Below are the notes from all libEnsemble releases.
GitHub issues are referenced, and can be viewed with hyperlinks on the github releases page.
- Fix wait_on_start type-instance condition checking. #1474
- Logging updates:
* Add VDEBUG logging level for the tracking of log message communications. #1486
* Show worker node in the log only when running in distributed mode and with DEBUG logging. #1486
* Update uneven distribution messaging. #1486
- Add scripts for plotting APOSMM optimization runs. #1461
- Convert test runner to Python. #1437
- Move dragonfly GP, heFFTe, ytopt, and Ax multitask tests to community examples. #1439
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12, 3.13
- Heterogeneous workflows tested on Polaris (ALCF) and Perlmutter (NERSC).
- See known issues section in the documentation.
- Fix under-utilized resource usage. #1398
* Fixes bug causing executor to wrongly increase processor counts when not all nodes are utilized.
* Fixes case where setting num_gpus to zero was treated as None.
- Add missing PerlmutterGPU specs (these were detected anyway). #1393
- Handle case where Perlmutter finds no partition. #1391
- Launch environment scripts in shell. #1392
- Add proxystore example (uses a proxy in history array). #1326
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Frontier (OLCF), Polaris (ALCF), and Perlmutter (NERSC).
- Note that tests have been recently run on Aurora (ALCF), but the system was unavailable at time of release.
- See known issues section in the documentation.
- Fix erroneous
nworkers
warning when using mpi4py
comms. #1383
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Frontier (OLCF), Polaris (ALCF), and Perlmutter (NERSC).
- Note that tests have been recently run on Aurora (ALCF), but the system was unavailable at time of release.
- See known issues section in the documentation.
- Add a
live_data
option for real-time data collection / plotting. #1310
nworkers
/is_manager
are set when Ensemble
object is created. #1331/ #1336
* This update locks the comms method when Ensemble
object is created.
- Add a
group_size
option to deal with unevenly resourced nodes. #1349
- Bug fix: Fix shutdown hang on worker error when using
gen_on_manager
. #1348
- Bug fix: Log level was locked to
INFO
when using class interface.
- Updated code to support
numpy
2.0.
Documentation:
Example user functions:
- Update
gpCAM
generators to work with latest interface.
- Change
one_d_func
to norm_eval
. Works with multiple dimensions. #1352 / #1354
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Frontier (OLCF), Polaris (ALCF), and Perlmutter (NERSC).
- Note that tests have been recently run on Aurora (ALCF), but the system was unavailable at time of release.
- Tests were also run on Bebop and Improv LCRC systems.
- See known issues section in the documentation.
- Support generator running on the manager (on a thread). #1216/#1290
* Set libE_specs["gen_on_manager"] = True
* Then run with nworkers equal to the number of simulation workers.
- Default to local comms when nworkers is supplied and no MPI runner is detected. #1169
- Parse args defaults to local comms when --nworkers (or -n) is set on the command line. #1169
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Frontier (OLCF), Polaris (ALCF), and Perlmutter (NERSC).
- Note that tests have been recently run on Aurora (ALCF), but the system was unavailable at time of release.
- Tests were also run on Bebop and Improv LCRC systems.
- See known issues section in the documentation.
- Bugfix: Some libE_specs were not passed through correctly when added after ensemble initialization. #1264
- platform_specs options are now merged with detected platforms, rather than replacing. #1265
- Ensure simulation directories are created when sim_input_dir is specified, likewise for gen dirs. #1266
Example user functions:
- Improved structure of gpCAM generator. #1260
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Frontier (OLCF), Polaris (ALCF), and Perlmutter (NERSC).
- Note that tests have been recently run on Aurora (ALCF), but the system was unavailable at time of release.
- Tests were also run on Bebop and Improv LCRC systems.
- See known issues section in the documentation.
- Fix documentation bug where pydantic models do not display correctly. #1249
- Improve internal efficiency. #1243 / #1249
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Aurora (ALCF), Frontier (OLCF), Polaris, and Perlmutter (NERSC).
