AQcompute_batch_Python
if number_unique_indices > threshold_number_of_unique_time_steps:
AQcomp = cls.AQcompute_once(F,L,Qc, dt,compute_derivatives=compute_derivatives,
grad_params_no=grad_params_no, P_inf=P_inf, dP_inf=dP_inf, dF=dF, dQc=dQc)
if print_verbose:
print("CDO: Continue-to-discrete INSTANTANEOUS object is created.")
print("CDO: Number of different time steps: %i" % (number_unique_indices,) )
else:
AQcomp = AQcompute_batch(F,L,Qc,dt,compute_derivatives=compute_derivatives,
grad_params_no=grad_params_no, P_inf=P_inf, dP_inf=dP_inf, dF=dF, dQc=dQc)
if print_verbose:
print("CDO: Continue-to-discrete BATCH object is created.")
print("CDO: Number of different time steps: %i" % (number_unique_indices,) )
print("CDO: Total size if its data: %i" % (AQcomp.total_size_of_data,) )
return AQcomp
@staticmethod
def lti_sde_to_descrete(F,L,Qc,dt,compute_derivatives=False,
grad_params_no=None, P_inf=None,
dP_inf=None, dF = None, dQc=None):
"""
Linear Time-Invariant Stochastic Differential Equation (LTI SDE):
dx(t) = F x(t) dt + L d \beta ,where
x(t): (vector) stochastic process
\beta: (vector) Brownian motion process
F, L: (time invariant) matrices of corresponding dimensions
Qc: covariance of noise.
This function rewrites it into the corresponding state-space form:
x_{k} = A_{k} * x_{k-1} + q_{k-1}; q_{k-1} ~ N(0, Q_{k-1})
Input:
--------------
F,L: LTI SDE matrices of corresponding dimensions
Qc: matrix (n,n)
Covarince between different dimensions of noise \beta.
n is the dimensionality of the noise.
dt: double or iterable
Time difference used on this iteration.
If dt is iterable, then A and Q_noise are computed for every
unique dt
compute_derivatives: boolean
Whether derivatives of A and Q are required.
grad_params_no: int
Number of gradient parameters
P_inf: (state_dim. state_dim) matrix
dP_inf
dF: 3D array
Derivatives of F
dQc: 3D array
Derivatives of Qc
dR: 3D array
Derivatives of R
Output:
--------------
A: matrix
A_{k}. Because we have LTI SDE only dt can affect on matrix
difference for different k.
Q_noise: matrix
Covariance matrix of (vector) q_{k-1}. Only dt can affect the
matrix difference for different k.
reconstruct_index: array
If dt was iterable return three dimensinal arrays A and Q_noise.
Third dimension of these arrays correspond to unique dt's.
This reconstruct_index contain indices of the original dt's
in the uninue dt sequence. A[:,:, reconstruct_index[5]]
is matrix A of 6-th(indices start from zero) dt in the original
sequence.
