More package extensions

.github/workflows/CI.yml

@@ -18,7 +18,6 @@ jobs:
       fail-fast: false
       matrix:
         version:
-          - '1.7'
           - '1'
         os:
           - ubuntu-latest

Project.toml

@@ -1,7 +1,7 @@
 name = "ITensorNetworks"
 uuid = "2919e153-833c-4bdc-8836-1ea460a35fc7"
 authors = ["Matthew Fishman <[email protected]> and contributors"]
-version = "0.9.0"
+version = "0.10.0"
 
 [deps]
 AbstractTrees = "1520ce14-60c1-5f80-bbc7-55ef81b5835c"
@@ -13,19 +13,16 @@ Dictionaries = "85a47980-9c8c-11e8-2b9f-f7ca1fa99fb4"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
 Graphs = "86223c79-3864-5bf0-83f7-82e725a168b6"
-GraphsFlows = "06909019-6f44-4949-96fc-b9d9aaa02889"
 ITensors = "9136182c-28ba-11e9-034c-db9fb085ebd5"
 IsApprox = "28f27b66-4bd8-47e7-9110-e2746eb8bed7"
 IterTools = "c8e1da08-722c-5040-9ed9-7db0dc04731e"
 KrylovKit = "0b1a1467-8014-51b9-945f-bf0ae24f4b77"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 NDTensors = "23ae76d9-e61a-49c4-8f12-3f1a16adf9cf"
 NamedGraphs = "678767b0-92e7-4007-89e4-4527a8725b19"
-Observers = "338f10d5-c7f1-4033-a7d1-f9dec39bcaa0"
 PackageExtensionCompat = "65ce6f38-6b18-4e1d-a461-8949797d7930"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
-Requires = "ae029012-a4dd-5104-9daa-d747884805df"
 SerializedElementArrays = "d3ce8812-9567-47e9-a7b5-65a6d70a3065"
 SimpleTraits = "699a6c99-e7fa-54fc-8d76-47d257e15c1d"
 SparseArrayKit = "a9a3c162-d163-4c15-8926-b8794fbefed2"
@@ -38,9 +35,15 @@ TupleTools = "9d95972d-f1c8-5527-a6e0-b4b365fa01f6"
 
 [weakdeps]
 EinExprs = "b1794770-133b-4de1-afb4-526377e9f4c5"
+GraphsFlows = "06909019-6f44-4949-96fc-b9d9aaa02889"
+Observers = "338f10d5-c7f1-4033-a7d1-f9dec39bcaa0"
+OMEinsumContractionOrders = "6f22d1fd-8eed-4bb7-9776-e7d684900715"
 
 [extensions]
 ITensorNetworksEinExprsExt = "EinExprs"
+ITensorNetworksGraphsFlowsExt = "GraphsFlows"
+ITensorNetworksObserversExt = "Observers"
+ITensorNetworksOMEinsumContractionOrdersExt = "OMEinsumContractionOrders"
 
 [compat]
 AbstractTrees = "0.4.4"
@@ -50,32 +53,35 @@ DataGraphs = "0.2.2"
 DataStructures = "0.18"
 Dictionaries = "0.4"
 Distributions = "0.25.86"
-DocStringExtensions = "0.8, 0.9"
+DocStringExtensions = "0.9"
 EinExprs = "0.6.4"
 Graphs = "1.8"
 GraphsFlows = "0.1.1"
-ITensors = "0.3.58, 0.4"
+ITensors = "0.4"
 IsApprox = "0.1"
 IterTools = "1.4.0"
 KrylovKit = "0.6, 0.7"
 NamedGraphs = "0.5.1"
-NDTensors = "0.2, 0.3"
+NDTensors = "0.3"
 Observers = "0.2"
+OMEinsumContractionOrders = "0.8.3"
 PackageExtensionCompat = "1"
-Requires = "1.3"
 SerializedElementArrays = "0.1"
 SimpleTraits = "0.9"
-SparseArrayKit = "0.2.1, 0.3"
+SparseArrayKit = "0.3"
 SplitApplyCombine = "1.2"
 StaticArrays = "1.5.12"
 StructWalk = "0.2"
 Suppressor = "0.2"
 TimerOutputs = "0.5.22"
 TupleTools = "1.4"
-julia = "1.7"
+julia = "1.10"
 