- Tests were also run on Bebop and Improv LCRC systems.
- See known issues section in the documentation.
New capabilities:
- Support for both Pydantic 1 and 2. #1135
- Support
object
dtype in history array. #1179 / #1181
- Users can add additional fields to output arrays in user functions. #1203
- Decorators to provide user function in/out specs. #1072
Fixes:
- Bug fix - Overwrite history file on completion when even when the pathname is unchanged. #1177
- Prevent duplicate save when using
save_every_k_gens
. #1154
- Add a
FAILED_TO_START
task status. #1229
- Set
ensemble.nworkers
when create ensemble object even when parse_args
is False. #1162
Platform support:
- Add platform support for Aurora. #1183
* Support for GPU tiles (new platform spec tiles_per_gpu).
* Add libE_specs option use_tiles_as_gpus to treat each tile as a GPU.
* Add Aurora platform guide.
- Add platform guide for Improv. #1235
- Detection of Perlmutter GPU nodes updated. #1211
- Make
srun
GPU setting default to gpus_per_task instead of gpus_per_node. #1206
- Remove Theta support and guide. #1200
Example user functions:
- Add gpCAM generator. #1189 / #1213 / #1220
- Support for IBCDFO local optimization methods in APOSMM. #998
- Add mock_sim to enable replay of a previous run using history file. #1207
- Fix Sine tutorial. #1168
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Aurora (ALCF), Frontier (OLCF), Polaris, and Perlmutter (NERSC).
- Tests were also run on Bebop and Improv LCRC systems.
- See known issues section in the documentation.
New capabilities:
- New history array save options in libE_specs. #1103/#1139/#1141
* save_H_on_completion saves history before exiting main libE function.
* save_H_with_date includes date and timestamp in the save.
* H_file_prefix provides prefix for saved file.
* save_H_on_completion defaults to True when save_every_k_gens/sims is set.
Support for Python versions:
- Adds support/testing for Python 3.12
- Removes testing of Python 3.8
- Tests were run on Linux and MacOS with Python versions 3.9, 3.10, 3.11, 3.12
- Heterogeneous workflows tested on Frontier (OLCF), Polaris, and Perlmutter (NERSC).
- Tests were also run on Bebop and Improv LCRC systems.
- See known issues section in the documentation.
New capabilities:
- libE_specs option final_gen_send returns last results to the generator (replaces final_fields). #1086
- libE_specs option reuse_output_dir allows reuse of workflow and ensemble directories. #1028 #1041
- libE_specs option calc_dir_id_width no. of digits for calc ID in output sim/gen directories. #1052 / #1066
- Added gen_num_procs and gen_num_gpus libE_specs (and persis_info) options for resourcing a generator. #1068
- Added gpu_env_fallback option to platform fields - specifies a GPU environment variable (for non-MPI usage). #1050
- New MPIExecutor submit() argument mpi_runner_type specifies an MPI runner for current call only. #1054
- Allow oversubscription when using the num_procs gen_specs["out"] option. #1058
- sim/gen_specs can use outputs in place of out to be consistent with inputs. #1075
- Executor can be obtained from libE_info (4th parameter) in user functions. #1078
Breaking changes:
- libE_specs option final_fields is removed in favor of final_gen_send. #1086
- libE_specs option kill_canceled_sims now defaults to False. #1062
- parse_args is not run automatically by Ensemble constructor.
