Feature globally germ aware fpr

jupyter_notebooks/Tutorials/algorithms/advanced/GST-FiducialPairReduction.ipynb

@@ -14,7 +14,10 @@
     "- **Global fiducial pair reduction (GFPR)** removes the same intelligently-selected set of fiducial pairs for all germ-powers.  This is a conceptually simple method of reducing the operation sequences, but it is the most computationally intensive since it repeatedly evaluates the number of amplified parameters for en *entire germ set*.  In practice, while it can give very large sequence reductions, its long run can make it prohibitive, and the \"per-germ\" reduction discussed next is used instead. \n",
     "<span style=\"color:red\">Note: this form of FPR is deprecated on the latest versions of pygsti's develop branch. We now recommend using per-germ FPR instead. Also note that the current implementation of per-germ FPR will in most cases return smaller experiment designs than the legacy global FPR does.</span>\n",
     "\n",
-    "- **Per-germ fiducial pair reduction (PFPR)** removes the same intelligently-selected set of fiducial pairs for all powers of a given germ, but different sets are removed for different germs.  Since different germs amplify different directions in model space, it makes intuitive sense to specify different fiducial pair sets for different germs.  Because this method only considers one germ at a time, it is less computationally intensive than GFPR, and thus more practical.  Note, however, that PFPR usually results in less of a reduction of the operation sequences, since it does not (currently) take advantage overlaps in the amplified directions of different germs (i.e. if $g_1$ and $g_3$ both amplify two of the same directions, then GST doesn't need to know about these from both germs).\n",
+    "- **Per-germ fiducial pair reduction (PFPR)** removes the same intelligently-selected set of fiducial pairs for all powers of a given germ, but different sets are removed for different germs.  Since different germs amplify different directions in model space, it makes intuitive sense to specify different fiducial pair sets for different germs.  Because this method only considers one germ at a time, it is less computationally intensive than GFPR, and thus more practical.\n",
+    "\n",
+    "- **Per-germ global fiducial pair reduction (PGGFPR)** removes the same intelligently-selected set of fiducial pairs for all powers of a given germ, but different sets are removed for different germs while also taking into account the amplificational properties of a germ set as a whole. This is a two-step process in which we first identify redundancy within a germ set itself due to overlapping amplified directions in parameter space and identifies a subset of amplified parameters for each germ such that collectively we have sensitivity to every direction. In the second stage we select a subset of fiducial pairs for each germ only requiring sensitivity to the subset of amplified parameters of that germ identified in the first stage. This is currently our most effective form of fiducial pair reduction in terms of potential experimental savings, capable with the right settings of achieving experimental designs approaching information theoretic lower bounds in size, but with fewer fiducial pairs comes the potential for detecting non-Markovian effects and potentially less robustness to those effects (the extent to which this is true, or if it is true at all is an active area of research), so caveat emptor. \n",
+    "\n",
     "- **Random per-germ power fiducial pair reduction (RFPR)** randomly chooses a different set of fiducial pairs to remove for each germ-power.  It is extremly fast to perform, as pairs are just randomly selected for removal, and in practice works well (i.e. does not impair Heisenberg-scaling) up until some critical fraction of the pairs are removed.  This reflects the fact that the direction detected by a fiducial pairs usually has some non-negligible overlap with each of the directions amplified by a germ, and it is the exceptional case that an amplified direction escapes undetected.  As such, the \"critical fraction\" which can usually be safely removed equals the ratio of amplified-parameters to germ-process-matrix-elements (typically $\\approx 1/d^2$ where $d$ is the Hilbert space dimension, so $1/4 = 25\\%$ for 1 qubit and $1/16 = 6.25\\%$ for 2 qubits).  RFPR can be combined with GFPR or PFPR so that some number of randomly chosen pairs can be added on top of the \"intelligently-chosen\" pairs of GFPR or PFPR.  In this way, one can vary the amount of sequence reduction (in order to trade off speed vs. robustness to non-Markovian noise) without inadvertently selecting too few or an especially bad set of random fiducial pairs.\n",
