Latticesupports specifying arbitrary edge content for each unit cell via the kwargcustom_edges. A generator for hexagonal lattices with coloured edges is implemented asnk.graph.KitaevHoneycomb.nk.graph.Gridagain supports colouring edges by direction. #1074- Fermionic hilbert space (
nkx.hilbert.SpinOrbitalFermions) and fermionic operators (nkx.operator.fermion) to treat systems with a finite number of Orbitals have been added to the experimental submodule. The operators are also integrated with OpenFermion. Those functionalities are still in development and we would welcome feedback. #1090
- The gradient for models with real-parameter is now multiplied by 2. If your model had real parameters you might need to change the learning rate and halve it. Conceptually this is a bug-fix, as the value returned before was wrong (see Bug Fixes section below for additional details) #1069
- The gradient obtained with
VarState.expect_and_gradfor models with real-parameters was off by a factor of $ 1/2 $ from the correct value. This has now been corrected. As a consequence, the correct gradient for real-parameter models is equal to the old times 2. If your model had real parameters you might need to change the learning rate and halve it. #1069 - Support for coloured edges in
nk.graph.Grid, removed in #724, is now restored. #1074 - Fixed bug that prevented calling
.quantum_geometric_tensoronnetket.vqs.ExactState. #1108 - Fixed bug where the gradient of
C->Cmodels (complex parameters, complex output) was computed incorrectly withnk.vqs.ExactState. #1110 - Fixed bug where
QGTJacobianDense.stateandQGTJacobianPyTree.statewould not correctly transform the starting pointx0ifholomorphic=False. #1115
- Support for Python 3.10 #952.
- The minimum optax version is now
0.1.1, which finally correctly supports complex numbers. The internal implementation of Adam which was introduced in 3.3 (#1069) has been removed. If an older version ofoptaxis detected, an import error is thrown to avoid providing wrong numerical results. Please update your optax version! #1097
- Allow
LazyOperator@densevectorfor operators such as lazyAdjoint,TransposeandSquared. #1068 - The logic to update the progress bar in
nk.experimental.TDVPhas been improved, and it should now display updates even if there are very sparsesave_steps. #1084 - The
nk.logging.TensorBoardLogis now lazily initialized to better work in an MPI environment. #1086 - Converting a
nk.operator.BoseHubbardto ank.operator.LocalOperatormultiplied by 2 the nonlinearityU. This has now been fixed. #1102
- Initialisation of all implementations of
DenseSymm,DenseEquivariant,GCNNnow defaults to truncated normals with Lecun variance scaling. For layers without masking, there should be no noticeable change in behaviour. For masked layers, the same variance scaling now works correctly. #1045 - Fix bug that prevented gradients of non-hermitian operators to be computed. The feature is still marked as experimental but will now run (we do not guarantee that results are correct). #1053
- Common lattice constructors such as
Honeycombnow accepts the same keyword arguments asLattice. #1046 - Multiplying a
QGTOnTheFlyrepresenting the real part of the QGT (showing up when the ansatz has real parameters) with a complex vector now throws an error. Previously the result would be wrong, as the imaginary part was casted away. #885
- The interface to define expectation and gradient function of arbitrary custom operators is now stable. If you want to define it for a standard operator that can be written as an average of local expectation terms, you can now define a dispatch rule for {ref}
netket.vqs.get_local_kernel_argumentsand {ref}netket.vqs.get_local_kernel. The old mechanism is still supported, but we encourage to use the new mechanism as it is more terse. #954 nk.optimizer.Adamnow supports complex parameters, and you can usenk.optimizer.split_complexto make optimizers process complex parameters as if they are pairs of real parameters. #1009- Chunking of
MCState.expectandMCState.expect_and_gradcomputations is now supported, which allows to bound the memory cost in exchange of a minor increase in computation time. #1006 (and discussions in #918 and #830) - A new variational state that performs exact summation over the whole Hilbert space has been added. It can be constructed with {ref}
nk.vqs.ExactStateand supports the same Jax neural networks as {ref}nk.vqs.MCState. #953 DenseSymmallows multiple input features. #1030- [Experimental] A new time-evolution driver {ref}
nk.experimental.TDVPusing the time-dependent variational principle (TDVP) has been added. It works with time-independent and time-dependent Hamiltonians and Liouvillians. #1012 - [Experimental] A set of JAX-compatible Runge-Kutta ODE integrators has been added for use together with the new TDVP driver. #1012
- The method
sample_nextinSamplerand exact samplers (ExactSamplerandARDirectSampler) is removed, and it is only defined inMetropolisSampler. The module functionnk.sampler.sample_nextalso only works withMetropolisSampler. For exact samplers, please use the methodsampleinstead. #1016 - The default value of