Updates to Object Oriented Ensemble interface:
- Added parse_args as option to Ensemble constructor. #1065
- The executor can be passed as an option to the Ensemble constructor. #1078
- Better handling of Ensemble.add_random_streams and ensemble.persis_info. #1074
Output changes:
- The worker ID suffix is removed from sim/gen output directories. #1041
- Separate ensemble.log and libE_stats.txt for different workflows directories. #1027 #1041
- Defaults to four digits for sim/gen ID in output directories (adds digits on overflow). #1052 / #1066
Bug fixes:
- Resolved PETSc/Open-MPI issue (when using the Executor). #1064
- Prevent mpi4py validation running during local comms (when using OO interface). #1065
Performance changes:
- Optimize kill_cancelled_sims function. #1043 / #1063
- safe_mode defaults to False (for performance). #1053
Updates to example functions:
- Multiple regression tests and examples ported to use OO ensemble interface. #1014
Update forces examples:
- Make persistent generator the default for both simple and GPU examples (inc. updated tutorials).
- Update to object oriented interface.
- Added separate variable resources example for forces GPU.
- Rename multi_task example to multi_app.
Documentation:
- General overhaul and simplification of documentation. #992
- Tested platforms include Linux, MacOS, Windows, and major systems such as Frontier (OLCF), Polaris, and Perlmutter (NERSC). The major system tests ran heterogeneous workflows.
- Tested Python versions: (Cpython) 3.7, 3.8, 3.9, 3.10, 3.11.
- See known issues section in the documentation.
- Fixes issues with workflow directories:
* Ensure relative paths are interpreted from where libEnsemble is run. #1020
* Create intermediate directories for workflow paths. #1017
- Fixes issue where libEnsemble pre-initialized a shared multiprocessing queue. #1026
- Tested platforms include Linux, MacOS, Windows and major systems including Frontier (OLCF), Polaris (ALCF), Perlmutter (NERSC), Theta (ALCF) and Bebop. The major system tests ran heterogeneous workflows.
Hotfix for breaking changes in Pydantic.
- Pin Pydantic to version < 2.
- Minor fixes for NumPy 1.25 deprecations.
- Tested platforms include Linux, MacOS, Windows and major systems including Frontier (OLCF) and Perlmutter (NERSC). The major system tests ran heterogeneous workflows.
- Tested Python versions: (Cpython) 3.7, 3.8, 3.9, 3.10, 3.11.
- See known issues section in the documentation.
New capabilities:
- Enhance portability and simplify the assignment of procs/GPUs to worker resources #928 / #983
* Auto-detect GPUs across systems (inc. Nvidia, AMD, and Intel GPUs).
* Auto-determination of GPU assignment method by MPI runner or provided platform.
* Portable auto_assign_gpus / match_procs_to_gpus and num_gpus arguments added to the MPI executor submit.
* Add set_to_gpus function (similar to set_to_slots).
* Allow users to specify known systems via option or environment variable.
* Allow users to specify their own system configurations.
* These changes remove a number of tweaks that were needed for particular platforms.
- Resource management supports GPU and non-GPU simulations in the same ensemble. #993
* User's can specify num_procs and num_gpus in the generator for each evaluation.
- Pydantic models are used for validating major libE input (input can be provided as classes or dictionaries). #878
- Added option to store output and ensemble directories in a workflow directory. #982
- Simplify user function interface. Valid user functions can accept <4 parameters and return <3 values. #971
- New option to parse settings from TOML. #745
- New dry_run option to libE() that checks scripts are valid and returns. #987
- Added an option to the executor submit function to pre-execute a script in the task environment. #996
Breaking changes:
- Removed old Balsam Executor. #921
- Ensemble class moved from libensemble.api to libensemble.ensemble. #1003
- Default to one resource set per simulation in dynamic scheduling mode. #996
Documentation:
- Added type hints/annotations for major modules/functions. #823
- Added Polaris Guide. #930
- Added Frontier Guide. #909
- Added PBS example scripts. #956 #930
- Streamlined and improved the readability of docs. #1004
Tests and Examples:
- Updated forces_gpu tutorial example. #956
* Source code edit is not required for the GPU version.
* Reports whether running on device or host.
* Increases problem size.
* Added versions with persistent generator and multi-task (GPU v non-GPU).
- Moved multiple tests, generators, and simulators to the community repo.
