slight changes to allow non-writing subchains in Piccard, and added more precision

[vhaasteren]

Added an option to not write the chain to file, and changed the _AMbuffer to be of longer length than the chain we are running at this point.

Added precision to the files that are written.

[jellis18]

Hi Rutger

I will take a look at this. Although I'm making a lot of changes to make
the sampler more modular. I'll incorporate the changes you want soon but I
need to do some testing. Although I think I'm pretty much done with a very
general rjmcmc!

Justin

Sent from my iPhone

On Mar 31, 2014, at 5:06 PM, Rutger notifications@github.com wrote:

Added an option to not write the chain to file, and changed the _AMbuffer
to be of longer length than the chain we are running at this point.

Added precision to the files that are written.

You can merge this Pull Request by running

git pull https://github.com/vhaasteren/PAL2 master

Or view, comment on, or merge it at:

#1
Commit Summary

• slight changes to allow non-writing subchains in Piccard

File Changes

• M PAL2/PALInferencePTMCMC.pyhttps://github.com/slight changes to allow non-writing subchains in Piccard, and added more precision #1/files#diff-0(19)

Patch Links:

• https://github.com/jellis18/PAL2/pull/1.patch
• https://github.com/jellis18/PAL2/pull/1.diff

Reply to this email directly or view it on
GitHubhttps://github.com//pull/1
.

[vhaasteren]

Hi Justin,

That's cool, take your time. My changes were very modest, as you'll see.
Just a thing renamed, and two or three if-clauses. I understand that you
must have changed a lot in the past few days, but if the outside interface
for regular MCMCs has not changed much I suspect it to be easy.

So you are good on track with the RJMCMC?

On Mon, Mar 31, 2014 at 3:11 PM, Justin Ellis notifications@github.comwrote:

Hi Rutger

I will take a look at this. Although I'm making a lot of changes to make
the sampler more modular. I'll incorporate the changes you want soon but I
need to do some testing. Although I think I'm pretty much done with a very
general rjmcmc!

Justin

Sent from my iPhone

On Mar 31, 2014, at 5:06 PM, Rutger notifications@github.com wrote:

Added an option to not write the chain to file, and changed the _AMbuffer
to be of longer length than the chain we are running at this point.

Added precision to the files that are written.

You can merge this Pull Request by running

git pull https://github.com/vhaasteren/PAL2 master

Or view, comment on, or merge it at:

#1
Commit Summary

• slight changes to allow non-writing subchains in Piccard

File Changes

• M PAL2/PALInferencePTMCMC.py<
  https://github.com/jellis18/PAL2/pull/1/files#diff-0>(19)

Patch Links:

• https://github.com/jellis18/PAL2/pull/1.patch
• https://github.com/jellis18/PAL2/pull/1.diff

Reply to this email directly or view it on
GitHubhttps://github.com//pull/1
.

Reply to this email directly or view it on GitHubhttps://github.com//pull/1#issuecomment-39148751
.

---

dr. Rutger van Haasteren, Einstein postdoctoral fellow
Jet Propulsion Laboratory, Pasadena, USA
E-mail: vhaasteren@gmail.com

[jellis18]

Hi Rutger

The interface is exactly the same just may chunks within the sample method
are now separate methods or functions that can be called outside of the
sample method.

Yes I have it all coded up now I just have to test it. I also think I found
a way to include jitter in the full GWB likelihood function without it
taking forever. I still have to code it up though.

Justin

Sent from my iPhone

On Mar 31, 2014, at 5:20 PM, Rutger notifications@github.com wrote:

Hi Justin,

That's cool, take your time. My changes were very modest, as you'll see.
Just a thing renamed, and two or three if-clauses. I understand that you
must have changed a lot in the past few days, but if the outside interface
for regular MCMCs has not changed much I suspect it to be easy.

So you are good on track with the RJMCMC?

On Mon, Mar 31, 2014 at 3:11 PM, Justin Ellis <notifications@github.com

wrote:

Hi Rutger

I will take a look at this. Although I'm making a lot of changes to make
the sampler more modular. I'll incorporate the changes you want soon but I
need to do some testing. Although I think I'm pretty much done with a very
general rjmcmc!

Justin

Sent from my iPhone

On Mar 31, 2014, at 5:06 PM, Rutger notifications@github.com wrote:

Added an option to not write the chain to file, and changed the _AMbuffer
to be of longer length than the chain we are running at this point.

Added precision to the files that are written.

