Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* Feature/GitHub actions (#5) * first try of GitHub Actions * go simpler this time * trying to fix MPI error * testing mpi * separate stage mpi install * fix * install openmpi-dev * try again * testing now works, try documentation now * new file for documentation * install mpi for docs * host everything on GitHub * using default deploydocs * fix repo * remove workaround * branches * ok devbranch * try deploying doc to my own repo * try again with TRAVIS_REPO_SLUG * fix typos * DOCUMENTER_KEY * deploy to Rimu.jl now * fix wrong repo * remove DOCUMENTER_KEY * final commit * added an example calculation with BHM with detailed descriptions * added a new mpi example script, the old one is bit outdated * updated documentation, still more to cover * minor changes * trying to fix source code not found * temporarily changes devbranch to this branch for testing * don't need repo in makedocs for github * trying to fix failed doctests * added logo * added an exmaple page in the documentation, let's see how that looks * modified index.md * set a hiden r_strat in BHM-example.jl to skip writeinfo * added RNG seed; fixed index.md * removed "et" from FCIQMC loop * added preview * updated Documenter.jl in Manifest.toml * final commit, change devbranch back to "develop" * test out the workaround with @doc for source links in the documentation * temporarily switch devbranch to current branch * source links to docstrings working now, switch devbranch back * expand run test on all feature/**, and doc for feature/doc-** branches * fixed typos * fixed typos in example scripts * updated example doc * changes as requested in review * comment out OvershootControl * updated install links * devbranch for docs changed back to default master * automate devbranch for deploydocs, testing * now it should deploy the documentation * remove feature/doc** for deployment * updated the MPI section in index.md (#6) Documentation update only. * Bugfix/compatj1.5 (#7) * changed == to === for comparison with nothing * fix a vector return bug in fciqmc! the bus could cause the wrong coefficient vector to be returned * fixed the tests to run with v1.5 as well as with earlier versions Random selections from a collection changed in Julia v1.5, which made it necessary to change all tests that depend on spawning events. * changed Julia version for workflows testing to latest * run CI on push to bugfix branches * fixed wrong version spec * Feature/momspacebhchain (#8) * new function numSandDoccupiedsites() * started implementing BoseHubbardMom1D removed superfluous code in Hamiltonians.jl; added missing methods for numSandDoccupiedsites(); added method for numOfHops() * added hop() and tests * BoseHubbardMom1D is now complete and exported beware of bugs!!! Added diagME() for momentum space added tests for BoseHubbardMom1D, also in lomc!() * add Momentum and rayleighQuotient() * tests for Momentum and rayleighQuotient() * fix modulo bug in momentum calculation * fixed bugs in BoseHubbardMom1D, numOfHops() and hops() fixed tests * renamed LinearOperator to AbstractHamiltonian * slashed some bugs in BoseHubbardMom1D * added the unexported build_sparse_matrix_from_LO() renamed rayleigh_quotient() * fixed and expanded tests in set BoseHubbardMom1D * introduce nearUniformONR() rename nearUniform(n,m) -> onr to nearUniformONR(), keep nearUniform() for BoseFS and BosonicHamiltonian * made address type a parameter in BoseHubbardMom1D * cosmetic change * fix test * fix a docstring typo * updated documentation for BoseHubbardMom1D * Feature/two norm (#10) * added `NoProjectionTwoNorm` as a new `ProjectionStrategy` * fix test on Julia v1.4 * added CI test with Julia v1.4 * fix workflow actions file * change workflows to test on Julia 1.5.2 * added Norm2Projector() * improve doc string * Bugfix/annihilation stochastic threshold (#11) * fixed sign bug in fciqmc_col() for IsStochasticWithThreshold need to fix broken tests * fixed tests and docstring ProjectedMemory * Feature/delayed-double-log-update (#12) * added DelayedDoubleLogUpdate shift strategy; fixed some typos * added DelayedDoubleLogUpdateAfterTW strategy * corrected docstrings * Feature/deltanoise (#13) * introduced DeltaMemory3 * add @setThreshold and @setDeterministic * removed wrong comment * add missing method for BoseFS * added growthWitness() and test * modified tests * fix broken test * improved @setThreshold, @setDeterministic, new function setThreshold90 added tests * constructors with keyword `capacity` for DVec, DFVec and FastDVec * make NormProjector() and Norm2Projector() faster Projection of Hamiltonian is now ommitted when specifying these projectors. * Add