- Added ytopt example. And updated heFFTe example. #943
- Support Python 3.11 #922
- Tested platforms include Linux, MacOS, Windows and major systems: Frontier (OLCF), Polaris (ALCF), and Perlmutter (NERSC). The major system tests ran heterogeneous workflows.
- Recent testing was also carried out on Summit (IBM Power9/LSF), but this was not possible at time of release.
- Tested Python versions: (Cpython) 3.7, 3.8, 3.9, 3.10, 3.11.
- See known issues section in the documentation.
New capabilities:
- New pair of utilities, liberegister and libesubmit (based on PSI/J), for easily preparing and launching libEnsemble workflows with local comms onto most machines and schedulers. #807
- New persistent support function to cancel sim_ids (request_cancel_sim_ids). #880
- keep_state option for persistent workers: this lets the manager know that the information being sent is intermediate. #880
Other enhancements:
- The Executor manager_poll() interface now sets consistent flags instead of literal strings. #877
- Some internal modules and the test suite now work on Windows. #869 #888
- Specifying the num_resource_sets libE_specs option instead of zero_resource_workers is now recommended except when using a fixed worker/resource mapping. Use
persis_info["gen_resources"]
to assign persistent generator resources (default is zero). #905
- An extraneous warning removed. #903
- Tested platforms include Linux, MacOS, Windows, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), Swing (A100 GPU system), Perlmutter (HPE Cray EX with A100 NVIDIA GPUs). For Perlmutter, see "Known issues" below.
- Tested Python versions: (Cpython) 3.7, 3.8, 3.9, 3.10.
- At time of testing on Perlmutter there was an issue running concurrent applications on a node, following a recent system update. This also affects previous versions of libEnsemble, and is being investigated.
- See known issues section in the documentation.
New capabilities:
- Support auto-detection of PBS node lists. #602
- Added configuration options for libE_stats.txt file. #743
- Support for spawn and forkserver multiprocessing start methods. #797
Updates to example functions:
Added simple dynamic sampling example. #833
Added heFFTe example. #844
Regression tests separated into problem examples and functionality tests. #839
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), Swing (A100 GPU system), Perlmutter (HPE Cray EX with A100 NVIDIA GPUs).
- Tested Python versions: (Cpython) 3.7, 3.8, 3.9, 3.10.
- The APOSMM generator function has been noted to operate slower than expected with the spawn multiprocessing start method. For this reason we recommend using fork with APOSMM, when using local comms (fork is the default method on Linux systems).
- See known issues section in the documentation.
This release has minimal changes, but a large number of touched lines.
- Reformatted code for black compliance, including string normalization. #811, #814, #821
- Added Spock and Crusher guides. #802
- User can now set
calc_status
to any string (for output in libE_stats). #808
- Added a workflows community initiative file. #817
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), Swing (A100 GPU system), Perlmutter (HPE Cray EX with A100 NVIDIA GPUs).
- Tested Python versions: (Cpython) 3.7, 3.8, 3.9, 3.10.
- See known issues section in the documentation.
Featured new capabilities:
- New Balsam Executor with multi-site capability (run user applications on remote systems). #631, #729
- Support for funcX (place user functions on remote systems). #712 / #713
- Added partial support for concurrent/futures interface. (cancel(), cancelled(), done(), running(), result(), exception() and context manager) #719
Breaking API / helper function changes:
See "Updating for libEnsemble v0.9.0" wiki for details:
https://github.com/Libensemble/libensemble/wiki/Updating-for-libEnsemble-v0.9.0
- Scheduler options moved from alloc_specs['user'] to libE_specs. #790
- BalsamMPIExecutor is now LegacyBalsamMPIExecutor. #729
- The exit_criteria elapsed_wallclock_time has been renamed wallclock_max. #750 (with a deprecation warning)
- Clearer and consistent naming of libE-protected fields in history array. #760
Updates to example functions:
- Moved some examples to new repository - [libe-community-examples](https://github.com/Libensemble/libe-community-examples) (VTMOP, DEAP, DeepDriveMD). #716, #721, #726
- Updates to Tasmanian examples to include asynchronous generator example. #727 / #732
- Added multi-task, multi-fidelity optimization regression tests using ax. #717 / #720
Other functionality enhancements:
- Non-blocking option added for persistent user function receives. #752
- Added match_slots option to resource scheduler. #746
Documentation:
- Added tutorial on assigning tasks to GPUs. #768
- Refactored Executor tutorial for simplicity. #749
- Added Perlmutter guide. #728
- Added Slurm guide. #728
- Refactored examples and tutorials - added exercises. #736 / #737
- Updated history array documentation with visual workflow example. #723
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), Swing (A100 GPU system), Perlmutter (HPE Cray EX with A100 NVIDIA GPUs).