 [extras]
 EinExprs = "b1794770-133b-4de1-afb4-526377e9f4c5"
+GraphsFlows = "06909019-6f44-4949-96fc-b9d9aaa02889"
+Observers = "338f10d5-c7f1-4033-a7d1-f9dec39bcaa0"
+OMEinsumContractionOrders = "6f22d1fd-8eed-4bb7-9776-e7d684900715"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]

ext/ITensorNetworksGraphsFlowsExt/ITensorNetworksGraphsFlowsExt.jl

@@ -0,0 +1,19 @@
+module ITensorNetworksGraphsFlowsExt
+using Graphs: AbstractGraph
+using GraphsFlows: GraphsFlows
+using ITensorNetworks: ITensorNetworks
+using NDTensors.AlgorithmSelection: @Algorithm_str
+
+function ITensorNetworks.mincut(
+  ::Algorithm"GraphsFlows",
+  graph::AbstractGraph,
+  source_vertex,
+  target_vertex;
+  capacity_matrix,
+  alg=GraphsFlows.PushRelabelAlgorithm(),
+)
+  # TODO: Replace with `Backend(backend)`.
+  return GraphsFlows.mincut(graph, source_vertex, target_vertex, capacity_matrix, alg)
+end
+
+end

ext/ITensorNetworksOMEinsumContractionOrdersExt/ITensorNetworksOMEinsumContractionOrdersExt.jl