n_chains_per_rankinSamplerand exact samplers is changed to 1, and specifyingn_chainsorn_chains_per_rankwhen constructing them is deprecated. Please changechain_lengthwhen callingsample. ForMetropolisSampler, the default value is changed fromn_chains = 16(across all ranks) ton_chains_per_rank = 16. #1017 GCNN_Parityallowed biasing both the parity-preserving and the parity-flip equivariant layers. These enter into the network output the same way, so having both is redundant and makes QGTs unstable. The biases of the parity-flip layers are now removed. The previous behaviour can be restored using the deprecatedextra_biasswitch; we only recommend this for loading previously saved parameters. Such parameters can be transformed to work with the new default usingnk.models.update_GCNN_parity. #1030- Kernels of
DenseSymmare now three-dimensional, not two-dimensional. Parameters saved from earlier implementations can be transformed to the new convention usingnk.nn.update_dense_symm. #1030
- The method
Sampler.samplesis added to return a generator of samples. The module functionsnk.sampler.sampler_state,reset,sample,samples, andsample_nextare deprecated in favor of the corresponding class methods. #1025 - Kwarg
in_featuresofDenseEquivariantis deprecated; the number of input features are inferred from the input. #1030 - Kwarg
out_featuresofDenseEquivariantis deprecated in favour offeatures. #1030
- The definitions of
MCStateandMCMixedStatehave been moved to an internal module,nk.vqs.mcthat is hidden by default. #954 - Custom deepcopy for
LocalOperatorto avoid buildingLocalOperatorfrom scratch each time it is copied #964
- The constructor of
TensorHilbert(which is used by the product operator*for inhomogeneous spaces) no longer fails when one of the component spaces is non-indexable. #1004 - The {ref}
nk.hilbert.random.flip_statemethod used byMetropolisLocalnow throws an error when called on a {ref}nk.hilbert.ContinuousHilberthilbert space instead of entering an endless loop. #1014 - Fixed bug in conversion to qutip for
MCMixedState, where the resulting shape (hilbert space size) was wrong. #1020 - Setting
MCState.samplernow recomputesMCState.chain_lengthaccording toMCState.n_samplesand the newsampler.n_chains. #1028 GCNN_Parityallowed biasing both the parity-preserving and the parity-flip equivariant layers. These enter into the network output the same way, so having both is redundant and makes QGTs unstable. The biases of the parity-flip layers are now removed. #1030
GraphOperator(andHeisenberg) now support passing a custom mapping of graph nodes to Hilbert space sites via the newacting_on_subspaceargument. This makes it possible to createGraphOperators that act on a subset of sites, which is useful in composite Hilbert spaces. #924PauliStringnow supports any Hilbert space with local size 2. The Hilbert space is now the optional first argument of the constructor. #960PauliStringnow can be multiplied and summed together, performing some simple algebraic simplifications on the strings they contain. They also lazily initialize their internal data structures, making them faster to construct but slightly slower the first time that their matrix elements are accessed. #955PauliStrings can now be constructed starting from anOpenFermionoperator. #956- In addition to nearest-neighbor edges,
Latticecan now generate edges between next-nearest and, more generally, k-nearest neighbors via the constructor argumentmax_neighbor_order. The edges can be distinguished by theircolorproperty (which is used, e.g., byGraphOperatorto apply different bond operators). #970 - Two continuous-space operators (
KineticEnergyandPotentialEnergy) have been implemented. #971 HeisenbergHamiltonians support different coupling strengths onGraphedges with different colors. #972.- The
little_groupandspace_group_irrepsmethods ofSpaceGroupBuildertake the wave vector as either varargs or iterables. #975 - A new
netket.experimentalsubmodule has been created and all experimental features have been moved there. Note that in contrast to the othernetketsubmodules,netket.experimentalis not imported by default. #976
- Moved
nk.vqs.variables_from_***tonk.experimental.vqsmodule. Also moved the experimental samplers tonk.sampler.MetropolisPtandnk.sampler.MetropolisPmaptonk.experimental.sampler. #976 operator.size, has been deprecated. If you were using this function, please transition tooperator.hilbert.size. #985
- A bug where
LocalOperator.get_conn_flattenedwould read out-of-bounds memory has been fixed. It is unlikely that the bug was causing problems, but it triggered warnings when running Numba with boundscheck activated. #966 - The dependency
python-igraphhas been updated toigraphfollowing the rename of the upstream project in order to work on conda. #986 - {attr}
~netket.vqs.MCState.n_samples_per_rankwas returning wrong values and has now been fixed. #987 - The
DenseSymmlayer now also accepts objects of typeHashableArrayassymmetriesargument. #989 - A bug where
VMC.info()was erroring has been fixed. #984
- Added Conversion methods
to_qobj()to operators and variational states, that produce QuTiP's qobjects. - A function
nk.nn.activation.reimhas been added that transforms a nonlinearity to act seperately on the real and imaginary parts - Nonlinearities
reim_seluandreim_reluhave been added - Autoregressive Neural Networks (ARNN) now have a
machine_powfield (defaults to 2) used to change the exponent used for the normalization of the wavefunction. #940.
