Release History¶
0.23.1 - Bugfix release¶
Bugfixes¶
Fix issue where if
None
was used foronline_analysis_interval
an error would be thrown (issue #708 PR `#710 <https://github.com/choderalab/openmmtools/pull/710`_)
0.23.0 - latest numba support and real time stats enhancements¶
Please note that there is an API breaking change. To ensure consistency of the data when appending real time stats make sure that you make the online_analysis_interval
of your MultiStateSampler
object match the checkpoint_interval
of your MultiStateReporter
. It will error if this is not the case.
Enhancements¶
Bugfixes¶
Fix metadata for netcdf files, specifying openmmtools for the
program
metadata (issue #694, PR #704).Real time statistics YAML file gets appended instead of overwritten when extending or resumimng simulations (issue #691, PR #692).
Error when resuming simulations with numba 0.57 fixed by avoiding using
numpy.MaskedArray
when deserializing.nc
files (issue #700, PR #701)
0.22.1 - Bugfix release¶
Bugfixes¶
Fixed issue where the error message thrown from openMM changed, so we need a case insensitive check. This was already fixed in most of the code base but one spot was missed. (PR #684)
0.22.0 - pymbar 4 support and gentle equilibration¶
Enhancements¶
Bugfixes¶
Users were not able to distinguish the exceptions caught during dynamics. Warnings are now raised when an exception is being caught (Issue #643 PR #658).
Deserializing MCMC moves objects from versions <=0.21.4 resulted in error finding the key. Fixed by catching the exception and raising a warning when key is not found (Issue #618 PR #675).
Different improvements in documentation strings and readthedocs documentation generation (Issues #620 #641 #548. PR #676)
Support for newer NetCDF versions (1.6 branch) by not using zlib compression for varying length variables. (PR #654).
0.21.5 - Bugfix release¶
Changed behaviors¶
Bugfixes¶
Velocities were being incorrectly updated as zeros when resuming simulations or broadcasting from different mpi processes. Fixed by specifying
ignore_velocities=False
in_propagate_replica
. Issue #531 (PR #602).Bug in equilibration detection #1: The user was allowed to specify
statistical_inefficiency
without specifyingn_equilibration_iterations
, which doesn’t make sense, asn_equilibration_iterations
andn_effective_max
cannot be computed fromstatistical_inefficiency
alone. Fixed by preventing user from specifyingstatistical_inefficiency
withoutn_equilibration_iterations
. Issue #609 (PR #610).Bug in equilibration detection #2: If the user specified
n_equilibration_iterations
but notstatistical_inefficiency
, the returnedn_equilibration_iterations
did not include number of equilibration iterations as computed from_get_equilibration_data_per_sample()
. Fixed by always including the_get_equilibration_data_per_sample()
result in in the returnedn_equilibration_iterations
. Issue #609 (PR #610).Bug in equilibration detection #3:
get_equilibration_data_per_sample
returns 0 forn_equilibration_iterations
. Fixed by always discarding the first time origin returned byget_equilibration_data_per_sample
. To control the amount of data discarded by the first time origin, the user can now specifymax_subset
when initializingMultiStateSamplerAnalyzer
. Issue #609 (PR #610).Deserializing simulations from
openmmtools<0.21.3
versions resulted in error. Fixed by catching the missing key,KeyError
exception, when deserializing. Issue #612, PR #613.Not specifying a subdirectory for the reporter file resulted in
PermissionError
when writing the real time analysis file. Fixed by usingos.path.join
for creating the output paths. Issue #615, PR #616.
Enhancements¶
LangevinDynamicsMove
now allowsconstraint_tolerance
parameter and public attribute, for specifying the fraction of the constrained distance within which constraints are maintained for the integrator (Refer to Openmm’s documentation for more information). Issue #608, PR #611.Platform is now reported in the logs in DEBUG mode. Issue #583, PR #605.
0.21.4 - Bugfix release¶
Bugfixes¶
Bug in statistical inefficiency computation – where self.max_n_iterations wasn’t being used – was fixed (#577).
