Merge branch 'develop' into feature-svb-qol-updates

.gitignore

@@ -32,6 +32,7 @@ doc/build
 *gst_checkpoints*
 *model_test_checkpoints*
 *standard_gst_checkpoints*
+*ibmqexperiment_checkpoint*
 
 # Serialization Testing Artifacts #
 ###################################
@@ -44,6 +45,7 @@ test/output/pylint/*
 test/output/individual_coverage/*/*
 test/test_packages/cmp_chk_files/Fake_Dataset_none.txt.cache
 **.noseids
+**test_ibmq**
 
 # Tutorial Notebook Untracked Files #
 ####################################

pygsti/algorithms/core.py

@@ -33,6 +33,8 @@
 from pygsti.baseobjs.resourceallocation import ResourceAllocation as _ResourceAllocation
 from pygsti.optimize.customlm import CustomLMOptimizer as _CustomLMOptimizer
 from pygsti.optimize.customlm import Optimizer as _Optimizer
+from pygsti import forwardsims as _fwdsims
+from pygsti import layouts as _layouts
 
 _dummy_profiler = _DummyProfiler()
 
@@ -400,7 +402,7 @@ def _construct_ab(prep_fiducials, effect_fiducials, model, dataset, op_label_ali
         for j, rhostr in enumerate(prep_fiducials):
             opLabelString = rhostr + estr  # LEXICOGRAPHICAL VS MATRIX ORDER
             dsStr = opLabelString.replace_layers_with_aliases(op_label_aliases)
-            expd_circuit_outcomes = opLabelString.expand_instruments_and_separate_povm(model)
+            expd_circuit_outcomes = model.expand_instruments_and_separate_povm(opLabelString)
             assert(len(expd_circuit_outcomes) == 1), "No instruments are allowed in LGST fiducials!"
             unique_key = next(iter(expd_circuit_outcomes.keys()))
             outcomes = expd_circuit_outcomes[unique_key]
@@ -429,7 +431,7 @@ def _construct_x_matrix(prep_fiducials, effect_fiducials, model, op_label_tuple,
         for j, rhostr in enumerate(prep_fiducials):
             opLabelString = rhostr + _circuits.Circuit(op_label_tuple, line_labels=rhostr.line_labels) + estr
             dsStr = opLabelString.replace_layers_with_aliases(op_label_aliases)
-            expd_circuit_outcomes = opLabelString.expand_instruments_and_separate_povm(model)
+            expd_circuit_outcomes = model.expand_instruments_and_separate_povm(opLabelString)
             dsRow_fractions = dataset[dsStr].fractions
             assert(len(expd_circuit_outcomes) == nVariants)
 
@@ -677,16 +679,10 @@ def run_gst_fit(mdc_store, optimizer, objective_function_builder, verbosity=0):
         if _np.linalg.norm(mdc_store.model.to_vector() - v_cmp) > 1e-6:
             raise ValueError("MPI ERROR: *different* MC2GST start models"
                              " given to different processors!")                   # pragma: no cover
-
-    #MEM from ..baseobjs.profiler import Profiler
-    #MEM debug_prof = Profiler(comm)
-    #MEM debug_prof.print_memory("run_gst_fit1", True)
-
+
     if objective_function_builder is not None:
         objective_function_builder = _objfns.ObjectiveFunctionBuilder.cast(objective_function_builder)
-        #MEM debug_prof.print_memory("run_gst_fit2", True)
         objective = objective_function_builder.build_from_store(mdc_store, printer)  # (objective is *also* a store)
-        #MEM debug_prof.print_memory("run_gst_fit3", True)
     else:
         assert(isinstance(mdc_store, _objfns.ObjectiveFunction)), \
             "When `objective_function_builder` is None, `mdc_store` must be an objective fn!"
@@ -705,14 +701,8 @@ def run_gst_fit(mdc_store, optimizer, objective_function_builder, verbosity=0):
 
     printer.log("Completed in %.1fs" % (_time.time() - tStart), 1)
 
-    #if target_model is not None:
-    #  target_vec = target_model.to_vector()
-    #  targetErrVec = _objective_func(target_vec)
-    #  return minErrVec, soln_gs, targetErrVec
     profiler.add_time("do_mc2gst: total time", tStart)
-    #TODO: evTree.permute_computation_to_original(minErrVec) #Doesn't work b/c minErrVec is flattened
-    # but maybe best to just remove minErrVec from return value since this isn't very useful
-    # anyway?
+
     return opt_result, objective
 