@@ -0,0 +1,185 @@
+module ITensorNetworksOMEinsumContractionOrdersExt
+using DocStringExtensions: TYPEDSIGNATURES
+using ITensorNetworks: ITensorNetworks
+using ITensors: ITensors, Index, ITensor, inds
+using NDTensors: dim
+using NDTensors.AlgorithmSelection: @Algorithm_str
+using OMEinsumContractionOrders: OMEinsumContractionOrders
+
+# OMEinsumContractionOrders wrapper for ITensors
+# Slicing is not supported, because it might require extra work to slice an `ITensor` correctly.
+
+const ITensorList = Union{Vector{ITensor},Tuple{Vararg{ITensor}}}
+
+# infer the output tensor labels
+# TODO: Use `symdiff` instead.
+function infer_output(inputs::AbstractVector{<:AbstractVector{<:Index}})
+  indslist = reduce(vcat, inputs)
+  # get output indices
+  iy = eltype(eltype(inputs))[]
+  for l in indslist
+    c = count(==(l), indslist)
+    if c == 1
+      push!(iy, l)
+    elseif c !== 2
+      error("Each index in a tensor network must appear at most twice!")
+    end
+  end
+  return iy
+end
+
+# get a (labels, size_dict) representation of a collection of ITensors
+function rawcode(tensors::ITensorList)
+  # we use id as the label
+  indsAs = [collect(Index{Int}, ITensors.inds(A)) for A in tensors]
+  ixs = collect.(inds.(tensors))
+  unique_labels = unique(reduce(vcat, indsAs))
+  size_dict = Dict([x => dim(x) for x in unique_labels])
+  index_dict = Dict([x => x for x in unique_labels])
+  return OMEinsumContractionOrders.EinCode(ixs, infer_output(indsAs)), size_dict, index_dict
+end
+
+"""
+$(TYPEDSIGNATURES)
+Optimize the contraction order of a tensor network specified as a vector tensors.
+Returns a [`NestedEinsum`](@ref) instance.
+### Examples
+```jldoctest
+julia> using ITensors, ITensorContractionOrders
+julia> i, j, k, l = Index(4), Index(5), Index(6), Index(7);
+julia> x, y, z = randomITensor(i, j), randomITensor(j, k), randomITensor(k, l);
+julia> net = optimize_contraction([x, y, z]; optimizer=TreeSA());
+```
+"""
+function optimize_contraction_nested_einsum(
+  tensors::ITensorList;
+  optimizer::OMEinsumContractionOrders.CodeOptimizer=OMEinsumContractionOrders.TreeSA(),
+)
+  r, size_dict, index_dict = rawcode(tensors)
+  # merge vectors can speed up contraction order finding
+  # optimize the permutation of tensors is set to true
+  res = OMEinsumContractionOrders.optimize_code(
+    r, size_dict, optimizer, OMEinsumContractionOrders.MergeVectors(), true
+  )
+  if res isa OMEinsumContractionOrders.SlicedEinsum   # slicing is not supported!
+    if length(res.slicing) != 0
+      @warn "Slicing is not yet supported by `ITensors`, removing slices..."
+    end
+    res = res.eins
+  end
+  return res
+end
+
+"""
+Convert NestedEinsum to contraction sequence, such as `[[1, 2], [3, 4]]`.
+"""
+function convert_to_contraction_sequence(net::OMEinsumContractionOrders.NestedEinsum)
+  if OMEinsumContractionOrders.isleaf(net)
+    return net.tensorindex
+  else
+    return convert_to_contraction_sequence.(net.args)
+  end
+end
+
+"""
+Convert the result of `optimize_contraction` to a contraction sequence.
+"""
+function optimize_contraction_sequence(
+  tensors::ITensorList; optimizer::OMEinsumContractionOrders.CodeOptimizer=TreeSA()
+)
+  res = optimize_contraction_nested_einsum(tensors; optimizer)
+  return convert_to_contraction_sequence(res)
+end
+
+"""
+    GreedyMethod(; method=MinSpaceOut(), nrepeat=10)
+
+The fast but poor greedy optimizer. Input arguments are:
+
+* `method` is `MinSpaceDiff()` or `MinSpaceOut`.
+    * `MinSpaceOut` choose one of the contraction that produces a minimum output tensor size,
+    * `MinSpaceDiff` choose one of the contraction that decrease the space most.
+* `nrepeat` is the number of repeatition, returns the best contraction order.
+"""
+function ITensorNetworks.contraction_sequence(
+  ::Algorithm"greedy", tn::Vector{ITensor}; kwargs...
+)
+  return optimize_contraction_sequence(
+    tn; optimizer=OMEinsumContractionOrders.GreedyMethod(; kwargs...)
+  )
+end
+
+"""
+    TreeSA(; sc_target=20, βs=collect(0.01:0.05:15), ntrials=10, niters=50,
+             sc_weight=1.0, rw_weight=0.2, initializer=:greedy, greedy_config=GreedyMethod(; nrepeat=1))
+
+Optimize the einsum contraction pattern using the simulated annealing on tensor expression tree.
+
+* `sc_target` is the target space complexity,
+* `ntrials`, `βs` and `niters` are annealing parameters, doing `ntrials` indepedent annealings, each has inverse tempteratures specified by `βs`, in each temperature, do `niters` updates of the tree.
+* `sc_weight` is the relative importance factor of space complexity in the loss compared with the time complexity.
+* `rw_weight` is the relative importance factor of memory read and write in the loss compared with the time complexity.
+* `initializer` specifies how to determine the initial configuration, it can be `:greedy` or `:random`. If it is using `:greedy` method to generate the initial configuration, it also uses two extra arguments `greedy_method` and `greedy_nrepeat`.
+* `nslices` is the number of sliced legs, default is 0.
+* `fixed_slices` is a vector of sliced legs, default is `[]`.
+
+### References
+* [Recursive Multi-Tensor Contraction for XEB Verification of Quantum Circuits](https://arxiv.org/abs/2108.05665)
+"""
+function ITensorNetworks.contraction_sequence(::Algorithm"tree_sa", tn; kwargs...)
+  return optimize_contraction_sequence(
+    tn; optimizer=OMEinsumContractionOrders.TreeSA(; kwargs...)
+  )
+end
+
+"""
+    SABipartite(; sc_target=25, ntrials=50, βs=0.1:0.2:15.0, niters=1000
+                  max_group_size=40, greedy_config=GreedyMethod(), initializer=:random)
+
+Optimize the einsum code contraction order using the Simulated Annealing bipartition + Greedy approach.
+This program first recursively cuts the tensors into several groups using simulated annealing,
+with maximum group size specifed by `max_group_size` and maximum space complexity specified by `sc_target`,
+Then finds the contraction order inside each group with the greedy search algorithm. Other arguments are:
+
+* `size_dict`, a dictionary that specifies leg dimensions,
+* `sc_target` is the target space complexity, defined as `log2(number of elements in the largest tensor)`,
+* `max_group_size` is the maximum size that allowed to used greedy search,
+* `βs` is a list of inverse temperature `1/T`,
+* `niters` is the number of iteration in each temperature,
+* `ntrials` is the number of repetition (with different random seeds),
+* `greedy_config` configures the greedy method,
+* `initializer`, the partition configuration initializer, one can choose `:random` or `:greedy` (slow but better).
+
+### References
+* [Hyper-optimized tensor network contraction](https://arxiv.org/abs/2002.01935)
+"""
+function ITensorNetworks.contraction_sequence(::Algorithm"sa_bipartite", tn; kwargs...)
+  return optimize_contraction_sequence(
+    tn; optimizer=OMEinsumContractionOrders.SABipartite(; kwargs...)
+  )
+end
+
+"""
+    KaHyParBipartite(; sc_target, imbalances=collect(0.0:0.005:0.8),
+                       max_group_size=40, greedy_config=GreedyMethod())
+
+Optimize the einsum code contraction order using the KaHyPar + Greedy approach.
+This program first recursively cuts the tensors into several groups using KaHyPar,
+with maximum group size specifed by `max_group_size` and maximum space complexity specified by `sc_target`,
+Then finds the contraction order inside each group with the greedy search algorithm. Other arguments are:
+
+* `sc_target` is the target space complexity, defined as `log2(number of elements in the largest tensor)`,
+* `imbalances` is a KaHyPar parameter that controls the group sizes in hierarchical bipartition,
+* `max_group_size` is the maximum size that allowed to used greedy search,
+* `greedy_config` is a greedy optimizer.
+
+### References
+* [Hyper-optimized tensor network contraction](https://arxiv.org/abs/2002.01935)
+* [Simulating the Sycamore quantum supremacy circuits](https://arxiv.org/abs/2103.03074)
+"""
+function ITensorNetworks.contraction_sequence(::Algorithm"kahypar_bipartite", tn; kwargs...)
+  return optimize_contraction_sequence(
+    tn; optimizer=OMEinsumContractionOrders.KaHyParBipartite(; kwargs...)
+  )
+end
+end