- Tested Python versions: (Cpython) 3.7, 3.8, 3.9, 3.10.
- Open-MPI does not work with direct MPI job launches in
mpi4py
comms mode,
since it does not support nested MPI launches.
(Either use local mode or the Balsam Executor.)
- See known issues section in the documentation for more issues.
Featured new capabilities:
- Variable resource workers (dynamic reassignment of resources to workers). #643
- Alternative libE interface. An Ensemble object is created and can be parameterized by a YAML file. #645
- Improved support classes/functions for alloc/gen/sims and executors.
- Many new example generator/simulators and workflows.
Breaking API / helper function changes:
See "Updating for libEnsemble v0.8.0" wiki for details:
https://github.com/Libensemble/libensemble/wiki/Updating-for-libEnsemble-v0.8.0
- Resources management is now independent of the executor. #345
- The
'persis_in'
field has been added to gen_specs/sim_specs (instead of being hard-coded in alloc funcs). #626 / #670
alloc support
module is now a class. #643 / #656
gen_support
module is replaced by Persistent Worker support module (now a class). #609 / #671
- Remove
libE_
prefix from the logger. #608
avail_worker_ids
function should specify EVAL_GEN_TAG
or EVAL_SIM_TAG
instead of True
. #615 #643
- Pass
libE_info
to allocation functions (allows more flexibility for user and efficiency improvements). #672
'given_back'
is now a protected libEnsemble field in the manager's history array. #651
- Several name changes to functions and parameters (See the wiki above for details). #529 / #659
Updates to example functions:
- Suite of distributed optimization methods for minimizing sums of convex functions. #647 / #649. Methods include:
- Added batched construction for Tasmanian example. #644
- Added Tasmanian dependency to Spack package. spack/spack#25762
- Added VTMOP source code and example usage. #676
- Added a multi-fidelity persistent_gp regression test. #683 / #684
- Added a DeepDriveMD inspired workflow. #630
- Created a persistent sim example. #614 / #615
- Added an example where cancellations are given from the alloc func. #677
Other functionality changes:
- A helper function for generic task polling loop has been added. #572 / #612
- Break main loop now happens when sim_max is returned rather than given out. #624
- Enable a final communication with gen. #620 / #628
- Logging updates - includes timestamps, enhanced debug logging, and libEnsemble version. #629 / #674
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), Swing (A100 GPU system).
- Tested Python versions: (Cpython) 3.6, 3.7, 3.8, 3.9, 3.10 [1].
[1] | A reduced set of tests were run for python 3.10 due to some unavailable test dependencies at time of release. |
- Open-MPI does not work with direct MPI job launches in
mpi4py
comms mode,
since it does not support nested MPI launches.
(Either use local mode or the Balsam Executor.)
- See known issues section in the documentation for more issues.