- The default initializer for
netket.models.GCNNhas been changed to fromjax.nn.selutonetket.nn.reim_selu#892 netket.nn.initializershas been deprecated in favor ofjax.nn.initializers#935.- Subclasses of
AbstractARNNmust define the fieldmachine_pow#940 nk.hilbert.HilbertIndexandnk.operator.spin.DTypeare now unexported (they where never intended to be visible). #904AbstractOperators have been renamedDiscreteOperators.AbstractOperators still exist, but have almost no functionality and they are intended as the base class for more arbitrary (eg. continuous space) operators. If you have defined a custom operator inheriting fromAbstractOperatoryou should change it to derive fromDiscreteOperator. #929
PermutationGroup.product_tablenow consumes less memory and is more performant. This is helpfull when working with large symmetry groups. #884 #891- Added size check to
DiscreteOperator.get_connand throw helpful error messages if those do not match. #927 - The internal
numba4jaxmodule has been factored out into a standalone library, named (how original)numba4jax. This library was never intended to be used by external users, but if for any reason you were using it, you should switch to the external library. #934 netket.jaxnow includes several batching utilities likebatched_vmapandbatched_vjp. Those can be used to build memory efficient batched code, but are considered internal, experimental and might change without warning. #925.
- Autoregressive networks now work with
Qubithilbert spaces. #937
- The default initializer for
netket.nn.Denselayers now matches the same default asflax.linen, and it islecun_normalinstead ofnormal(0.01)#869 - The default initializer for
netket.nn.DenseSymmlayers is now chosen in order to give variance 1 to every output channel, therefore defaulting tolecun_normal#870
- DenseSymm now accepts a mode argument to specify whever the symmetries should be computed with a full dense matrix or FFT. The latter method is much faster for sufficiently large systems. Other kwargs have been added to satisfy the interface. The api changes are also reflected in RBMSymm and GCNN. #792
- The so-called legacy netket in
netket.legacyhas been removed. #773
- The methods
expectandexpect_and_gradofMCStatenow use dispatch to select the relevant implementation of the algorithm. They can therefore be expanded and overridden without editing NetKet's source code. #804 netket.utils.mpi_availablehas been moved tonetket.utils.mpi.availableto have a more consistent api interface (all mpi-related properties in the same submodule). #827netket.logging.TBLoghas been renamed tonetket.logging.TensorBoardLogfor better readability. A deprecation warning is now issued if the older name is used #827- When
MCStateinitializes a model by callingmodel.init, the call is now jitted. This should speed it up for non-trivial models but might break non-jit invariant models. #832 operator.get_conn_paddednow supports arbitrarily-dimensioned bitstrings as input and reshapes the output accordingly. #834- NetKet's implementation of dataclasses now support
pytree_node=True/Falseon cached properties. #835 - Plum version has been bumped to 1.5.1 to avoid broken versions (1.4, 1.5). #856.
- Numba version 0.54 is now allowed #857.
- Fix Progress bar bug. #810
- Make the repr/printing of history objects nicer in the REPL. #819
- The field
MCState.modelis now read-only, to prevent user errors. #822 - The order of the operators in
PauliStringdoes no longer influences the estimate of the number of non-zero connected elements. #836
- The {ref}
netket.utils.groupsubmodule provides utilities for geometrical and permutation groups.Lattice(and its specialisations likeGrid) use these to automatically construct the space groups of lattices, as well as their character tables for generating wave functions with broken symmetry. #724 - Autoregressive neural networks, sampler, and masked linear layers have been added to
models,samplerandnn#705.