Bug in estimated performance in realtime yaml file – fixed by iterating through all MCMC moves (#578)
Potential bug fixed by explicitly updating and broadcasting thermodynamic states in replicas, when used in an MPI (distributed) context. Issue #579 (#587).
Bug in handling unsampled states in realtime/offline analysis – fixed by using
MultiStateSampler._unsampled_states
to build the mbar estimate array. Issue #592 (#593)
Enhancements¶
0.21.3 - Bugfix release¶
Bugfixes¶
Bug in replica mixing in MPI multi-GPU runs–where some replicas were simulated in incorrect states–was fixed (#449) & (#562).
Velocities are now stored in the checkpoint file to eliminate issue with “cold restart”. Fixes issue #531 (#555).
Documentation now correctly builds via CI. Fixes issue #548 (#554).
Enhancements¶
Real time MBAR analysis and timing information is now produced in yaml format at user-specified intervals (#565), (#561) & (#572).
Information of what CUDA devices are available is now provided in log output (#570).
Replica exchanges are now attempted during equilibration phase to enhance mixing (#556).
An example of resuming a MultiStateSampler simulation using API is now provided (#569)
0.21.2 - Bugfix release¶
Bugfixes¶
Fixed UnboundLocalError when using a string to specify platform in
platform_supports_precision
(#551).
0.21.1 - Bugfix release¶
Bugfixes¶
More streamlined context cache usage using instance attributes (#547).
Improved docstring and examples for
MultiStateSampler
object.
0.21.0 - Bugfix release¶
Bugfixes¶
Fixes TestAbsoluteAlchemicalFactory.test_overlap NaNs (#534)
Try closing reporter in test for windows fix (#535)
Follow NEP 29 and test newer python versions and drop old ones (#542)
Update to handle the new OpenMM 7.6 package namespace (#528)
Context cache usage cleanup (#538 <https://github.com/choderalab/openmmtools/pull/538>_). Avoiding memory issues and more streamlined API usage of ContextCache objects.
Known issues¶
Correctly raises an error when a
CustomNonbondedForce
made by OpenMM’sLennardJonesGenerator
is detected (#511)
Enhancement¶
0.20.3 - Bugfix release¶
Bugfixes¶
Fixes [#505](https://github.com/choderalab/openmmtools/issues/505): GPU contexts would silently fail to enable ‘mixed’ precision; corrects reporting of available precisions
0.20.2 - Bugfix release¶
Remove leftover support for python 2.7
Cleanup¶
Remove leftover six imports and xrange (#504)
0.20.1 - Bugfix release¶
Enhancements¶
openmmtools.utils.get_available_platforms()
and.get_fastest_platform()
now filter OpenMM Platforms based on specified minimum precision support, which defaults tomixed
Bugfixes¶
Misc¶
Resolve
numpy 1.20
DeprecationWarning
aboutnp.float
0.20.0 - Periodic alchemical integrators¶
Enhancements¶
Add PeriodicNonequilibriumIntegrator, a simple extension of AlchemicalNonequilibriumLangevinIntegrator that supports periodic alchemical protocols
0.19.1 - Bugfix release¶
Bugfixes¶
Fixed a crash during the restraint unbiasing for systems with an unexpected order of atoms of receptor and ligands (#462).
0.19.0 - Multiple alchemical regions¶
New features¶
Added support in
AbsoluteAlchemicalFactory
for handling multiple independent alchemical regions (#438).Added support for anisotropic and membrane barostats in ThermodynamicState (#437).
Added support for platform properties in ContextCache (e.g. for mixed and double precision CUDA in multistate sampler) (#437).
Bugfixes¶
Known issues¶
Using parallel MPI processes causes poor mixing of the odd thermodynamic states while the mixing of the even states is normal. We’re still investigating whether the issue is caused by a change in the MPI library or an internal bug. For now, we recommend running calculations using only 1 GPU (see also #449 and yank#1130).