     "\n",
     "## Preliminaries\n",
@@ -24,21 +27,30 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 12,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Gate operation labels =  [Label(('Gxpi2', 0)), Label(('Gypi2', 0))]\n"
+     ]
+    }
+   ],
    "source": [
-    "#Import pyGSTi and the \"stardard 1-qubit quantities for a model with X(pi/2), Y(pi/2), and idle gates\"\n",
+    "#Import pyGSTi and the \"stardard 1-qubit quantities for a model with X(pi/2), Y(pi/2)\"\n",
     "import pygsti\n",
     "import pygsti.circuits as pc\n",
-    "from pygsti.modelpacks import smq1Q_XYI\n",
+    "from pygsti.modelpacks import smq1Q_XY\n",
+    "import numpy as np\n",
     "\n",
     "#Collect a target model, germ and fiducial strings, and set \n",
     "# a list of maximum lengths.\n",
-    "target_model = smq1Q_XYI.target_model()\n",
-    "prep_fiducials = smq1Q_XYI.prep_fiducials()\n",
-    "meas_fiducials = smq1Q_XYI.meas_fiducials()\n",
-    "germs = smq1Q_XYI.germs()\n",
+    "target_model = smq1Q_XY.target_model()\n",
+    "prep_fiducials = smq1Q_XY.prep_fiducials()\n",
+    "meas_fiducials = smq1Q_XY.meas_fiducials()\n",
+    "germs = smq1Q_XY.germs()\n",
     "maxLengths = [1,2,4,8,16,32]\n",
     "\n",
     "opLabels = list(target_model.operations.keys())\n",
@@ -58,9 +70,25 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 2,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "** Without any reduction ** \n",
+      "L=1: 92 operation sequences\n",
+      "L=2: 168 operation sequences\n",
+      "L=4: 285 operation sequences\n",
+      "L=8: 448 operation sequences\n",
+      "L=16: 616 operation sequences\n",
+      "L=32: 784 operation sequences\n",
+      "\n",
+      "784 experiments to run GST.\n"
+     ]
+    }
+   ],
    "source": [
     "#Make list-of-lists of GST operation sequences\n",
     "fullStructs = pc.create_lsgst_circuit_lists(\n",
@@ -155,6 +183,117 @@
     "print(\"\\n%d experiments to run GST.\" % len(pfprExperiments))"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Per-germ Fiducial Pair Reduction (PFPR) with Greedy Search Heuristics\n",
+    "\n",
+    "In addition to the implementation of per-germ fiducial pair reduction above, which supports either a brute force sequential or random search heuristic, there is also an implementation using a greedy search heuristic combined with fast low-rank update-based techniques for significantly faster execution, particularly when generating experiment designs for two-or-more qubits. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "------  Per Germ (L=1) Fiducial Pair Reduction --------\n",
+      "Progress: [##################################################] 100.0% -- Circuit(Gxpi2:0Gxpi2:0Gypi2:0@(0)) germ (5 params)\n",
+      "\n",
+      "Per-germ FPR to keep the pairs:\n",
+      "Qubit 0 ---|Gxpi2|---\n",
+      ": [(0, 1), (3, 1), (3, 3), (5, 5)]\n",
+      "Qubit 0 ---|Gypi2|---\n",
+      ": [(0, 3), (2, 3), (5, 2), (4, 4)]\n",
+      "Qubit 0 ---|Gxpi2|-|Gypi2|---\n",
+      ": [(3, 4), (5, 2), (5, 5), (5, 4)]\n",
+      "Qubit 0 ---|Gxpi2|-|Gxpi2|-|Gypi2|---\n",
+      ": [(0, 2), (1, 2), (1, 4), (3, 0), (4, 4), (0, 4)]\n",
+      "\n",
+      "Per-germ FPR reduction (greedy heuristic)\n",
+      "L=1: 56 operation sequences\n",
+      "L=2: 61 operation sequences\n",
+      "L=4: 71 operation sequences\n",
+      "L=8: 89 operation sequences\n",
+      "L=16: 107 operation sequences\n",
+      "L=32: 125 operation sequences\n",
+      "\n",
+      "125 experiments to run GST.\n"
+     ]
+    }
+   ],
+   "source": [
+    "fid_pairsDict = pygsti.alg.find_sufficient_fiducial_pairs_per_germ_greedy(target_model, prep_fiducials, meas_fiducials,\n",
+    "                                                germs, verbosity=1)\n",
+    "print(\"\\nPer-germ FPR to keep the pairs:\")\n",
+    "for germ,pairsToKeep in fid_pairsDict.items():\n",
+    "    print(\"%s: %s\" % (str(germ),pairsToKeep))\n",
+    "\n",