- The
netket.graph.Gridclass has been removed. {ref}netket.graph.Gridwill now return an instance of {ref}graph.Latticesupporting the same API but with new functionalities related to spatial symmetries. Thecolor_edgesoptional keyword argument has been removed without deprecation. #724 MCState.n_discardhas been renamedMCState.n_discard_per_chainand the old binding has been deprecated #739.nk.optimizer.qgt.QGTOnTheFlyoptioncentered=Truehas been removed because we are now convinced the two options yielded equivalent results.QGTOnTheFlynow always behaves as ifcentered=False#706.
networkXhas been replaced byigraph, yielding a considerable speedup for some graph-related operations #729.netket.hilbert.randommodule now usesplum-dispatch(throughnetket.utils.dispatch) to select the correct implementation ofrandom_stateandflip_state. This makes it easy to define new hilbert states and extend their functionality easily. #734.- The AbstractHilbert interface is now much smaller in order to also support continuous Hilbert spaces. Any functionality specific to discrete hilbert spaces (what was previously supported) has been moved to a new abstract type
netket.hilbert.DiscreteHilbert. Any Hilbert space previously subclassing {ref}netket.hilbert.AbstractHilbertshould be modified to subclass {ref}netket.hilbert.DiscreteHilbert#800.
nn.to_arrayandMCState.to_array, ifnormalize=False, do not subtract the logarithm of the maximum value from the state #705.- Autoregressive networks now work with Fock space and give correct errors if the hilbert space is not supported #806.
- Autoregressive networks are now much (x10-x100) faster #705.
- Do not throw errors when calling
operator.get_conn_flattened(states)with a jax array #764. - Fix bug with the driver progress bar when
step_size != 1#747.
- Group Equivariant Neural Networks have been added to
models#620 - Permutation invariant RBM and Permutation invariant dense layer have been added to
modelsandnn.linear#573 - Add the property
acceptancetoMetropolisSampler'sSamplerState, computing the MPI-enabled acceptance ratio. #592. - Add
StateLog, a new logger that stores the parameters of the model during the optimization in a folder or in a tar file. #645 - A warning is now issued if NetKet detects to be running under
mpirunbut MPI dependencies are not installed #631 operator.LocalOperators now do not return a zero matrix element on the diagonal if the whole diagonal is zero. #623.logger.JSONLognow automatically flushes at every iteration if it does not consume significant CPU cycles. #599- The interface of Stochastic Reconfiguration has been overhauled and made more modular. You can now specify the solver you wish to use, NetKet provides some dense solvers out of the box, and there are 3 different ways to compute the Quantum Geometric Tensor. Read the documentation to learn more about it. #674
- Unless you specify the QGT implementation you wish to use with SR, we use an automatic heuristic based on your model and the solver to pick one. This might affect SR performance. #674
- For all samplers,
n_chainsnow sets the total number of chains across all MPI ranks. This is a breaking change compared to the old API, wheren_chainswould set the number of chains on a single MPI rank. It is still possible to set the number of chains per MPI rank by specifyingn_chains_per_rankinstead ofn_chains. This change, while breaking allows us to be consistent with the interface of {ref}variational.MCState, wheren_samplesis the total number of samples across MPI nodes. MetropolisSampler.reset_chainhas been renamed toMetropolisSampler.reset_chains. Likewise in the constructor of all samplers.- Briefly during development releases
MetropolisSamplerState.acceptance_ratioreturned the percentage (not ratio) of acceptance.acceptance_ratiois now deprecated in favour of the correctacceptance. models.Jastrownow internally symmetrizes the matrix before computing its value #644MCState.evaluatehas been renamed toMCState.log_value#632nk.optimizer.SRno longer accepts keyword argument relative to the sparse solver. Those should be passed inside the closure orfunctools.partialpassed assolverargument.nk.optimizer.sr.SRLazyCGandnk.optimizer.sr.SRLazyGMREShave been deprecated and will soon be removed.- Parts of the
LatticeAPI have been overhauled, with deprecations of several methods in favor of a consistent usage ofLattice.positionfor real-space location of sites andLattice.basis_coordsfor location of sites in terms of basis vectors.Lattice.siteshas been added, which provides a sequence ofLatticeSiteobjects combining all site properties. Furthermore,Latticenow provides lookup of sites from their position viaid_from_positionusing a hashing scheme that works across periodic boundaries. #703 #715 nk.variationalhas been renamed tonk.vqsand will be removed in a future release.