 
@@ -888,10 +878,30 @@ def _max_array_types(artypes_list):  # get the maximum number of each array type
     #The ModelDatasetCircuitsStore
     printer.log('Precomputing CircuitOutcomeProbabilityArray layouts for each iteration.', 2)
     precomp_layouts = []
+
+    #pre-compute a dictionary caching completed circuits for layout construction performance.
+    unique_circuits = list({ckt for circuit_list in circuit_lists for ckt in circuit_list})
+    if isinstance(mdl.sim, (_fwdsims.MatrixForwardSimulator, _fwdsims.MapForwardSimulator)):
+        precomp_layout_circuit_cache = mdl.sim.create_copa_layout_circuit_cache(unique_circuits, mdl, dataset=dataset)
+    else:
+        precomp_layout_circuit_cache = None
+
     for i, circuit_list in enumerate(circuit_lists):
         printer.log(f'Layout for iteration {i}', 2)
-        precomp_layouts.append(mdl.sim.create_layout(circuit_list, dataset, resource_alloc, array_types, verbosity= printer - 1))
-
+        precomp_layouts.append(mdl.sim.create_layout(circuit_list, dataset, resource_alloc, array_types, verbosity= printer - 1,
+                                                    layout_creation_circuit_cache = precomp_layout_circuit_cache))
+
+    #precompute a cache of possible outcome counts for each circuits to accelerate MDC store creation
+    if isinstance(mdl, _models.model.OpModel):
+        if precomp_layout_circuit_cache is not None: #then grab the split circuits from there.
+            expanded_circuit_outcome_list = mdl.bulk_expand_instruments_and_separate_povm(unique_circuits, 
+                                                                                          completed_circuits= precomp_layout_circuit_cache['completed_circuits'].values())
+        else:
+            expanded_circuit_outcome_list = mdl.bulk_expand_instruments_and_separate_povm(unique_circuits)        
+        outcome_count_by_circuit_cache = {ckt: len(outcome_tup) for ckt,outcome_tup in zip(unique_circuits, expanded_circuit_outcome_list)}
+    else:
+        outcome_count_by_circuit_cache = {ckt: mdl.compute_num_outcomes(ckt) for ckt in unique_circuits}
+
     with printer.progress_logging(1):
         for i in range(starting_index, len(circuit_lists)):
             circuitsToEstimate = circuit_lists[i]
@@ -908,7 +918,8 @@ def _max_array_types(artypes_list):  # get the maximum number of each array type
             mdl.basis = start_model.basis  # set basis in case of CPTP constraints (needed?)
             initial_mdc_store = _objfns.ModelDatasetCircuitsStore(mdl, dataset, circuitsToEstimate, resource_alloc,
                                                                   array_types=array_types, verbosity=printer - 1, 
-                                                                  precomp_layout = precomp_layouts[i])
+                                                                  precomp_layout = precomp_layouts[i],
+                                                                  outcome_count_by_circuit=outcome_count_by_circuit_cache)
             mdc_store = initial_mdc_store
 
             for j, obj_fn_builder in enumerate(iteration_objfn_builders):

pygsti/circuits/__init__.py

@@ -10,7 +10,7 @@
 # http://www.apache.org/licenses/LICENSE-2.0 or in the LICENSE file in the root pyGSTi directory.
 #***************************************************************************************************
 
-from .circuit import Circuit
+from .circuit import Circuit, SeparatePOVMCircuit
 from .circuitlist import CircuitList
 from .circuitstructure import CircuitPlaquette, FiducialPairPlaquette, \
     GermFiducialPairPlaquette, PlaquetteGridCircuitStructure

pygsti/circuits/circuit.py

@@ -1095,13 +1095,12 @@ def _proc_lines_arg(self, lines):
     def _proc_key_arg(self, key):
         """ Pre-process the key argument used by many methods """
         if isinstance(key, tuple):
-            if len(key) != 2: return IndexError("Index must be of the form <layerIndex>,<lineIndex>")
-            layers = key[0]
-            lines = key[1]
+            if len(key) != 2: 
+                return IndexError("Index must be of the form <layerIndex>,<lineIndex>")
+            else:
+                return key[0], key[1]
         else:
-            layers = key
-            lines = None
-        return layers, lines
+            return key, None
 
     def _layer_components(self, ilayer):
         """ Get the components of the `ilayer`-th layer as a list/tuple. """
@@ -1191,22 +1190,41 @@ def extract_labels(self, layers=None, lines=None, strict=True):
             `layers` is a single integer and as a `Circuit` otherwise.
             Note: if you want a `Circuit` when only selecting one layer,
             set `layers` to a slice or tuple containing just a single index.
+            Note that the returned circuit doesn't retain any original
+            metadata, such as the compilable layer indices or occurence id.
         """
         nonint_layers = not isinstance(layers, int)
 
