Greater parralellization of dirty_map

[kiyo-masui]

Not yet ready to merge.

@cjanderson23 Take a look at my changes to this function and see if you understand how this could be used to provide the threading your looking for.

[cjanderson23]

Is it okay if I merge this?

[kiyo-masui]

Not until you have a working feature. Checkout this branch and develop the feature in this branch. Then merge it in when the feature is complete. Going forward, we will be using the following branching model: http://scottchacon.com/2011/08/31/github-flow.html

[cjanderson23]

Okay. I'd like to work on this branch and either create a new parallel_dirty_map_thread.py or modify parallel_dirty_map.py to take a threading_subdivide input boolean. Since this branch doesn't have parallel_dirty_map.py, to get it should I do something like 'git checkout merge_all parallel_dirty_map.py' and then commit?

[kiyo-masui]

Merge merge_all into this branch. I can do that actually. One sec.

Kiyo

On Tue, Feb 11, 2014 at 2:26 PM, cjanderson23 notifications@github.comwrote:

Okay. I'd like to work on this branch and either create a new
parallel_dirty_map_thread.py or modify parallel_dirty_map.py to take a
threading_subdivide input boolean. Since this branch doesn't have
parallel_dirty_map.py, to get it should I do something like 'git checkout
merge_all parallel_dirty_map.py' and then commit?

Reply to this email directly or view it on GitHubhttps://github.com//pull/5#issuecomment-34816031
.

[cjanderson23]

Okay, I wrote parallel_dirty_map_tcs.py, which subdivides the chunks of the covariance by ra and distributes the work to the threads. I tested it on a small map subsection, and the noise covariance is identical to that produced by parallel_dirty_map.py. I also had to change dirty_map.py, but it should be backwards compatible. I think it's ready to merge if you approve.

[kiyo-masui]

Looks like parallel_dirty_map_tcs.py and parallel_dirty_map.py only differ by a few lines. Do you really need two programs or can you merge them into one to make them differ in the parameters they take?

This looks really good by the way. I'm glad you managed to implement this so quickly.

[cjanderson23]

Yeah, you're right. A single file would make more sense. Let me make sure I understand how to do it correctly. I think I do this by adding some key like thread_divide with a default value (probably False) to the params_init dictionary and then the .ini or .pipe file sets the value by assigning dm_thread_divide= True or False. Is that right?

[kiyo-masui]

Yes that is correct.

On Thu, Feb 13, 2014 at 3:31 PM, cjanderson23 notifications@github.comwrote:

Yeah, you're right. A single file would make more sense. Let me make sure
I understand how to do it correctly. I think I do this by adding some key
like thread_divide with a default value (probably False) to the params_init
dictionary and then the .ini or .pipe file sets the value by assigning
dm_thread_divide= True or False. Is that right?

Reply to this email directly or view it on GitHubhttps://github.com//pull/5#issuecomment-35039595
.