You can merge this Pull Request by running

git pull https://github.com/vhaasteren/PAL2 master

Or view, comment on, or merge it at:

#1
Commit Summary

• slight changes to allow non-writing subchains in Piccard

File Changes

• M PAL2/PALInferencePTMCMC.py<
  https://github.com/jellis18/PAL2/pull/1/files#diff-0>(19)

Patch Links:

• https://github.com/jellis18/PAL2/pull/1.patch
• https://github.com/jellis18/PAL2/pull/1.diff

Reply to this email directly or view it on
GitHubhttps://github.com//pull/1
.

Reply to this email directly or view it on GitHub<
https://github.com/jellis18/PAL2/pull/1#issuecomment-39148751>
.

---

dr. Rutger van Haasteren, Einstein postdoctoral fellow
Jet Propulsion Laboratory, Pasadena, USA
E-mail: vhaasteren@gmail.com

Reply to this email directly or view it on
GitHubhttps://github.com//pull/1#issuecomment-39149550
.

[vhaasteren]

Yes I have it all coded up now I just have to test it.

Cool, looking forward to it!

I also think I found
a way to include jitter in the full GWB likelihood function without it
taking forever. I still have to code it up though.

Anything fancy? I have been trying to include it in the diagonal noise
matrix 'N', which then becomes block-diagonal. In theory that should be
very fast, but my Python implementation takes forever, even if you use a
Woodbury expansion for the blocks. I was thinking of writing a C-callback
function just for that. Even in the Gibbs sampler, the Jitter is proving to
be a PITA

---

dr. Rutger van Haasteren, Einstein postdoctoral fellow
Jet Propulsion Laboratory, Pasadena, USA
E-mail: vhaasteren@gmail.com

[jellis18]

I'm away from my computer now but in essence you can just write the
covariance with jitter but not coarse graining the red noise. I wrote it
out yesterday an I think it should be faster than what I had before by a
factor of q/p)^3 where q is the number of epochs and p is the number of
frequencies. This is about a factor of 1000!

Sent from my iPhone

On Mar 31, 2014, at 5:27 PM, Rutger notifications@github.com wrote:

Yes I have it all coded up now I just have to test it.

Cool, looking forward to it!

I also think I found
a way to include jitter in the full GWB likelihood function without it
taking forever. I still have to code it up though.

Anything fancy? I have been trying to include it in the diagonal noise
matrix 'N', which then becomes block-diagonal. In theory that should be
very fast, but my Python implementation takes forever, even if you use a
Woodbury expansion for the blocks. I was thinking of writing a C-callback
function just for that. Even in the Gibbs sampler, the Jitter is proving to
be a PITA

---

dr. Rutger van Haasteren, Einstein postdoctoral fellow
Jet Propulsion Laboratory, Pasadena, USA
E-mail: vhaasteren@gmail.com

Reply to this email directly or view it on
GitHubhttps://github.com//pull/1#issuecomment-39150204
.

[vhaasteren]

Hmm, we'll see

On Mon, Mar 31, 2014 at 3:48 PM, Justin Ellis notifications@github.comwrote:

I'm away from my computer now but in essence you can just write the
covariance with jitter but not coarse graining the red noise. I wrote it
out yesterday an I think it should be faster than what I had before by a
factor of q/p)^3 where q is the number of epochs and p is the number of
frequencies. This is about a factor of 1000!

Sent from my iPhone

On Mar 31, 2014, at 5:27 PM, Rutger notifications@github.com wrote:

Yes I have it all coded up now I just have to test it.

Cool, looking forward to it!

I also think I found
a way to include jitter in the full GWB likelihood function without it
taking forever. I still have to code it up though.

Anything fancy? I have been trying to include it in the diagonal noise
matrix 'N', which then becomes block-diagonal. In theory that should be
very fast, but my Python implementation takes forever, even if you use a
Woodbury expansion for the blocks. I was thinking of writing a C-callback
function just for that. Even in the Gibbs sampler, the Jitter is proving to
be a PITA

---

dr. Rutger van Haasteren, Einstein postdoctoral fellow
Jet Propulsion Laboratory, Pasadena, USA
E-mail: vhaasteren@gmail.com

Reply to this email directly or view it on
GitHubhttps://github.com//pull/1#issuecomment-39150204

.

Reply to this email directly or view it on GitHubhttps://github.com//pull/1#issuecomment-39152006
.

---

dr. Rutger van Haasteren, Einstein postdoctoral fellow
Jet Propulsion Laboratory, Pasadena, USA
E-mail: vhaasteren@gmail.com