dA: 3D array
Derivatives of A
dQ: 3D array
Derivatives of Q
"""
# Dimensionality
n = F.shape[0]
if not isinstance(dt, collections.Iterable): # not iterable, scalar
# The dynamical model
A = matrix_exponent(F*dt)
if np.any( np.isnan(A)):
A = linalg.expm3(F*dt)
# The covariance matrix Q by matrix fraction decomposition ->
Phi = np.zeros((2*n,2*n))
Phi[:n,:n] = F
Phi[:n,n:] = L.dot(Qc).dot(L.T)
Phi[n:,n:] = -F.T
AB = matrix_exponent(Phi*dt)
AB = np.dot(AB, np.vstack((np.zeros((n,n)),np.eye(n))))
Q_noise_1 = linalg.solve(AB[n:,:].T,AB[:n,:].T)
Q_noise_2 = P_inf - A.dot(P_inf).dot(A.T)
# The covariance matrix Q by matrix fraction decomposition <-
if compute_derivatives:
dA = np.zeros([n, n, grad_params_no])
dQ = np.zeros([n, n, grad_params_no])
#AA = np.zeros([2*n, 2*n, nparam])
FF = np.zeros([2*n, 2*n])
AA = np.zeros([2*n, 2*n, grad_params_no])
for p in range(0, grad_params_no):
FF[:n,:n] = F
FF[n:,:n] = dF[:,:,p]
FF[n:,n:] = F
# Solve the matrix exponential
AA[:,:,p] = matrix_exponent(FF*dt)
# Solve the differential equation
#foo = AA[:,:,p].dot(np.vstack([m, dm[:,p]]))
#mm = foo[:n,:]
#dm[:,p] = foo[n:,:]
# The discrete-time dynamical model*
if p==0:
A = AA[:n,:n,p]
Q_noise_2 = P_inf - A.dot(P_inf).dot(A.T)
Q_noise = Q_noise_2
#PP = A.dot(P).dot(A.T) + Q_noise_2
# The derivatives of A and Q
dA[:,:,p] = AA[n:,:n,p]
dQ[:,:,p] = dP_inf[:,:,p] - dA[:,:,p].dot(P_inf).dot(A.T) \
- A.dot(dP_inf[:,:,p]).dot(A.T) - A.dot(P_inf).dot(dA[:,:,p].T) # Rewrite not ro multiply two times
else:
dA = None
dQ = None
Q_noise = Q_noise_2
# Innacuracies have been observed when Q_noise_1 was used.
#Q_noise = Q_noise_1
# Return
return A, Q_noise,None, dA, dQ
else: # iterable, array
# Time discretizations (round to 14 decimals to avoid problems)
dt_unique, tmp, reconstruct_index = np.unique(np.round(dt,8),
return_index=True,return_inverse=True)
del tmp
# Allocate space for A and Q
A = np.empty((n,n,dt_unique.shape[0]))
Q_noise = np.empty((n,n,dt_unique.shape[0]))
if compute_derivatives:
dA = np.empty((n,n,grad_params_no,dt_unique.shape[0]))
dQ = np.empty((n,n,grad_params_no,dt_unique.shape[0]))
else:
dA = None
dQ = None
# Call this function for each unique dt
for j in range(0,dt_unique.shape[0]):
A[:,:,j], Q_noise[:,:,j], tmp1, dA_t, dQ_t = ContDescrStateSpace.lti_sde_to_descrete(F,L,Qc,dt_unique[j],
compute_derivatives=compute_derivatives, grad_params_no=grad_params_no, P_inf=P_inf, dP_inf=dP_inf, dF = dF, dQc=dQc)
if compute_derivatives:
dA[:,:,:,j] = dA_t
dQ[:,:,:,j] = dQ_t
# Return
return A, Q_noise, reconstruct_index, dA, dQ
def matrix_exponent(M):
"""
The function computes matrix exponent and handles some special cases
"""
if (M.shape[0] == 1): # 1*1 matrix
Mexp = np.array( ((np.exp(M[0,0]) ,),) )
else: # matrix is larger
method = None
try:
Mexp = linalg.expm(M)
method = 1
except (Exception,) as e:
Mexp = linalg.expm3(M)
method = 2
finally:
if np.any(np.isnan(Mexp)):
if method == 2:
raise ValueError("Matrix Exponent is not computed 1")
else:
Mexp = linalg.expm3(M)
method = 2
if np.any(np.isnan(Mexp)):
raise ValueError("Matrix Exponent is not computed 2")
return Mexp
def balance_matrix(A):
"""
Balance matrix, i.e. finds such similarity transformation of the original
matrix A: A = T * bA * T^{-1}, where norms of columns of bA and of rows of bA
are as close as possible. It is usually used as a preprocessing step in
eigenvalue calculation routine. It is useful also for State-Space models.
See also:
[1] Beresford N. Parlett and Christian Reinsch (1969). Balancing
a matrix for calculation of eigenvalues and eigenvectors.
Numerische Mathematik, 13(4): 293-304.
Input:
----------------------
A: square matrix
Matrix to be balanced
Output:
----------------
bA: matrix
Balanced matrix
T: matrix
Left part of the similarity transformation
T_inv: matrix
Right part of the similarity transformation.