0.18.3 - Storage enhancements and bugfixes¶
Bugfixes¶
Enhancements¶
Writing on disk is much faster when the checkpoint_interval of multi-state samplers is large. This was due to the dimension of the netcdf chunk size increasing with the checkpoint interval and surpassing the dimension of the netcdf chunk cache. The chunk size of the iteration dimension is now always set to 1 (#432).
0.18.2 - Bugfix release¶
Bugfixes¶
A bug in the multistate samplers where``logsumexp`` was imported from
scipy.misc
(now inscipy.special
) was fixed (#423).Improve the robustness of opening the netcdf file on resuming of the multi-state samplers by setting the environment variable HDF5_USE_FILE_LOCKING to FALSE after 4 failed attempts (#426).
Fixed a crash during exception handling (#426).
Other¶
Update build infrastructure to match MolSSI cookiecutter (#424, #426).
0.18.1 - Bugfix release¶
This is a minor bugfix release.
New features¶
Improvements for
HostGuest*
classes - addoemols
,host_oemol
, andguest_oemol
properties to retrieve OpenEye ToolkitOEMol
objects (requires toolkit license and installation) - these classes can now accept overridingkwargs
Bugfixes¶
openmmtools.multistate
experimental API warning is only issued whenopenmmtools.multistate
is importedAlchemicalNonequilibriumLangevinIntegrator.reset()
now correctly resets the nonequilibrium work
0.18.0 - Added multistate samplers¶
New features¶
Add a number of classes that can use MCMC to sample from multiple thermodynamic states: -
MultiStateSampler
: sample independently from multiple thermodynamic states -ReplicaExchangeSampler
: replica exchange among thermodynamic states -SAMSSampler
: self-adjusted mixture sampling (SAMS) samplingAll samplers can use MPI via the mpiplus package
0.17.0 - Removed Py2 support, faster exact PME treatment¶
New features¶
Add
GlobalParameterFunction
that allows to enslave aGlobalParameter
to an arbitrary function of controlling variables (#380).Allow to ignore velocities when building the dict representation of a
SamplerState
. This can be useful for example to save bandwidth when sending aSamplerState
over the network and velocities are not required (#386).Add
DoubleWellDimer_WCAFluid
andDoubleWellChain_WCAFluid
test systems (#389).
Enhancements¶
New implementation of the exact PME handling that uses the parameter offset feature in OpenMM 7.3. This comes with a
considerable speed improvement over the previous implementation (#380).
- Exact PME is now the default for the alchemical_pme_treatment
parameter in the constructor of
AbsoluteAchemicalFactory
(#386).
- It is now possible to have multiple composable states exposing the same attributes/getter/setter in a
CompoundThermodynamicState
(#380).
Bug fixes¶
Fixed a bug involving the
NoseHooverChainVelocityVerletIntegrator
withSystem
with constraints. The constraints were not taken into account when calculating the number of degrees of freedom resulting in the temperature not converging to the target value. (#384)Fixed a bug affecting
reduced_potential_at_states
when computing the reduced potential of systems in different ``AlchemicalState``s when the same alchemical parameter appeared in force objects split in different force groups. (#385)
Deprecated and API breaks¶
Python 2 and 3.5 is not supported anymore.
The
update_alchemical_charges
attribute of ``AlchemicalState`, which was deprecated in 0.16.0, has now been removed since it doesn’t make sense with the new parameter offset implementation.The methods
AlchemicalState.get_alchemical_variable
andAlchemicalState.set_alchemical_variable
have been deprecated. UseAlchemicalState.get_alchemical_function
andAlchemicalState.set_alchemical_function
instead.
0.16.0 - Py2 deprecated, GlobalParameterState class, SamplerState reads CVs¶
New features¶
Add ability for
SamplerState
to access new OpenMM Custom CV Force Variables (#362).SamplerState.update_from_context
now has keywords to support finer grain updating from the Context. This is only recommended for advanced users (#362).Added the new class
states.GlobalParameterState
designed to simplify the implementation of composable states that control global variables (#363).Allow restraint force classes to be controlled by a parameter other than
lambda_restraints
. This will enable multi-restraints simulations (#363).
