juancamilog

import numpy as np
import theano
import theano.tensor as T
from theano.compile import function
from theano.tensor.nlinalg import matrix_inverse
from theano.tensor.slinalg import solve
from theano.tensor.nlinalg import det
from theano.tensor.slinalg import cholesky
from theano.ifelse import ifelse
from theano.gradient import verify_grad

juancamilog

juancamilog@juancamilog:~/workspace/antoinette$ gdb python jacobian_segfault.py 
GNU gdb (Ubuntu 7.7.1-0ubuntu5~14.04.2) 7.7.1
Copyright (C) 2014 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:

juancamilog

import numpy as np
import theano
import theano.tensor as T

from theano import function as F, shared as S
from theano.sandbox.linalg import psd
from theano.tensor.nlinalg import matrix_inverse,det
from time import time

if __name__=='__main__':

juancamilog

import theano
import theano.tensor as T
from theano.sandbox.linalg import psd,matrix_inverse,det,cholesky
import numpy as np
from time import time

def SEard(loghyp,X):
    ''' Squared exponential kernel with diagonal scaling matrix (one lengthscale per dimension)'''
    N,idims = X.shape

juancamilog

import theano
import theano.tensor as T
from theano.sandbox.linalg import psd,matrix_inverse,det,cholesky
import numpy as np
from time import time

def SEard(loghyp,X):
    ''' Squared exponential kernel with diagonal scaling matrix (one lengthscale per dimension)'''
    N,idims = X.shape

juancamilog

import theano
import theano.tensor as T
from theano.sandbox.linalg import psd,matrix_inverse,det,cholesky
import numpy as np
from time import time

def SEard(loghyp,X):
    ''' Squared exponential kernel with diagonal scaling matrix (one lengthscale per dimension)'''
    N,idims = X.shape

juancamilog

import theano
import theano.tensor as T
from theano.sandbox.linalg import psd,matrix_inverse,det,cholesky
from theano.tensor.slinalg import solve_lower_triangular
import numpy as np
from time import time

def SEard(loghyp,X):
    ''' Squared exponential kernel with diagonal scaling matrix (one lengthscale per dimension)'''
    N,idims = X.shape

juancamilog

@register_stabilize
@local_optimizer([Dot, Dot22])
def inv_as_solve(node):
    if not imported_scipy:
        return False
    if isinstance(node.op, (Dot, Dot22)):
        l, r = node.inputs
        if l.owner and l.owner.op == matrix_inverse:

            if CHECK l.owner.inputs[0] is PSD HERE:

juancamilog

import theano
import numpy as np

# inti vars
s = theano.tensor.matrix('s')
y = theano.tensor.vector('y')
r = theano.tensor.slinalg.Solve()(s, y)

# jacobian using scan
dr1 = theano.tensor.jacobian(r.flatten(), s)

juancamilog