API additions:
- Active receive option added that allows irregular manager/worker communication patterns. (#527 / #595)
- A mechanism is added for the cancellation/killing of previously issued evaluations. (#528 / #595 / #596)
- A submit function is added in the base
Executor
class that runs a serial application locally. (#531 / #595)
- Added libEnsemble history array protected fields: returned_time, last_given_time, and last_gen_time. (#590)
- Updated libE_specs options (
mpi_comm
and profile
). (#547 / #548)
- Explicit seeding of random streams in
add_unique_random_streams()
is now possible. (#542 / #545)
Updates to example functions:
- Added Surmise calibration generator function and two examples (regression tests). (#595)
Other changes:
- Better support for uneven worker to node distribution (including at sub-node level). (#591 / #600)
- Fixed crash when running on Windows. (#534)
- Fixed crash when running with empty persis_info. (#571 / #578)
- Error handling has been made more robust. (#592)
- Improve
H0
processing (esp. for pre-generated, but not evaluated points). (#536 / #537)
- A global
sim_id
is now given, rather than a local count, in _libE_stats.txt_. Also a global gen count is given. (#587, #588)
- Added support for Python 3.9. (#532 / Removed support for Python 3.5. (#562)
- Improve SLURM nodelist detection (more robust). (#560)
- Add check that user does not change protected history fields (Disable via
libE_specs['safe_mode'] = False
). (#541)
- Added
print_fields.py
script for better interrogating the output history files. (#558)
- In examples,
is_master
changed to is_manager
to be consistent with manager/worker nomenclature. (#524)
Documentation:
- Added tutorial Borehole Calibration with Selective Simulation Cancellation. (#581 / #595)
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm).
- Tested Python versions: (Cpython) 3.6, 3.7, 3.8, 3.9.
- Open-MPI does not work with direct MPI job launches in
mpi4py
comms mode, since it does not support nested MPI launches
(Either use local mode or Balsam Executor).
- See known issues section in the documentation for more issues.
Dependencies:
psutils
is now a required dependency. (#478 #491)
API additions:
- Executor updates:
- Addition of a zero-resource worker option for persistent gens (does not allocate nodes to gen). (#500)
- Multiple applications can be registered to the Executor (and submitted) by name. (#498)
- Wait function added to Tasks. (#499)
- Gen directories can now be created with options analogous to those for sim dirs. (#349 / #489)
Other changes:
- Improve comms efficiency (Repack fields when NumPy version 1.15+). (#511)
- Fix multiprocessing error on macOS/Python3.8 (Use 'fork' instead of 'spawn'). (#502 / #503)
Updates to example functions:
- Allow APOSMM to trigger ensemble exit when condition reached. (#507)
- Improvement in how persistent APOSMM shuts down subprocesses (preventing PETSc MPI-abort). (#478)
Documentation:
- APOSMM Tutorial added. (#468)
- Writing guide for user functions added to docs (e.g., creating sim_f, gen_f, alloc_f). (#510)
- Addition of posters and presentations section to docs (inc. Jupyter notebooks/binder links). (#492 #497)
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), and Bridges (HPE system at PSC).
- Cori (Cray XC40/Slurm) was not tested with release code due to system issues.
- Tested Python versions: (Cpython) 3.5, 3.6, 3.7, 3.8.
- We currently recommend running in Central mode on Bridges, as distributed runs are experiencing hangs.
- Open-MPI does not work with direct MPI job launches in mpi4py comms mode, since it does not support nested MPI launches
(Either use local mode or Balsam Executor).
- See known issues section in the documentation for more issues.
Breaking API changes:
- Job_controller/Job renamed to Executor/Task and
launch
function to submit
. (#285)
- Executors/Resources/Utils moved into sub-packages.
utils
now in package tools
. (#285)
- sim/gen/alloc support functions moved into
tools
sub-package. (#285)
- Restructuring of sim directory creation with
libE_specs
configuration options.
