Skip to content

Latest commit

 

History

History
658 lines (484 loc) · 38.3 KB

releasehistory.rst

File metadata and controls

658 lines (484 loc) · 38.3 KB

Release History

0.24.1 - Differential storage of positions and velocities

Enhancements

  • MultiStateReporter now accepts variable position and velocity checkpointing intervals. Note that when resuming simulations users have to specify again the keyword arguments for the reporter (Pull Request #767).

0.24.0 - pyMBAR Behavior Changes + HIP Platform Added

Bug Fixes

Behavior Changes

  • Use robust solver for pyMBAR by default. pyMBAR 3 & 4 used two different solvers by default. We now use the robust solver by default regardless of the pyMBAR version. We still respect whichever solver is specified in analysis_kwargs (i.e analysis_kwargs["solver_protocol"] = "robust" when creating the analyzer, but now set the solver to "robust" if no solver is specified. This should improve convergence performance (Pull Request #735).

Enhancements

  • Added OpenMM's "HIP" platform as a selectable platform. With the release of OpenMM 8.2, the "HIP" platform is now available to use on compatible AMD GPUs. This update will allow openmmtools to automatically select the HIP platform if it is available (Pull Request #753).
  • Added effective_length to MultiStateSamplerAnalyzer (Pull Request #589).
  • Create alchemy subpackage (Pull Request #721).

Testing

0.23.1 - Bugfix release

Bugfixes

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

  • Running with NVIDIA GPUs in Exclusive Process mode now raises a warning (issue #697, PR #699)

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

  • Openmmtools now supports both Pymbar 3 and 4 versions. (PR #659)
  • Gentle equilibration protocol utility function available in openmmtools.utils.gentle_equilibration (PR #669).
  • Timing information for multiple state sampler is now reported by default (PRs #679 and #671).

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

  • LangevinDynamicsMove now uses openmm.LangevinMiddleIntegrator (a BAOAB integrator) instead of openmm.LangevinIntegrator (an OBABO integrator). Issue #599 (PR #600).

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 specifying n_equilibration_iterations, which doesn't make sense, as n_equilibration_iterations and n_effective_max cannot be computed from statistical_inefficiency alone. Fixed by preventing user from specifying statistical_inefficiency without n_equilibration_iterations. Issue #609 (PR #610).
  • Bug in equilibration detection #2: If the user specified n_equilibration_iterations but not statistical_inefficiency, the returned n_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 returned n_equilibration_iterations. Issue #609 (PR #610).
  • Bug in equilibration detection #3: get_equilibration_data_per_sample returns 0 for n_equilibration_iterations. Fixed by always discarding the first time origin returned by get_equilibration_data_per_sample. To control the amount of data discarded by the first time origin, the user can now specify max_subset when initializing MultiStateSamplerAnalyzer. 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 using os.path.join for creating the output paths. Issue #615, PR #616.

Enhancements

  • LangevinDynamicsMove now allows constraint_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

  • DHFR test system does not require parmed as dependency, since OpenMM can now handle prmtop/inpcrd files. Issue #539 (#588).
  • MultiStateSamplerAnalyzer now allows to manually specify n_equilibrium_iterations and statistical_inefficiency parameters. (#586).

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).
  • Failing windows CI (issue #567) is fixed. (#573)

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's LennardJonesGenerator is detected (#511)

Enhancement

Full Changelog

0.20.3 - Bugfix release

Bugfixes

  • Fixes [#505](#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 to mixed

Bugfixes

  • Replace the cython accelerated all-swap replica mixing scheme with a numba implementation for better stability, and portability, and speed
  • Fixes incorrect temperature spacing in ParallelTemperingSampler constructor
  • Do unit conversion first to improve precision PR #501 (fixes issue #500)

Misc

  • Resolve numpy 1.20 DeprecationWarning about np.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

  • The multistate samplers now issue experimental API warnings via logger.warn() rather than warnings.warn() (#446).
  • Fix return value in states.reduced_potential_at_states (#444).

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

  • Fixed a bug in multistateanalyzer.py where a function was imported from openmmtools.utils instead of openmmtools.multistate.utils (#430).
  • Fixed a few imprecisions in the documentation (#432).

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 in scipy.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

0.18.1 - Bugfix release

This is a minor bugfix release.