scientific paper references * formatting fix * Another formatting fix * bump version * Feature/deltanoise (#15) * introduced DeltaMemory3 * add @setThreshold and @setDeterministic * removed wrong comment * add missing method for BoseFS * added growthWitness() and test * modified tests * fix broken test * improved @setThreshold, @setDeterministic, new function setThreshold90 added tests * constructors with keyword `capacity` for DVec, DFVec and FastDVec * make NormProjector() and Norm2Projector() faster Projection of Hamiltonian is now ommitted when specifying these projectors. * yet another formatting fix * define adjoint of Hamiltonian and missing*dv * towards faster projected energy * improve adjoint of AbstractHamiltonian, document `h_adjoint()` * Introduce `AbstractProjector` supertype. * fix bug * new exported function copytight() * add tests for copytight() * Rework ReportingStrategy for faster calculation of projected energy ReportingStrategy types now contain two projector fields new external dependency Setfield new unexported Rimu.refine_r_strat * commment only * testset "ReportingStrategy internals" * remove spurious method * fix bug in tests * fix compatibility with Julia v1.4 * replace ks() by faster version * BitstringAdresses.i_onr() fast and non-allocating internal OccupationNumberIterator single argument onr() returns SVector{M,Int32} * faster diagME() for BoseHubbardMom1D * fix test and bug * new faster Hamiltonian HubbardMom1D() with tests * return Tuple from fciqmc_col!() to avoid allocation time identified by profiling * extended Hops to take another field; faster Hops for HubbardMom1D * remove allocations using Setfield.jl new unsafe_shift_left() in BitStringAddresses and hop() for HubbardMom1D fciqmc_col!() with HubbardMom1D is now non-allocating * bump version number * update package versions * removed reference counting to be compatible with MPI.jl v0.16.0 updated package dependencies for Rimu and Rimu/test and Rimu/docs * Bugfix/GitHub actions (#16) * added env variable * try @latest for actions/checkout and setup-julia * @latest does not work, try @v2 and @v1 * Feature/embarrassinglydistributed (#18) * make `autoblock()` handle missing values * new module EmbarrassinglyDistributed with 4 functions * remove tests for old julia version * added comments to document unit test * part 2 * improve docstrings * update julia version for docummenter * Feature/BoseFS2C (#19) * introduced DeltaMemory3 * add @setThreshold and @setDeterministic * removed wrong comment * add missing method for BoseFS * added growthWitness() and test * modified tests * fix broken test * improved @setThreshold, @setDeterministic, new function setThreshold90 added tests * constructors with keyword `capacity` for DVec, DFVec and FastDVec * make NormProjector() and Norm2Projector() faster Projection of Hamiltonian is now ommitted when specifying these projectors. * yet another formatting fix * define adjoint of Hamiltonian and missing*dv * towards faster projected energy * improve adjoint of AbstractHamiltonian, document `h_adjoint()` * Introduce `AbstractProjector` supertype. * fix bug * new exported function copytight() * add tests for copytight() * Rework ReportingStrategy for faster calculation of projected energy ReportingStrategy types now contain two projector fields new external dependency Setfield new unexported Rimu.refine_r_strat * commment only * testset "ReportingStrategy internals" * remove spurious method * fix bug in tests * fix compatibility with Julia v1.4 * replace ks() by faster version * BitstringAdresses.i_onr() fast and non-allocating internal OccupationNumberIterator single argument onr() returns SVector{M,Int32} * BoseFS2C for bsa and bsb with same type only * BoseFS2C works with different types of BitStringAddressType now * BoseFS2C works with onr now * BoseFS2C from two BoseFS * faster diagME() for BoseHubbardMom1D * fix test and bug * new faster Hamiltonian HubbardMom1D() with tests * return Tuple from fciqmc_col!() to avoid allocation time identified by profiling * reorganising code blocks * added benchmark functions BoseFS2C_, will remove later * extended Hops to take another field; faster Hops for HubbardMom1D * remove allocations using Setfield.jl new unsafe_shift_left() in BitStringAddresses and hop() for HubbardMom1D fciqmc_col!