E.g: When sim_input_dir
is given, directories for each sim are created. (#267)
- User can supply a file called
node_list
(replaces worker_list
). (#455)
API additions:
- Added gen_funcs.rc configuration framework with option to select APOSMM Optimizers for import. (#444)
- Provide
alloc_specs
defaults via alloc_funcs.defaults module. (#325)
- Added
extra_args
option to the Executor submit function to allow addition of arbitrary MPI runner options. (#445)
- Added
custom_info
argument to MPI Executor to allow overriding of detected settings. (#448)
- Added
libE_specs
option to disable log files. (#368)
Other changes:
- Added libEnsemble Conda package, hosted on conda-forge.
- Bugfix: Intermittent failures with repeated libE calls under mpi4py comms.
Every libE call now uses its own duplicate of provided communicator and closes out. (#373/#387)
- More accurate timing in libE_stats.txt. (#318)
- Addition of new post-processing scripts.
Updates to example functions:
- Persistent APOSMM is now the recommended APOSMM (aposmm.py renamed to old_aposmm.py). (#435)
- New alloc/gen func: Finite difference parameters with noise estimation. (#350)
- New example gen func: Tasmanian UQ generator. (#351)
- New example gen func: Deap/NSGA2 generator. (#407)
- New example gen func to interface with VTMOP.
- New example sim func: Borehole. (#367)
- New example use-case: WarpX/APOSMM. (#425)
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), Cori (Cray XC40/Slurm), and Bridges (HPE system at PSC).
- Tested Python versions: (Cpython) 3.5, 3.6, 3.7, 3.8.
- We currently recommended running in Central mode on Bridges as distributed runs are experiencing hangs.
- See known issues section in the documentation for more issues.
API changes:
- sim/gen/alloc_specs options that do not directly involve these routines are moved to libE_specs (see docs). (#266, #269)
- sim/gen/alloc_specs now require user-defined attributes to be added under the
'user'
field (see docs and examples). (#266, #269)
- Addition of a utils module to help users create calling scripts. Includes an argument parser and utility functions. (#308)
- check_inputs() function is moved to the utils module. (#308)
- The libE_specs option
nprocesses
has been changed to nworkers
. (#235)
New example functions:
- Addition of a persistent APOSMM generator function. (#217)
Other changes:
- Overhaul of documentation, including HPC platform guides and a new pdf structure. (inc. #232, #282)
- Addition of OpenMP threading and GPU support to forces test. (#250)
- Balsam job_controller now tested on Travis. (#47)
- Tested platforms include Linux, MacOS, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), Bebop (Cray CS400/Slurm), and Cori (Cray XC40/Slurm).
- Tested Python versions: (Cpython) 3.5, 3.6, 3.7
- These are unchanged from v0.5.0.
- A known issues section has now been added to the documentation.
- Code has been restructured to meet xSDK package policies for interoperable ECP software (version 0.5.0). #208
- The use of MPI.COMM_WORLD has been removed. Uses a duplicate of COMM_WORLD if no communicator passed (any process not in communicator returns with an exit code of 3). #108
- All output from libEnsemble goes via logger. MANAGER_WARNING level added. This level and above are echoed to stderr by default. API option to change echo level.
- Simulation directories are created only during sim_f calls are suffixed by _worker. #146
- New user function libE.check_inputs() can be used to check valid configuration of inputs. Can be called in serial or under MPI (see libE API). #65
- Installation option has been added to install dependencies used in tests
pip install libensemble[extras]
.
- A profiling option has been added to sim_specs. #170
- Results comparison scripts have been included for convenience.
- Tested platforms include Linux, MacOS (New), Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), and Bebop (Cray CS400/Slurm).
- Tested Python versions: (Cpython) 3.5, 3.6, 3.7
- Note Support has been removed for Python 3.4 since it is officially retired. Also NumPy has removed support.
- These are unchanged from v0.5.0.
- Fixed LSF resource detection for large jobs on LSF systems (e.g., Summit). #184
- Added support for macOS. #182
- Improved the documentation (including addition of beginner's tutorial and FAQ).
- Tested platforms include Local Linux, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), and Bebop (Cray CS400/Slurm).
- Tested Python versions: (Cpython) 3.4, 3.5, 3.6, 3.7.
