IMP Reference Guide
2.18.0
The Integrative Modeling Platform
|
News May 7 2016 - update 2
Here is a list of recent feature additions and bug fixes in PMI2:
–Charles
News May 7 2016
We have updated PMI significantly, completely rewriting the topology building tools (you may see the new code referred to as "PMI2". Instead of using the monolithic "representation" class (in representation.py) we are now using a set of modular classes with a bit more flexibility:
Changes in IMP that are helpful in PMI:
–Charles
News May 14 2014
From Ben Webb:
PMI was added as an IMP submodule. But what might not be obvious is that a submodule is tied to a specific revision (or git hash) so doesn't automatically update whenever PMI is updated. So far I've been updating PMI in IMP on a rather ad hoc basis (essentially whenever I make changes to PMI itself), but you might want to do this too when you add new features.
It's pretty easy: if you maintain PMI outside of IMP somewhere, just go into your IMP checkout, cd modules/pmi, then git checkout <githash>
where <githash> is the version of PMI you want to live in IMP. Then you can go back up into IMP proper and git commit/git push that change. Or you can do what I do and work on the copy of PMI that's in IMP directly rather than checking it out separately. To do that, go into modules/pmi in your IMP checkout and run git checkout develop
to get on the PMI develop branch. Then you can run all the normal git commands (e.g. git pull) in there. (Although note that you can't push by default; that's easy to fix though by running git remote set-url --push origin git@github.com:salilab/pmi.git
.) When you're done with your git push, you can cd up into IMP proper and commit the change in PMI githash there (without having to remember what it is).
News April 24 2014
The main branch is now develop
(not master
) and the module is included in IMP by default.
News February 28 2014
The clustering analysis calculates the localization densities
News February 25 2014
First working version of the clustering analysis. For the moment each cluster directory will only contain pdbs and rmfs of the structures, but soon I'll add statistics, features and localization densities.
To have full speed install the mpi4py python library
News February 10 2014
Checkout the macros in macros.py
News February 6 2014
Added a FAQ section in the wiki. There will be ipython notebook tutorials soon. Best IMP version so far: ee1763c6859a29ab37415c8454d16549268d9668
Previous versions had a bug in the RigidClosePairsFinder bug
News February 4 2014:
Best IMP version so far: b700fff93ca45ba1551c8aa9d697c805fcb126dc
Previous versions might be considerably slow due to a bug
News January 30 2014:
Now the developed git branch is master and not resolution-zero.
If you want to use pmi, after you've freshly cloned it, you don't have to checkout resolution-zero anymore: it is the default branch that you get when you clone it.
The resolution-zero branch does not exist anymore, it was copied into resolution-zero-old.
To see what branch you're in, run (into the pmi source code directory):
git branch
If you want to update the code and you still are in resolution-zero branch, just run :
git checkout master git pull
Note that the interface is also changing, so you'll probably get deprecated warnings in your standard output more and more. grep deprecated
to get the deprecation warnings, which might be lost in the middle of many other messages. Change your python script according to what the warnings say. The old version of pmi (mainly used by Peter and SJ) is still available under the tag "v0.1". To get it:
git checkout tags/v0.1