() with HubbardMom1D is now non-allocating * bump version number * update package versions * BoseFS2C type stable version; start coding BoseHubbard2CMom1D * Hops works with BoseFS2C; fciqmc also seems working fine in real space * fixed function orders * fixed BH2Cinteraction * BoseHubbard2CMom1D but missing interaction2c * fixed numOfHops for BH2CMom1D * trying to correct diagME * very ugly code for hop(h::BH2CMom1D, add::BoseFS2C, chosen), but works * an example file for testing * update comments * attempts to fix ham * fixed hopacross2adds * added BoseHubbardMom1D2C for benchmarking * fixed diagME, now getting correct energy * everything seems working correctly now * updated docstrings * new subtype TwoComponentBosonicHamiltonian * changing code order; fixed typos * added testsets for BoseFS2C and TwoComponentBosonicHamiltonian * export TwoComponentBosonicHamiltonian, fixed docstrings * fxied typos * fixing problems for pull request * type unstable version of BoseFS2C() * removed add from BoseHubbardMom1D2C, added m; code much faster * added Bose.show() for BoseFS2C * updated docstring * more docstring change * removed unnecessary fields from real and mom BH1d2C * now V is a type param for both Real and Mom 1D2C ham Co-authored-by: Joachim Brand <j.brand@massey.ac.nz> * fixing type parameter for HubbardMom1D (#20) * Fix problem with reverse in `BoseFS` constructor. (#21) * Fix weird performance problem with reverse. Julia wasn't happy with `reverse(onr)`. This problem is not there on the Julia `master` branch, but it exists on 1.5 and 1.6. * Apply fix to other constructors as well. * Feature/complexnoise (#23) * introduced variable type shift and norm * fixed bugs and introduced test * changed order of arguments in norm_project() * changed norm_project!() to four-argument call structure and updated docs * add cRandn() and ComplexNoiseCancellation * preparing for complex noise MPI sync/broadcast of complex noise; bug fix in ComplexNoiseCancellation Prepare fciqmc!, fciqmc_step!, fciqmc_col! and applyMemorynoise! for complex coefficients new testset ComplexNoiseCancellation * fixed bugs, better error message with ComplexNoiseCancellation extended testset * gW() new version of growth witness * prepare Blocking for complex data * sensible and compile-time evaluated defaults for StochasticStyle make DoubleLogUpdate a parametric type * new function `walkernumber` generalising the one-norm for complex walkes * new StocjasticStyle `IsStochastic2Pop` new `walkernumber(w)` to replace norm(w,1) dispatching on StochasticStyle. intro fciqmc_col(::IsStochastic2Pop,...) but not tested yet. * replace for loop by mapreduce() in fciqmc_step!() * precalculating type of stats by running fciqmc_step!() once runtests.jl now fails because of rng. * fixed tests and remove testing version < v1.5 * added preliminary testset for complex walkers spawning for IsStochastic2Pop still seems to be buggy as results are not sensible * fixed bug in using imaginary part of shift * unexported function purge_negative_walkers!() needs to be generalised for mpi and threading * new MemoryStrategy PurgeNegatives() * IsStochastic2PopInitiator() and IsStochastic2PopWithThreshold() The trait IsStochastic2PopWithThreshold is introduced but the relevant methods are still missing. * improve multithreading fixed multithreading for complex walkers more efficient threading algorithm using SpittablesBase and added other algorithms using ThreadsX (currently not used) consolitdated non-threaded fciqmc_step!() for use with and without MPI * fix tests test in "fciqmc with BoseFS" of multithreaded code was updated with new values after code has changed * replace sync_MPI_cRand() with sync_cRand() defined in ConsistentRNG * Make CRNGs be seeded at runtime when module loads CRNGs now becomes Ref on a vector of of RNGs; length is decided by Threads.nthreads() at runtime. * trying to fix DataFrames dependency bug * Add module RimuIO for saving DataFrames in Arrow format files. * add consistency and sanity checking during RNG seeding check_crng_independence(v) for doing a one-off sanity check * MPI rework 1: moving MPI functionality into module RMPI needs more tests * Load but do not export RMPI. * new function mpi_seed_CRNGs!() * rename save and load into save_df and load_df * add mpi example script and use it for testing mpi; fix small bug in fciqmc * clean up * fix Julia using system installed MPI * free DataFrames and cleanup Project.toml and actions * new function smoothen(); also fixed buf in gW. * fix GMTest of EmbarrassinglyDistributed * don't re-export Distributed; replace `mapreduce` -> `sum` * fixes after PR review. Removed extraneous using statment. Renamed `my_stats` into `combine_stats`. Replaced `reduce` and `mapreduce` with more descriptive `sum` and `all`. * New DVec tests (#22) * Implement test suite for `AbstractDVec` interface * fix docstring * Add KrylovKit tests * Move other DFVec tests to DictVectors.jl * Add test for BoseFS, formatting * Delete Manifests (#26) * Add coverage to Github action (#25) * add code coverage