- These are unchanged from v0.5.0.
- Added local (multiprocessing) and TCP options for manager/worker communications, in addition to mpi4py. (#42).
- Example: libEnsemble can be run on MOM/launch nodes (e.g., those of ALCF/Theta & OLCF/Summit) and can remotely detect compute resources.
- Example: libEnsemble can be run on a system without MPI.
- Example: libEnsemble can be run with a local manager and remote TCP workers.
- Added support for Summit/LSF scheduler in job controller.
- MPI job controller detects and retries launches on failure; adding resilience. (#143)
- Job controller supports option to extract/print job times in libE_stats.txt. (#136)
- Default logging level changed to INFO. (#164)
- Logging interface added, which allows user to change logging level and file. (#110)
- All worker logging and calculation stats are routed through manager.
- libEnsemble can be run without a gen_func, for example, when using a previously computed random sample. (#122)
- Aborts dump persis_info with the history.
- This version no longer supports Python 2.
- Tested platforms include Local Linux, Theta (Cray XC40/Cobalt), Summit (IBM Power9/LSF), and Bebop (Cray CS400/Slurm).
- Open-MPI does not work with direct MPI job launches in mpi4py comms mode, since it does not support nested MPI launches
(Either use local mode or Balsam job controller).
- Local comms mode (multiprocessing) may fail if MPI is initialized before forking processors. This is thought to be responsible for issues combining with PETSc.
- Remote detection of logical cores via LSB_HOSTS (e.g., Summit) returns number of physical cores since SMT info not available.
- TCP mode does not support (1) more than one libEnsemble call in a given script or (2) the auto-resources option to the job controller.
- Logging no longer uses root logger (also added option to change libEnsemble log level). (#105)
- Added wait_on_run option for job controller launch to block until jobs have started. (#111)
- persis_info can be passed to sim as well as gen functions. (#112)
- Postprocessing scripts added to create performance/utilization graphs. (#102)
- New scaling test added (not part of current CI test suite). (#114)
- Separated job controller classes into different modules including a base class (API change).
- Added central_mode run option to distributed type (MPI) job_controllers (API addition). (#93)
- Made poll and kill job methods (API change).
- In job_controller, set_kill_mode is removed and replaced by a wait argument for a hard kill (API change).
- Removed register module - incorporated into job_controller (API change).
- APOSMM has improved asynchronicity when batch mode is false (with new example). (#96)
- Manager errors (instead of hangs) when alloc_f or gen_f don't return work when all workers are idle. (#95)
- Open-MPI is not supported with direct MPI launches since nested MPI launches are not supported.
- Issues with killing jobs have been fixed. (#21)
- Fixed job_controller manager_poll to work with multiple jobs. (#62)
- API change: persis_info now included as an argument to libE and is returned from libE instead of gen_info
- Gen funcs: aposmm_logic module renamed to aposmm.
- New example gen and allocation functions.
- Updated Balsam launch script (with new Balsam workflow).
- History is dumped to file on manager or worker exception and MPI aborted (with exit code 1). (#46)
- Default logging level changed to DEBUG and redirected to file ensemble.log.
- Added directory of standalone tests (comms, job kills, and nested MPI launches).
- Improved and sped up unit tests. (#68)
- Considerable documentation enhancements.
- Open-MPI is not supported with direct MPI launches since nested MPI launches are not supported.
- Added job_controller interface (for portable user scripts).
- Added support for using the Balsam job manager. Enables portability and dynamic scheduling.
- Added autodetection of system resources.
- Scalability testing: Ensemble performed with 1023 workers on Theta (Cray XC40) using Balsam.
- Tested MPI libraries: MPICH and Intel MPI.
- Killing MPI jobs does not work correctly on some systems (including Cray XC40 and CS400). In these cases, libEnsemble continues, but processes remain running.
- Open-MPI does not work correctly with direct launches (and has not been tested with Balsam).