- Fix
operator.BoseHubbardusage under jax Hamiltonian Sampling #662 - Fix
SROnTheFlyforR->Cmodels with non homogeneous parameters #661 - Fix MPI Compilation deadlock when computing expectation values #655
- Fix bug preventing the creation of a
hilbert.SpinHilbert space with odd sites and evenS. #641 - Fix bug #635 preventing the usage of
NumpyMetropolisSamplerwithMCState.expect#635 - Fix bug #635 where the
graph.Latticewas not correctly computing neighbours because of floating point issues. #633 - Fix bug the Y Pauli matrix, which was stored as its conjugate. #618 #617 #615
-
Hilbert space constructors do not store the lattice graph anymore. As a consequence, the constructor does not accept the graph anymore.
-
Special Hamiltonians defined on a lattice, such as {class}
operator.BoseHubbard, {class}operator.Isingand {class}operator.Heisenberg, now require the graph to be passed explicitly through agraphkeyword argument. -
{class}
operator.LocalOperatornow default to real-valued matrix elements, except if you construct them with a complex-valued matrix. This is also valid for operators such as :func:operator.spin.sigmaxand similars. -
When performing algebraic operations {code}
*, -, +on pairs of {class}operator.LocalOperator, the dtype of the result iscomputed using standard numpy promotion logic.- Doing an operation in-place {code}
+=, -=, *=on a real-valued operator will now fail if the other is complex. While this might seem annoying, it's useful to ensure that smaller types such asfloat32orcomplex64are preserved if the user desires to do so.
- Doing an operation in-place {code}
-
{class}
AbstractMachinehas been removed. It's functionality is now split among the model itself, which is defined by the user and {class}variational.MCStatefor pure states or {class}variational.MCMixedStatefor mixed states.-
The model, in general is composed by two functions, or an object with two functions: an
init(rng, sample_val)function, accepting a {func}jax.random.PRNGKeyobject and an input, returning the parameters and the state of the model for that particular sample shape, and a {code}apply(params, samples, **kwargs)function, evaluating the model for the given parameters and inputs. -
Some models (previously machines) such as the RBM (Restricted Boltzmann Machine) Machine, NDM (Neural Density Matrix) or MPS (Matrix Product State ansatz) are available in {ref}
Pre-built models. -
Machines, now called models, should be written using Flax or another jax framework.
-
Serialization and deserialization functionality has now been moved to {ref}
netket.variational.MCState, which support the standard Flax interface through MsgPack. See Flax docs for more information -
{code}
AbstractMachine.init_random_parametersfunctionality has now been absorbed into {meth}netket.vqs.VariationalState.init_parameters, which however has a different syntax.
-
-
{ref}
Samplers <Sampler>now require the Hilbert space upon which they sample to be passed in to the constructor. Also note that several keyword arguments of the samplers have changed, and new one are available. -
It's now possible to change {ref}
Samplers <Sampler>dtype, which controls the type of the output. By default they use double-precision samples (np.float64). Be wary of type promotion issues with your models. -
{ref}
Samplers <Sampler>no longer take a machine as an argument. -
{ref}
Samplers <Sampler>are now immutable (frozen)dataclasses(defined throughflax.struct.dataclass) that only hold the sampling parameters. As a consequence it is no longer possible to change their settings such asn_chainsorn_sweepswithout creating a new sampler. If you wish to update only one parameter, it is possible to construct the new sampler with the updated value by using thesampler.replace(parameter=new_value)function. -
{ref}
Samplers <Sampler>are no longer stateful objects. Instead, they can construct an immutable state object {ref}netket.sampler.init_state, which can be passed to sampling functions such as {ref}netket.sampler.sample, which now return also the updated state. However, unless you have particular use-cases we advise you use the variational state {ref}MCStateinstead. -
The {ref}
netket.optimizermodule has been overhauled, and now only re-exports flax optim module. We advise not to use netket's optimizer but instead to use optax . -
The {ref}
netket.optimizer.SRobject now is only a set of options used to compute the SR matrix. The SR matrix, now calledquantum_geometric_tensorcan be obtained by calling {meth}variational.MCState.quantum_geometric_tensor. Depending on the settings, this can be a lazy object. -
netket.Vmchas been renamed to {ref}netket.VMC -
{ref}
netket.models.RBMreplaces the old {code}RBMmachine, but has real parameters by default. -
As we rely on Jax, using {code}
dtype=floator {code}dtype=complex, which are weak types, will sometimes lead to loss of precision because they might be converted tofloat32. Use {code}np.float64or {code}np.complex128instead if you want double precision when defining your models.