         #Shortcut for common case when lines == None and when we're only taking a layer slice/index
-        if lines is None:
-            assert(layers is not None)
-            if nonint_layers is False: return self.layertup[layers]
-            if isinstance(layers, slice) and strict is True:  # if strict=False, then need to recompute line labels
-                return Circuit._fastinit(self._labels[layers], self._line_labels, not self._static)
+        if lines is None and layers is not None:
+            if self._static:
+                if not nonint_layers:
+                    return self._labels[layers]
+                if isinstance(layers, slice) and strict is True:  # if strict=False, then need to recompute line labels
+                    #can speed this up a measurably by manually computing the new hashable tuple value and hash
+                    if not self._line_labels in (('*',), ()):
+                        new_hashable_tup = self._labels[layers] + ('@',) + self._line_labels
+                    else:
+                        new_hashable_tup = self._labels[layers]
+                    ret = Circuit.__new__(Circuit)
+                    return ret._copy_init(self._labels[layers], self._line_labels, not self._static, hashable_tup= new_hashable_tup, precomp_hash=hash(new_hashable_tup))
+            else:
+                if not nonint_layers:
+                    return self.layertup[layers]
+                if isinstance(layers, slice) and strict is True:  # if strict=False, then need to recompute line labels
+                    return Circuit._fastinit(self._labels[layers], self._line_labels, not self._static)
+        #otherwise assert both are not None:
+
 
         layers = self._proc_layers_arg(layers)
         lines = self._proc_lines_arg(lines)
         if len(layers) == 0 or len(lines) == 0:
-            return Circuit._fastinit(() if self._static else [],
-                                     tuple(lines) if self._static else lines,
-                                     not self._static) if nonint_layers else None  # zero-area region
+            if self._static:
+                return Circuit._fastinit((), tuple(lines), False)  # zero-area region
+            else:
+                return Circuit._fastinit(() if self._static else [],
+                                        tuple(lines) if self._static else lines,
+                                        not self._static)  # zero-area region
 
         ret = []
         if self._static:
@@ -3447,7 +3465,7 @@ def num_gates(self):
         """
         if self._static:
             def cnt(lbl):  # obj a Label, perhaps compound
-                if lbl.is_simple():  # a simple label
+                if lbl.IS_SIMPLE:  # a simple label
                     return 1 if (lbl.sslbls is not None) else 0
                 else:
                     return sum([cnt(sublbl) for sublbl in lbl.components])
@@ -4428,107 +4446,6 @@ def done_editing(self):
         self._hashable_tup = self.tup
         self._hash = hash(self._hashable_tup)
 