Enhancements¶
Global variables of integrators are now automatically copied over the integrator returned by
ContextCache.get_context
. It is possible to specify exception throughContextCache.INCOMPATIBLE_INTEGRATOR_ATTRIBUTES
(#364).
Others¶
Integrator
MCMCMove``s now attempt to recover from NaN automatically by default (with ``n_restart_attempts
set to 4) (#364).
Deprecated¶
Python2 is officially deprecated. Support will be dropped in future versions.
Deprecated the signature of
IComposableState._on_setattr
to fix a bug where the objects were temporarily left in an inconsistent state when an exception was raised and caught.Deprecated
update_alchemical_charges
inAlchemicalState
in anticipation of the new implementation of the exact PME that will be based on theNonbondedForce
offsets rather thanupdateParametersInContext()
.
0.15.0 - Restraint forces¶
0.14.0 - Exact treatment of alchemical PME electrostatics, water cluster test system, optimizations¶
New features¶
Enhancements¶
Bug fixes¶
Copy thermodynamic state on compound state initialization. (#320)
0.13.4 - Barostat/External Force Bugfix, Restart Robustness¶
Bug fixes¶
Fixed implementation bug where
CustomExternalForce
restraining atoms to absolute coordinates caused an issue when a Barostat was used (#310)
Enhancements¶
MCMC Integrators now attempt to re-initialize the
Context
object on the last restart attempt when NaN’s are encountered. This has internally been shown to correct some instances where normally resetting positions does not work around the NaN’s. This is a slow step relative to just resetting positions, but better than simulation crashing.
0.13.3 - Critical Bugfix to SamplerState Context Manipulation¶
Critical Fixes¶
SamplerState.apply_to_context()
applies box vectors before positions are set to prevent a bug on non-Reference OpenMM Platforms which can re-order system atoms. (#305)
Additional Fixes¶
0.13.2 - SamplerState Slicing and BitWise And/Or Ops¶
Added support for SamplerState slicing (#298)
Added bit operators and
and or
to math_eval
(#301)
0.13.1 - Bugfix release¶
0.13.0 - Alternative reaction field models, Langevin splitting MCMCMove¶
New Features¶
Storage Interface module with automatic disk IO handling
Option for shifted or switched Reaction Field
LangevinSplittingDynamic
MCMC move with specifiable sub step orderingNose-Hoover Chain Thermostat
Bug Fixes¶
Many doc string cleanups
Tests are based on released versions of OpenMM
Tests also compare against development OpenMM, but do not fail because of it
Fixed bug in Harmonic Oscillator tests’ error calculation
Default collision rate in Langevin Integrators now matches docs
0.12.1 - Add virtual sites support in alchemy¶
0.12.0 - GB support in alchemy and new forces module¶
New features¶
Add AbsoluteAlchemicalFactory support for all GB models (#250)
Added
forces
andforcefactories
modules implementingUnishiftedReactionFieldForce
andreplace_reaction_field
respectively. The latter has been moved fromAbsoluteAlchemicalFactory
(#253)Add
restrain_atoms
to restrain molecule conformation through an harmonic restrain (#255)
Bug fixes¶
0.11.2 - Bugfix release¶
Hotfix in fringe Python2/3 compatibility issue when using old style serialization systems in Python 2
0.11.1 - Optimizations¶
0.11.0 - Conda forge installation¶
New Features¶
LangevinIntegrator
now setsmeasure_heat=False
by default for increased performance (#211)AbsoluteAlchemicalFactory
now supportsdisable_alchemical_dispersion_correction
to prevent 600x slowdowns with nonequilibrium integration (#218)We now require conda-forge as a dependency for testing and deployment (#216)
Conda-forge added as channel to conda packages
0.10.0 - Optimizations of ThermodynamicState, renamed AlchemicalFactory¶
BREAKS API: Renamed AlchemicalFactory to AbsoluteAlchemicalFactory (#206)
Major optimizations of ThermodynamicState (#200, #205)
Keep in memory only a single System object per compatible state
Fast copy/deepcopy
Enable custom optimized serialization for multiple states
Added readthedocs documentation (#191)
Bugfix for serialization of context when NaN encountered (#199)