ext/ITensorNetworksObserversExt/ITensorNetworksObserversExt.jl

@@ -0,0 +1,9 @@
+module ITensorNetworksObserversExt
+using ITensorNetworks: ITensorNetworks
+using Observers.DataFrames: AbstractDataFrame
+using Observers: Observers
+
+function ITensorNetworks.update_observer!(observer::AbstractDataFrame; kwargs...)
+  return Observers.update!(observer; kwargs...)
+end
+end

src/ITensorNetworks.jl

@@ -1,7 +1,6 @@
 module ITensorNetworks
 include("lib/BaseExtensions/src/BaseExtensions.jl")
 include("lib/ITensorsExtensions/src/ITensorsExtensions.jl")
-include("observers.jl")
 include("visualize.jl")
 include("graphs.jl")
 include("abstractindsnetwork.jl")
@@ -33,6 +32,7 @@ include("caches/beliefpropagationcache.jl")
 include("contraction_tree_to_graph.jl")
 include("gauging.jl")
 include("utils.jl")
+include("update_observer.jl")
 include("solvers/local_solvers/eigsolve.jl")
 include("solvers/local_solvers/exponentiate.jl")
 include("solvers/local_solvers/dmrg_x.jl")
@@ -66,11 +66,7 @@ include("lib/ModelHamiltonians/src/ModelHamiltonians.jl")
 include("lib/ModelNetworks/src/ModelNetworks.jl")
 
 using PackageExtensionCompat: @require_extensions
-using Requires: @require
 function __init__()
   @require_extensions
-  @require OMEinsumContractionOrders = "6f22d1fd-8eed-4bb7-9776-e7d684900715" include(
-    "requires/omeinsumcontractionorders.jl"
-  )
 end
 end

src/apply.jl

@@ -27,7 +27,6 @@ using ITensors.ITensorMPS: siteinds
 using KrylovKit: linsolve
 using LinearAlgebra: eigen, norm, svd
 using NamedGraphs: NamedEdge, has_edge
-using Observers: Observers
 
 function full_update_bp(
   o,

src/contract_approx/mincut.jl

@@ -1,8 +1,8 @@
 using AbstractTrees: Leaves, PostOrderDFS
 using Combinatorics: powerset
 using Graphs: dijkstra_shortest_paths, weights
-using GraphsFlows: GraphsFlows
 using NamedGraphs: NamedDiGraph
+using NDTensors.AlgorithmSelection: Algorithm
 
 # a large number to prevent this edge being a cut
 MAX_WEIGHT = 1e32
@@ -37,6 +37,18 @@ function binary_tree_structure(tn::ITensorNetwork, outinds::Vector)
   return _binary_tree_structure(tn, outinds; maximally_unbalanced=false)
 end
 
+function mincut(graph::AbstractGraph, source_vertex, target_vertex; backend, kwargs...)
+  # TODO: Replace with `Backend(backend)`.
+  return mincut(Algorithm(backend), graph, source_vertex, target_vertex; kwargs...)
+end
+
+# TODO: Replace with `backend::Backend`.
+function mincut(
+  backend::Algorithm, graph::AbstractGraph, source_vertex, target_vertex; kwargs...
+)
+  return error("Backend `$backend` not implemented for `mincut`.")
+end
+
 """
 Calculate the mincut between two subsets of the uncontracted inds
 (source_inds and terminal_inds) of the input tn.
@@ -52,7 +64,7 @@ function _mincut(tn::ITensorNetwork, source_inds::Vector, terminal_inds::Vector)
   tn = disjoint_union(
     ITensorNetwork([ITensor(source_inds...), ITensor(terminal_inds...)]), tn
   )
-  return GraphsFlows.mincut(tn, (1, 1), (2, 1), weights(tn))
+  return mincut(tn, (1, 1), (2, 1); backend="GraphsFlows", capacity_matrix=weights(tn))
 end
 
 """