tracking * fix action * change version to 1.5 so the latest 1.5.x gets used * fix coveralls part * DVec2 (#24) * Move stochastic styles to DictVectors * Clean up DVec and interface * Apply cleanup to DFVec and FastDVec * DVec2 implementation and tests * move things around * minor formatting and commented code removal * change `ADV*Iterators` to `FDV*Iterators` as they are `FastDVec` specific * more generic functions * improve test coverage * fix problems from review * Unexport experimental `StochasticStyle` types. Remove references in docstrings to the unexported features. `IsStochasticNonlinear` `IsStochastic2Pop` `IsStochastic2PopInitiator` `IsStochastic2PopWithThreshold` `IsSemistochastic` Co-authored-by: Joachim Brand <j.brand@massey.ac.nz> * CI Benchmarking (#27) * add benchmarks * add benchmark action * tweak numbers to prevent capacity overflow and get consistent timings * update benchmarks (#32) * Fix LOStructure for BoseHubbardReal1D2C (#33) * Bugfix/gh actions mpi (#34) * debug test first * change JULIA_MPIEXEC=mpiexec * change MPI test to print hello only * try mpirun * more debug * Pkg.add("MPI") * try running from bash * test again * test * run mpi test outside julia * using @__DIR__ for mpi_df.arrow * test output file * move things around * add @. for julia mpi * run single core julia first * change openmpi version to 2.1.1 * change openmpi version to 2.1.1-8 * version again * version policy * do not apt update * using native julia MPI * using MPi * export PATH * fix PATH * remove MPI installation * run MPI in julia * fix * undo previous fix * fix $OUTPUT_FILE var * clean up * Bugfix/gh actions mpi (#35) * debug test first * change JULIA_MPIEXEC=mpiexec * change MPI test to print hello only * try mpirun * more debug * Pkg.add("MPI") * try running from bash * test again * test * run mpi test outside julia * using @__DIR__ for mpi_df.arrow * test output file * move things around * add @. for julia mpi * run single core julia first * change openmpi version to 2.1.1 * change openmpi version to 2.1.1-8 * version again * version policy * do not apt update * using native julia MPI * using MPi * export PATH * fix PATH * remove MPI installation * run MPI in julia * fix * undo previous fix * fix $OUTPUT_FILE var * clean up * checks if tests run locally * BitStringAddresses cleanup (#28) * reogranize benchmarks, move MPI tests to the end * benchmark updates * move BitStringAddress code around, remove BStringAddress * implement interface for BSAdd* * Use the interface with BSAdd where applicable * add `chunks` * rename some functions and types, tests * fix interfaces, code reordering * remove maxBSLength, fix Base.:~ * add specialized onr function, remove s_onr * formatting, remove bitaddr * optimize numberoccupiedsites * optimize bosehubbardinteraction * clean up and document implementations * hopnextneighbour improvement: better bit masks * implement occupied_orbitals iterator * implement numSandDoccupiedsites, hopnextneighbour and bosehubbardinteraction * fix hopnextneighbour * Add BitString * fix tests and bugs * remove abstract.jl, typos * tweaks, remove unused imports * hardcode UInt64, remove x_bit_mask * docstring tweaks, export reordering * fix test that sometimes fails due to choice of random bitstring * change BitString `show` method * remove unnecessary if statement, tweak tests * remove bitstring-specific functions from BoseFS * use new benchmarks * simplify KrylovKit test * Hamiltonians (#36) * add HubbardReal1D, ExtendedHubbardReal1D, refactor HubbardMom1D constructor * rework dimension, add starting_address, rename Momentum, remove Bosonic abstract types * implement AbstractHops * add num_modes for hamiltonians, change HubbardMom1D * reorder type parameters, tweak `show` * implement specialized hops for Mom1D2C, Real1D2C * more tests, fix isless and nearUniform for addresses * remove unnecessary `@with_kw` * code organization and docstrings * rename `hops`, `diagME` and related functions * HubbardReal1D in benchmarks * dont remove ghost bits in usigned right shift * remove unused benchmarks * Julia 1.6 compatibility (#38) * Julia 1.6 compatibility * Fix bug with dict allocation * Fix numbers again * Remove Rimu from test Project.toml (#39) * Fix Hubbard momentum-space diagonal elements (#41) * Add missing t factor to diagonal elements in HubbardMom1D * Move formula and code around, add tests * Importance Sampling (#37) * implement importance sampling * rename to gutzwiller, add new `LOStructure` * implement adjoint, a `LOAdjoint` trait, use said trait in refine_r_strat * merge `LOStructure` and `LOAdjoint`. Rename `HermitianLO` to `Hermitian` for consistency * use new function names * remove diagMEs * remove hops in tests * Implement GuidingVectorSampling * improve tests * docstrings and minor improvements * remove rogue HermitianLO * fix `\\tilde`s * Fix bug with wrong order, improve docstrings. * Simplify offdiagonal code * simd sum (#42) * fixed definition of kinetic energy (#43) * Fix bug with empty reduce error (#44) * Fix bug with empty reduce error * Rework fix with allocate_statss * Use new ArrowTypes interface (#46) * Rework RMPI directory structure (#50) * Bugfix/mpi one sided with buffers (#51) * use Buffer interface with one-sided communication protocol. Needs `MPI.jl#master`! * add tests for new MPI version * fix package spec * make actions test new mpi scripts * fix another bug * make compatible with develop branch * MPI updates (#45) * MPI without memory leak * Split RMPI to directory * add async strategy * try slightly different exchange * Remove async, add all_to_all * docstrings, faster sorting * Add barrier * optimize sorting * better sort for alltoall * set default strategy to point-to-point * Rename mpi_default to point_to_point in test * Feature/fciqmc files (#52) * let git ignore vscode config files * split fciqmc.jl into multiple files; use user installed julia executable if available * add newline to .gitignore * New `StochasticStyle` (#54) * Implement new stochastic style * Fixes * fix diagonal step * undo fix diagonal step * Create new style for comparison * Fix errors in new style * Add docstring to and unexport `IsStochasticWithThresholdAndInitiator` * Initiator docstring * Fix unexporting * Fix `IsStochasticWithThresholdAndInititator`, add basic tests (#55) * Fix `IsStochasticWithThresholdAndInititator`, add basic tests * fix broken test * Feature/stats tools (#56) Note: After merging this PR, functions from `Blocking.jl` are no longer re-exported by `Rimu`. To access data analysis feature in scripts use ``` using Rimu, Rimu.StatsTools ``` or, to use the old features from `Blocking.jl` ``` using Rimu, Rimu.Blocking ``` * stop reexporting Blocking, include StatsTools * fix Blocking tests * new module StatsTools * add DataFrames dependency * update Project.toml * style * add @inbounds * fix tests actually * `smoothen()` now preserves mean and wraps periodically; remove kwarg `pad` * add tests for `StatsTools` * add blocking and ratio estimators to StatsTools; TODO: tests * rename block_inplace! to blocker! * tests for blocking.jl * use Parameters * ratio_of_means() * docstring for RatioBlockingResult * fix blocking for complex time series; TODO: tests * fix errors * doc string change * blocks field in BlockingResult; show method; docstrings * fix passing through significance levels * clean up * remove Parameters.jl from StatsTools * fix up ratio_of_means(), add tests * towards ratio_of_means() for complex time series; WIP * show() method for complex ratio_of_means, new tests * fix dependencies * fix module import * fix again * rename block_and_test() -> blocking_analysis(); new mean_and_se() fix \pm methods, median for comples, add tests * undo changes to Blocking; remove StructArrays dependency * test that show methods run without error * `AbstractMatrix` Hamiltonian fix (#57) * `AbstractMatrix` Hamiltonian fix * fix fciqmc_col! for matrix and vector; test matrix lomc! against Hamiltonian lomc! Co-authored-by: Joachim Brand <j.brand@massey.ac.nz> * Feature/matrix hamiltonian (#58) * `AbstractMatrix` Hamiltonian fix * fix fciqmc_col! for matrix and vector; test matrix lomc! against Hamiltonian lomc! * remove restriction to AbstractFockAddress in AbstractOffDiagonals * New wrapper MatrixHamiltonian for using abstract matrices for fciqmc. TODO: write tests * add LOStructure for MatrixHamiltonian; move tests to test/Hamiltonians.jl * remove starting address test; pass through dimension of importance sampled Hamiltonian Co-authored-by: mtsch <matijacufar@gmail.com> * Feature/cleanup stochastic styles (#59) * cleaning up unused StochasticStyles * fix errors * more cleanup * faster integerization in integer walker fciqmc * even more cleanup * DoubleLogSumUpdate * PopsProjector() for CW * NoProjectionAccumulator <: ProjectStrategy fix test for NoProjectionAccumulator * New `Norm1ProjectorPPop` for `dot`; added test; norm projectors now return floats * Update replica fciqmc!; Allow for complex walkers; report projectors; make xHy optional * replay changes to replica `fciqmc!()` * replay changes to `norm_project!()` add method for `NoProjectionAccumulator` * fix bugs in fciqmc_col! * fix compile and test errors Co-authored-by: Ray Yang <mingrui.yang@outlook.com> Co-authored-by: mtsch <matijacufar@gmail.com>
- Loading branch information
3 people
authored
May 13, 2021
1 parent
5c52287
commit 7a58a8d