+    "pfprStructs_greedy = pc.create_lsgst_circuit_lists(\n",
+    "    opLabels, prep_fiducials, meas_fiducials, germs, maxLengths,\n",
+    "    fid_pairs=fid_pairsDict) #note: fid_pairs arg can be a dict too!\n",
+    "\n",
+    "print(\"\\nPer-germ FPR reduction (greedy heuristic)\")\n",
+    "for L,strct in zip(maxLengths,pfprStructs_greedy):\n",
+    "    print(\"L=%d: %d operation sequences\" % (L,len(strct)))\n",
+    "\n",
+    "pfprExperiments_greedy = pc.create_lsgst_circuits(\n",
+    "    opLabels, prep_fiducials, meas_fiducials, germs, maxLengths,\n",
+    "    fid_pairs=fid_pairsDict)\n",
+    "print(\"\\n%d experiments to run GST.\" % len(pfprExperiments_greedy))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Per-germ Global Fiducial Pair Reduction (PFPR)\n",
+    "\n",
+    "As mentioned above, the per-germ global FPR scheme is a two step process. First we identify a reduced set of amplified parameters for each germ to require sensitivity to, and then next we identify reduced sets of fiducials with sensitivity to those particular parameters."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#Note that we are setting the assume_real flag to True below as we know that we am working in the Pauli basis and as such the \n",
+    "#process matrices for the germs will be real-valued allowing for memory savings and somewhat faster performance. \n",
+    "#If you're working with a non-hermitian basis or aren't sure keep this set to it's default value of False.\n",
+    "#Likewise, float_type specifies the numpy data type to use, and is primarily useful either in conjunction with\n",
+    "#assume_real, or when needing to fine-tune the memory requirements of the algorithm (running this algorithm for\n",
+    "#more than 2-qubits can be very memory intensive). When running this function for more than two-qubits, consider\n",
+    "#setting the mode kwarg to 'RRQR', which is typically significantly faster for larger qubit counts, but is slightly\n",
+    "#less performant in terms of the cost function of the returned solutions.\n",
+    "germ_set_spanning_vectors, _ = pygsti.alg.germ_set_spanning_vectors(target_model, germs, assume_real=True, float_type= np.double)\n",
+    "\n",
+    "#Next use this set of vectors to find a sufficient reduced set of fiducial pairs.\n",
+    "#Alternatively this function can also take as input a list of germs\n",
+    "fid_pairsDict = pygsti.alg.find_sufficient_fiducial_pairs_per_germ_global(target_model, prep_fiducials, meas_fiducials,\n",
+    "                                                germ_vector_spanning_set=germ_set_spanning_vectors, verbosity=1)\n",
+    "print(\"\\nPer-germ Global FPR to keep the pairs:\")\n",
+    "for germ,pairsToKeep in fid_pairsDict.items():\n",
+    "    print(\"%s: %s\" % (str(germ),pairsToKeep))\n",
+    "\n",
+    "pggfprStructs = pc.create_lsgst_circuit_lists(\n",
+    "    opLabels, prep_fiducials, meas_fiducials, germs, maxLengths,\n",
+    "    fid_pairs=fid_pairsDict) #note: fid_pairs arg can be a dict too!\n",
+    "\n",
+    "print(\"\\nPer-germ Global FPR reduction\")\n",
+    "for L,strct in zip(maxLengths,pggfprStructs):\n",
+    "    print(\"L=%d: %d operation sequences\" % (L,len(strct)))\n",
+    "\n",
+    "pggfprExperiments = pc.create_lsgst_circuits(\n",
+    "    opLabels, prep_fiducials, meas_fiducials, germs, maxLengths,\n",
+    "    fid_pairs=fid_pairsDict)\n",
+    "print(\"\\n%d experiments to run GST.\" % len(pggfprExperiments))"
+   ]
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -218,6 +357,12 @@
     "print(\"\\n------ GST with PFPR sequences ------\")\n",
     "pfpr_results = runGST(pfprStructs, pfprExperiments)\n",
     "\n",
+    "print(\"\\n------ GST with PFPR sequences (greedy heuristic) ------\")\n",
+    "pfpr_results_greedy = runGST(pfprStructs_greedy, pfprExperiments_greedy)\n",
+    "\n",
+    "print(\"\\n------ GST with PGGFPR sequences ------\")\n",
+    "pggfpr_results = runGST(pggfprStructs, pggfprExperiments)\n",
+    "\n",
     "print(\"\\n------ GST with RFPR sequences ------\")\n",
     "rfpr_results = runGST(rfprStructs, rfprExperiments)"
    ]
@@ -241,6 +386,10 @@
     "                                        ).write_html(\"tutorial_files/example_gfpr_report\")\n",
     "pygsti.report.construct_standard_report(pfpr_results, title=\"Per-germ FPR Report Example\"\n",