Added tests for Python 3.6 (#184)
0.9.4 - Nonequilibrium integrators overhaul¶
Major changes¶
Overhaul of
LangevinIntegrator
and subclasses to better support nonequilibrium integratorsAdd true reaction-field support to
AlchemicalFactory
Add some alchemical test systems
Updates to openmmtools.integrators.LangevinIntegrator
and friends¶
API-breaking changes¶
The nonequilibrium integrators are now called
AlchemicalNonequilibriumLangevinIntegrator
andExternalPerturbationLangevinIntegrator
, and both are subclasses of a commonNonequilibriumLangevinIntegrator
that provides a consistent interface to setting and gettingprotocol_work
AlchemicalNonequilibriumLangevinIntegrator
now has a defaultalchemical_functions
to eliminate need for every test to treat it as a special case (#180)The
get_protocol_work()
method allows you to retrieve the protocol work from anyNonequilibriumLangevinIntegrator
subclass and returns a unit-bearing work. The optionaldimensionless=True
argument returns a dimensionless float in units of kT.Integrator global variables now store all energies in natural OpenMM units (kJ/mol) but the new accessor methods (see below) should b used instead of getting integrator global variables for work and heat. (#181)
Any private methods for adding steps to the integrator have been prepended with
_
to hide them from the public API.
New features¶
Order of arguments for all
LangevinIntegrator
derivatives matchesopenmm.LangevinIntegrator
so it can act as a drop-in replacement. (#176)The
get_shadow_work()
andget_heat()
methods are now available for anyLangevinIntegrator
subclass, as well as the corresponding propertiesshadow_work
and heat. The functions also supportdimensionless=True.
(#163)The
shadow_work
andheat
properties were added to all LangevinIntegrator subclasses, returning the values of these properties (if the integrator was constructed with the appropriatemeasure_shadow_work=True
ormeasure_heat=True
flags) as unit-bearing quantitiesThe
get_protocol_work()
andget_total_work()
methods are now available for anyNonequilibriumLangevinIntegrator
, returning unit-bearing quantities unlessdimensionless=True
is provided in which case they return the work in implicit units of kT.get_total_work()
requires the integrator to have been constructed withmeasure_shadow_work=True
.The
protocol_work
andtotal_work
properties were added to allNonequilibriumLangevinIntegrator
subclasses, and return the unit-bearing work quantities.total_work
requires the integrator to have been constructed withmeasure_shadow_work=True
.The subclasses have been reworked to support any kwargs that the base classes support, and defaults have all been made consistent.
Various reset() methods have been added to reset statistics for all
LangevinIntegrator
subclasses.All custom integrators support
.pretty_format()
and.pretty_print()
with optional highlighting of specific step types.
Bugfixes¶
Zero-step perturbations now work correctly (#177)
AlchemicalNonequilibriumLangevinIntegrator
now correctly supports multipleH
steps.
Internal changes¶
Adding new LangevinIntegrator step methods now uses a
self._register_step_method(step_string, callback_function, supports_force_groups=False)
call to simplify this process.Code duplication has been reduced through the use of calling base class methods whenever possible.
run_nonequilibrium_switching()
test now uses BAR to test dragging a harmonic oscillator and tests a variety of integrator splittings(["O { V R H R V } O", "O V R H R V O", "R V O H O V R", "H R V O V R H"])
.Integrator tests use deterministic PME and mixed precision when able.
Updates to openmmtools.alchemy.AlchemicalFactory¶
Reaction field electrostatics now removes the shift, setting
c_rf = 0
.A convenience method AlchemicalFactory.replace_reaction_field() has been added to allow fully-interacting systems to be modified to force
c_rf = 0
by recoding reaction-field electrostatics as aCustomNonbondedForce
New openmmtools.testsystems
classes¶
AlchemicalWaterBox was added, which has the first water molecule in the system alchemically modified