"""
if len(A.shape) != 2 or (A.shape[0] != A.shape[1]):
raise ValueError('balance_matrix: Expecting square matrix')
N = A.shape[0] # matrix size
gebal = sp.linalg.lapack.get_lapack_funcs('gebal',(A,))
bA, lo, hi, pivscale, info = gebal(A, permute=True, scale=True,overwrite_a=False)
if info < 0:
raise ValueError('balance_matrix: Illegal value in %d-th argument of internal gebal ' % -info)
# calculating the similarity transforamtion:
def perm_matr(D, c1,c2):
"""
Function creates the permutation matrix which swaps columns c1 and c2.
Input:
--------------
D: int
Size of the permutation matrix
c1: int
Column 1. Numeration starts from 1...D
c2: int
Column 2. Numeration starts from 1...D
"""
i1 = c1-1; i2 = c2-1 # indices
P = np.eye(D);
P[i1,i1] = 0.0; P[i2,i2] = 0.0; # nullify diagonal elements
P[i1,i2] = 1.0; P[i2,i1] = 1.0
return P
P = np.eye(N) # permutation matrix
if (hi != N-1): # there are row permutations
for k in range(N-1,hi,-1):
new_perm = perm_matr(N, k+1, pivscale[k])
P = np.dot(P,new_perm)
if (lo != 0):
for k in range(0,lo,1):
new_perm = perm_matr(N, k+1, pivscale[k])
P = np.dot(P,new_perm)
D = pivscale.copy()
D[0:lo] = 1.0; D[hi+1:N] = 1.0 # thesee scaling factors must be set to one.
#D = np.diag(D) # make a diagonal matrix
T = np.dot(P,np.diag(D)) # similarity transformation in question
T_inv = np.dot(np.diag(D**(-1)),P.T)
#print( np.max(A - np.dot(T, np.dot(bA, T_inv) )) )
return bA.copy(), T, T_inv
def balance_ss_model(F,L,Qc,H,Pinf,P0,dF=None,dQc=None,dPinf=None,dP0=None):
"""
Balances State-Space model for more numerical stability
This is based on the following:
dx/dt = F x + L w
y = H x
Let T z = x, which gives
dz/dt = inv(T) F T z + inv(T) L w
y = H T z
"""
bF,T,T_inv = balance_matrix(F)
bL = np.dot( T_inv, L)
bQc = Qc # not affected
bH = np.dot(H, T)
bPinf = np.dot(T_inv, np.dot(Pinf, T_inv.T))
#import pdb; pdb.set_trace()
# LL,islower = linalg.cho_factor(Pinf)
# inds = np.triu_indices(Pinf.shape[0],k=1)
# LL[inds] = 0.0
# bLL = np.dot(T_inv, LL)
# bPinf = np.dot( bLL, bLL.T)
bP0 = np.dot(T_inv, np.dot(P0, T_inv.T))
if dF is not None:
bdF = np.zeros(dF.shape)
for i in range(dF.shape[2]):
bdF[:,:,i] = np.dot( T_inv, np.dot( dF[:,:,i], T))
else:
bdF = None
if dPinf is not None:
bdPinf = np.zeros(dPinf.shape)
for i in range(dPinf.shape[2]):
bdPinf[:,:,i] = np.dot( T_inv, np.dot( dPinf[:,:,i], T_inv.T))
# LL,islower = linalg.cho_factor(dPinf[:,:,i])
# inds = np.triu_indices(dPinf[:,:,i].shape[0],k=1)
# LL[inds] = 0.0
# bLL = np.dot(T_inv, LL)
# bdPinf[:,:,i] = np.dot( bLL, bLL.T)
else:
bdPinf = None
if dP0 is not None:
bdP0 = np.zeros(dP0.shape)
for i in range(dP0.shape[2]):
bdP0[:,:,i] = np.dot( T_inv, np.dot( dP0[:,:,i], T_inv.T))
else:
bdP0 = None
bdQc = dQc # not affected
# (F,L,Qc,H,Pinf,P0,dF,dQc,dPinf,dP0)
return bF, bL, bQc, bH, bPinf, bP0, bdF, bdQc, bdPinf, bdP0, T
Reference in a new issue View git blame Copy permalink