     "                                        ).write_html(\"tutorial_files/example_pfpr_report\")\n",
+    "pygsti.report.construct_standard_report(pfpr_results_greedy, title=\"Per-germ FPR (Greedy Heuristic) Report Example\"\n",
+    "                                        ).write_html(\"tutorial_files/example_pfpr_greedy_report\")\n",
+    "pygsti.report.construct_standard_report(pggfpr_results, title=\"Per-germ Global FPR Report Example\"\n",
+    "                                        ).write_html(\"tutorial_files/example_pggfpr_report\")\n",
     "pygsti.report.construct_standard_report(rfpr_results, title=\"Random FPR Report Example\"\n",
     "                                        ).write_html(\"tutorial_files/example_rfpr_report\")"
    ]
@@ -251,18 +400,28 @@
    "source": [
     "If all has gone well, the [Standard GST](tutorial_files/example_stdstrs_report/main.html),\n",
     "[GFPR](tutorial_files/example_gfpr_report/main.html),\n",
-    "[PFPR](tutorial_files/example_pfpr_report/main.html), and\n",
+    "[PFPR](tutorial_files/example_pfpr_report/main.html),\n",
+    "[PFPR (Greedy)](tutorial_files/example_pfpr_greedy_report/main.html)\n",
+    "[PGGFPR](tutorial_files/example_pggfpr_report/main.html)\n",
+    "and\n",
     "[RFPR](tutorial_files/example_rfpr_report/main.html),\n",
     "reports may now be viewed.\n",
     "The only notable difference in the output are \"gaps\" in the color box plots which plot quantities such as the log-likelihood across all operation sequences, organized by germ and fiducials.  "
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3",
+   "display_name": "New_FPR",
    "language": "python",
-   "name": "python3"
+   "name": "new_fpr"
   },
   "language_info": {
    "codemirror_mode": {
@@ -274,9 +433,9 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.5"
+   "version": "3.9.13"
   }
  },
  "nbformat": 4,
- "nbformat_minor": 1
+ "nbformat_minor": 4
 }

pygsti/algorithms/core.py

@@ -856,7 +856,23 @@ def _max_array_types(artypes_list):  # get the maximum number of each array type
         ret = ()
         for artype, cnt in max_cnts.items(): ret += (artype,) * cnt
         return ret
-
+
+    #These lines were previously in the loop below, but we should be able to move it out from there so we can use it
+    #in precomputing layouts:
+    method_names = optimizer.called_objective_methods
+    array_types = optimizer.array_types + \
+                _max_array_types([builder.compute_array_types(method_names, mdl.sim)
+                                  for builder in iteration_objfn_builders + final_objfn_builders])
+
+    #precompute the COPA layouts. During the layout construction there are memory availability checks,
+    #So by doing it this way we should be able to reduce the number of instances of running out of memory before the end.
+    #The ModelDatasetCircuitsStore
+    printer.log('Precomputing CircuitOutcomeProbabilityArray layouts for each iteration.', 2)
+    precomp_layouts = []
+    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))
+
     with printer.progress_logging(1):
         for i in range(starting_index, len(circuit_lists)):
             circuitsToEstimate = circuit_lists[i]
@@ -871,12 +887,9 @@ def _max_array_types(artypes_list):  # get the maximum number of each array type
             if circuitsToEstimate is None or len(circuitsToEstimate) == 0: continue
 
             mdl.basis = start_model.basis  # set basis in case of CPTP constraints (needed?)
-            method_names = optimizer.called_objective_methods
-            array_types = optimizer.array_types + \
-                _max_array_types([builder.compute_array_types(method_names, mdl.sim)
-                                  for builder in iteration_objfn_builders + final_objfn_builders])
             initial_mdc_store = _objfns.ModelDatasetCircuitsStore(mdl, dataset, circuitsToEstimate, resource_alloc,
-                                                                  array_types=array_types, verbosity=printer - 1)
+                                                                  array_types=array_types, verbosity=printer - 1, 
+                                                                  precomp_layout = precomp_layouts[i])
             mdc_store = initial_mdc_store
 
             for j, obj_fn_builder in enumerate(iteration_objfn_builders):