New features

  • Improvements for HostGuest* classes - add oemols, host_oemol, and guest_oemol properties to retrieve OpenEye Toolkit OEMol objects (requires toolkit license and installation) - these classes can now accept overriding kwargs

Bugfixes

  • openmmtools.multistate experimental API warning is only issued when openmmtools.multistate is imported
  • AlchemicalNonequilibriumLangevinIntegrator.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) sampling
  • All 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 a GlobalParameter 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 a SamplerState over the network and velocities are not required (#386).
  • Add DoubleWellDimer_WCAFluid and DoubleWellChain_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 with System 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 and AlchemicalState.set_alchemical_variable have been deprecated. Use AlchemicalState.get_alchemical_function and AlchemicalState.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 through ContextCache.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 in AlchemicalState in anticipation of the new implementation of the exact PME that will be based on the NonbondedForce offsets rather than updateParametersInContext().

0.15.0 - Restraint forces

  • Add radially-symmetric restraint custom forces (#336).
  • Copy Python attributes of integrators on deepcopy() (#336).
  • Optimization of states.CompoundThermodynamicState deserialization (#338).
  • Bugfixes (#332, #343).

0.14.0 - Exact treatment of alchemical PME electrostatics, water cluster test system, optimizations

New features

  • Add a WaterCluster testsystem (#322)
  • Add exact treatment of PME electrostatics in alchemy.AbsoluteAlchemicalFactory. (#320)
  • Add method in ThermodynamicState for the efficient computation of the reduced potential at a list of states. (#320)

Enhancements

  • When a SamplerState is applied to many ``Context``s, the units are stripped only once for optimization. (#320)

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

  • LibYAML is now optional (#304)
  • Fix AppVeyor testing against Python 3.4 (now Python 3.5/3.6 and NumPy 1.12) (#307)
  • Release History now included in online Docs

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

  • Fix pickling of CompoundThermodynamicState (#284).
  • Add missing term to OBC2 GB alchemical Force (#288).
  • Generalize forcefactories.restrain_atoms() to non-protein receptors (#290).
  • Standardize integrator global variables in ContextCache (#291).

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 ordering
  • Nose-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

  • Fixed AbsoluteAlchemicalFactory treatment of virtual sites that were previously ignored (#259).
  • Add possibility to add ions to the WaterBox test system (#259).

0.12.0 - GB support in alchemy and new forces module

New features

  • Add AbsoluteAlchemicalFactory support for all GB models (#250)
  • Added forces and forcefactories modules implementing UnishiftedReactionFieldForce and replace_reaction_field respectively. The latter has been moved from AbsoluteAlchemicalFactory (#253)
  • Add restrain_atoms to restrain molecule conformation through an harmonic restrain (#255)

Bug fixes

  • Bugfix for testsystems that use implicit solvent (#250)
  • Bugfix for ContextCache: two consecutive calls retrieve the same Context with same thermodynamic state and no integrator (#252)

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

  • Adds Drew-Dickerson DNA dodecamer test system (#223)
  • Bugfix and optimization to ContextCache (#235)
  • Compress serialized ThermodynamicState strings for speed and size (#232)
  • Backwards compatible with uncompressed serialized ThermodynamicStates

0.11.0 - Conda forge installation

New Features

  • LangevinIntegrator now sets measure_heat=False by default for increased performance (#211)
  • AbsoluteAlchemicalFactory now supports disable_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)

  • Added tests for integrators (#186, #187)

0.9.4 - Nonequilibrium integrators overhaul

Major changes

  • Overhaul of LangevinIntegrator and subclasses to better support nonequilibrium integrators
  • Add 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 and ExternalPerturbationLangevinIntegrator, and both are subclasses of a common NonequilibriumLangevinIntegrator that provides a consistent interface to setting and getting protocol_work
  • AlchemicalNonequilibriumLangevinIntegrator now has a default alchemical_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 any NonequilibriumLangevinIntegrator subclass and returns a unit-bearing work. The optional dimensionless=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 matches openmm.LangevinIntegrator so it can act as a drop-in replacement. (#176)
  • The get_shadow_work() and get_heat() methods are now available for any LangevinIntegrator subclass, as well as the corresponding properties shadow_work and heat. The functions also support dimensionless=True. (#163)
  • The shadow_work and heat properties were added to all LangevinIntegrator subclasses, returning the values of these properties (if the integrator was constructed with the appropriate measure_shadow_work=True or measure_heat=True flags) as unit-bearing quantities
  • The get_protocol_work() and get_total_work() methods are now available for any NonequilibriumLangevinIntegrator, returning unit-bearing quantities unless dimensionless=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 with measure_shadow_work=True.
  • The protocol_work and total_work properties were added to all NonequilibriumLangevinIntegrator subclasses, and return the unit-bearing work quantities. total_work requires the integrator to have been constructed with measure_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 multiple H 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 a CustomNonbondedForce

New openmmtools.testsystems classes

  • AlchemicalWaterBox was added, which has the first water molecule in the system alchemically modified