You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
add function mdivide_left_spd and mdivide_right_spd #22
To expose them, we have to decide how to deal with checking
that the input truly is symmetric and positive definite. That's
probably not necessary for nested uses.
I need you to give me instances of these matrices:
matrix A is not symmetric
matrix A is not positive definite
matrix A is not square
matrix A that is ok; matrix b that is not compatible with A
Just pick these things out of a hat. I don't care what they are, as long as you provide the actual instances of the matrices.
Are there any other conditions where we should stop A?
NaN?
Inf?
-Inf?
zero size?
Are there any conditions where matrix b should fail?
NaN?
Inf?
-Inf?
zero size?
There are going to be issues just based on inspection. The check_pos_definite comes before checking for a square matrix. I don't know how safe that function is.
OK, I can do that. If there are any NaN or +/- Inf, it should fail the
check_pos_definite but we may want to have check_pos_definite catch those
specifically. This is another case where we want to compute the
Eigen::LLT<> and then check it, rather than doing a Cholesky decomposition
to see if we can do the same Cholesky decomposition.
I need you to give me instances of these matrices:
matrix A is not symmetric
matrix A is not positive definite
matrix A is not square
matrix A that is ok; matrix b that is not compatible with A Just
pick these things out of a hat. I don't care what they are, as long as you
provide the actual instances of the matrices.
Are there any other conditions where we should stop A?
NaN?
Inf? -Inf?
zero size?
Are there any conditions where matrix b should fail?
NaN?
Inf?
-Inf?
zero size?
There are going to be issues just based on inspection. The
check_pos_definite comes before checking for a square matrix. I don't know
how safe that function is.
The branch Rob just finished up today has two new function signatures for
check_pos_definite: one for LDLT and one for LLT. We can definitely use
that.
By the way, it's not safe to assume that our checks catch NaN or that Eigen
stops with NaN values.
OK, I can do that. If there are any NaN or +/- Inf, it should fail the
check_pos_definite but we may want to have check_pos_definite catch those
specifically. This is another case where we want to compute the
Eigen::LLT<> and then check it, rather than doing a Cholesky decomposition
to see if we can do the same Cholesky decomposition.
I need you to give me instances of these matrices:
matrix A is not symmetric
matrix A is not positive definite
matrix A is not square
matrix A that is ok; matrix b that is not compatible with A Just
pick these things out of a hat. I don't care what they are, as long as
you
provide the actual instances of the matrices.
Are there any other conditions where we should stop A?
NaN?
Inf? -Inf?
zero size?
Are there any conditions where matrix b should fail?
NaN?
Inf?
-Inf?
zero size?
There are going to be issues just based on inspection. The
check_pos_definite comes before checking for a square matrix. I don't
know
how safe that function is.
The branch Rob just finished up today has two new function signatures for
check_pos_definite: one for LDLT and one for LLT. We can definitely use
that.
By the way, it's not safe to assume that our checks catch NaN or that Eigen
stops with NaN values.
OK, I can do that. If there are any NaN or +/- Inf, it should fail the
check_pos_definite but we may want to have check_pos_definite catch those
specifically. This is another case where we want to compute the
Eigen::LLT<> and then check it, rather than doing a Cholesky
decomposition
to see if we can do the same Cholesky decomposition.
I need you to give me instances of these matrices:
matrix A is not symmetric
matrix A is not positive definite
matrix A is not square
matrix A that is ok; matrix b that is not compatible with A Just
pick these things out of a hat. I don't care what they are, as long as
you
provide the actual instances of the matrices.
Are there any other conditions where we should stop A?
NaN?
Inf? -Inf?
zero size?
Are there any conditions where matrix b should fail?
NaN?
Inf?
-Inf?
zero size?
There are going to be issues just based on inspection. The
check_pos_definite comes before checking for a square matrix. I don't
know
how safe that function is.
Activity
syclik commentedon Jul 6, 2015
From @PeterLi2016 on July 30, 2014 17:1
Is there a reason these aren't exposed? They have some tests implemented.
syclik commentedon Jul 6, 2015
From @bob-carpenter on July 30, 2014 17:14
To expose them, we have to decide how to deal with checking
that the input truly is symmetric and positive definite. That's
probably not necessary for nested uses.
On Jul 30, 2014, at 1:01 PM, Peter Li notifications@github.com wrote:
syclik commentedon Jul 6, 2015
I'm suggesting we remove the implementation as part of #1407.
syclik commentedon Jul 6, 2015
@bgoodri, this is what I need from you:
I need you to give me instances of these matrices:
Just pick these things out of a hat. I don't care what they are, as long as you provide the actual instances of the matrices.
Are there any other conditions where we should stop A?
-Inf?
Are there any conditions where matrix b should fail?
There are going to be issues just based on inspection. The check_pos_definite comes before checking for a square matrix. I don't know how safe that function is.
syclik commentedon Jul 6, 2015
From @bgoodri on March 25, 2015 23:43
OK, I can do that. If there are any NaN or +/- Inf, it should fail the
check_pos_definite but we may want to have check_pos_definite catch those
specifically. This is another case where we want to compute the
Eigen::LLT<> and then check it, rather than doing a Cholesky decomposition
to see if we can do the same Cholesky decomposition.
On Wed, Mar 25, 2015 at 7:39 PM, Daniel Lee notifications@github.com
wrote:
syclik commentedon Jul 6, 2015
The branch Rob just finished up today has two new function signatures for
check_pos_definite: one for LDLT and one for LLT. We can definitely use
that.
By the way, it's not safe to assume that our checks catch NaN or that Eigen
stops with NaN values.
On Wed, Mar 25, 2015 at 7:43 PM, bgoodri notifications@github.com wrote:
syclik commentedon Jul 6, 2015
From @bgoodri on March 25, 2015 23:53
Eigen doesn't stop but the .diagonal().array() <= 0).any() check should
return false if either there are NaNs or divisions by Inf.
On Wed, Mar 25, 2015 at 7:46 PM, Daniel Lee notifications@github.com
wrote:
syclik commentedon Jul 6, 2015
@bgoodri, that statement is incorrect. Eigen's transforms swallows both nan and inf. Pop this into src/test/eigen_test.cpp:
the output I get:
syclik commentedon Jul 6, 2015
From @bgoodri on March 26, 2015 3:25
I see what you mean now. We need to change the check from "any
non-negative" to "not all positive". Like
!(ldlt.vectorD().array() > 0.0).all();
On Wed, Mar 25, 2015 at 10:33 PM, Daniel Lee notifications@github.com
wrote:
syclik commentedon Aug 3, 2016
@bgoodri, want to add these tests so we can get these functions exposed?
bob-carpenter commentedon Aug 17, 2016
These exist in the math lib, so I'm closing this issue an opening a stan-dev/stan issue: stan-dev/stan#2016
Merge pull request #22 from charlesm93/issue-21-rename