-    def expand_instruments_and_separate_povm(self, model, observed_outcomes=None):
-        """
-        Creates a dictionary of :class:`SeparatePOVMCircuit` objects from expanding the instruments of this circuit.
-
-        Each key of the returned dictionary replaces the instruments in this circuit with a selection
-        of their members.  (The size of the resulting dictionary is the product of the sizes of
-        each instrument appearing in this circuit when `observed_outcomes is None`).  Keys are stored
-        as :class:`SeparatePOVMCircuit` objects so it's easy to keep track of which POVM outcomes (effects)
-        correspond to observed data.  This function is, for the most part, used internally to process
-        a circuit before computing its outcome probabilities.
-
-        Parameters
-        ----------
-        model : Model
-            The model used to provide necessary details regarding the expansion, including:
-
-            - default SPAM layers
-            - definitions of instrument-containing layers
-            - expansions of individual instruments and POVMs
-
-        Returns
-        -------
-        OrderedDict
-            A dict whose keys are :class:`SeparatePOVMCircuit` objects and whose
-            values are tuples of the outcome labels corresponding to this circuit,
-            one per POVM effect held in the key.
-        """
-        complete_circuit = model.complete_circuit(self)
-        expanded_circuit_outcomes = _collections.OrderedDict()
-        povm_lbl = complete_circuit[-1]  # "complete" circuits always end with a POVM label
-        circuit_without_povm = complete_circuit[0:len(complete_circuit) - 1]
-
-        def create_tree(lst):
-            subs = _collections.OrderedDict()
-            for el in lst:
-                if len(el) > 0:
-                    if el[0] not in subs: subs[el[0]] = []
-                    subs[el[0]].append(el[1:])
-            return _collections.OrderedDict([(k, create_tree(sub_lst)) for k, sub_lst in subs.items()])
-
-        def add_expanded_circuit_outcomes(circuit, running_outcomes, ootree, start):
-            """
-            """
-            cir = circuit if start == 0 else circuit[start:]  # for performance, avoid uneeded slicing
-            for k, layer_label in enumerate(cir, start=start):
-                components = layer_label.components
-                #instrument_inds = _np.nonzero([model._is_primitive_instrument_layer_lbl(component)
-                #                               for component in components])[0]  # SLOWER than statement below
-                instrument_inds = _np.array([i for i, component in enumerate(components)
-                                             if model._is_primitive_instrument_layer_lbl(component)])
-                if instrument_inds.size > 0:
-                    # This layer contains at least one instrument => recurse with instrument(s) replaced with
-                    #  all combinations of their members.
-                    component_lookup = {i: comp for i, comp in enumerate(components)}
-                    instrument_members = [model._member_labels_for_instrument(components[i])
-                                          for i in instrument_inds]  # also components of outcome labels
-                    for selected_instrmt_members in _itertools.product(*instrument_members):
-                        expanded_layer_lbl = component_lookup.copy()
-                        expanded_layer_lbl.update({i: components[i] + "_" + sel
-                                                   for i, sel in zip(instrument_inds, selected_instrmt_members)})
-                        expanded_layer_lbl = _Label([expanded_layer_lbl[i] for i in range(len(components))])
-
-                        if ootree is not None:
-                            new_ootree = ootree
-                            for sel in selected_instrmt_members:
-                                new_ootree = new_ootree.get(sel, {})
-                            if len(new_ootree) == 0: continue  # no observed outcomes along this outcome-tree path
-                        else:
-                            new_ootree = None
-
-                        add_expanded_circuit_outcomes(circuit[0:k] + Circuit((expanded_layer_lbl,)) + circuit[k + 1:],
-                                                      running_outcomes + selected_instrmt_members, new_ootree, k + 1)
-                    break
-
-            else:  # no more instruments to process: `cir` contains no instruments => add an expanded circuit
-                assert(circuit not in expanded_circuit_outcomes)  # shouldn't be possible to generate duplicates...
-                elabels = model._effect_labels_for_povm(povm_lbl) if (observed_outcomes is None) \
-                    else tuple(ootree.keys())
-                outcomes = tuple((running_outcomes + (elabel,) for elabel in elabels))
-                expanded_circuit_outcomes[SeparatePOVMCircuit(circuit, povm_lbl, elabels)] = outcomes
-
-        ootree = create_tree(observed_outcomes) if observed_outcomes is not None else None  # tree of observed outcomes
-        # e.g. [('0','00'), ('0','01'), ('1','10')] ==> {'0': {'00': {}, '01': {}}, '1': {'10': {}}}
-
-        if model._has_instruments():
-            add_expanded_circuit_outcomes(circuit_without_povm, (), ootree, start=0)
-        else:
-            # It may be helpful to cache the set of elabels for a POVM (maybe within the model?) because
-            # currently the call to _effect_labels_for_povm may be a bottleneck.  It's needed, even when we have
-            # observed outcomes, because there may be some observed outcomes that aren't modeled (e.g. leakage states)
-            if observed_outcomes is None:
-                elabels = model._effect_labels_for_povm(povm_lbl)
-            else:
-                possible_lbls = set(model._effect_labels_for_povm(povm_lbl))
-                elabels = tuple([oo for oo in ootree.keys() if oo in possible_lbls])
-            outcomes = tuple(((elabel,) for elabel in elabels))
-            expanded_circuit_outcomes[SeparatePOVMCircuit(circuit_without_povm, povm_lbl, elabels)] = outcomes
-
-        return expanded_circuit_outcomes
-
-
 class CompressedCircuit(object):
     """
     A "compressed" Circuit that requires less disk space.