The pygsti
package provides multiple ways to use its core Gate Set Tomography (GST) algorithms. This tutorial will show you how to work with pyGSTi's GST protocol objects to perform GST in different ways with a minimial amount of effort. In order to run the GST protocol there are 3 essential ingredients: 1) an "experiment design" specifying the structure of the GST circuits and how the data should be collected, 2) the outcome counts for the circuits specified by the experiment design, and 3) a desired, or "target", Model
. The GST overview tutorial, gave an end-to-end example of how to construct a GST experiment design, run GST, and generate a report. This tutorial focuses on the first and second steps in more detail; related information about circuit construction and report generation can be found in the GST circuits tutorial and report generation tutorial.
There are two different Protocol
objects within pyGSTi for running GST:
GateSetTomography
- runs a single model optimization based on a given initial model that can have any parameterization you like. This protocol can be run on any GateSetTomographyDesign
experiment design, which only needs a target model (to describe what gates occur in the circuits) and a list of circuit lists to specify the circuits used for each iteration of the model optimization.
StandardGST
- runs multiple model optimizations based on an ExplicitOpModel
target model by parameterizing this model in different ways. The target model is expected to be a part of the experiment design, and only StandardGSTDesign
-type experiment designs are allowed since the usual germs-and-fiducials structure of the GST circuits is expected.
Overall, the GateSetTomography
protocol is more flexible than the StandardGST
protocol, but requires a little more work to get going because its inputs are more complicated. Both protocols return a ModelEstimateResults
object when they are run.
from __future__ import print_function
import pygsti
In the DataSet tutorial we simulate the circuits required by a GST experiment design and save the results. In this tutorial, we'll be analyzing that data. This illustrates a typical workflow where at some earlier time you setup an experiment (a "GST experiment in this case) and save the experiment design to disk and at some later time (after the data has been collected) you want to analyze it. Now is that later time, and we start by reading the the data we've collected.
data = pygsti.io.load_data_from_dir("../tutorial_files/Example_GST_Data")
GateSetTomography
¶This protocol performs a single model optimization, and so computes a single GST estimate given a DataSet
, a target Model
, and other parameters. (The returned ModelEstimateResults
object may sometimes contain multiple related estimates in certain cases, but in these cases all the estimates are closely related.) The experiment design provides all of the information about the GST circuits, in this case a standard (prep_fiducial + germ^power + meas_fiducial) set, so the only thing needed by the protocol is an initial Model
to optimize. Thus, the GateSetTomography
protocol is essentially just a model optimizer that you give an initial point. Importantly, this initial point (a Model
) also specifies the parameterization, i.e. the space of parameters that are optimized over.
Minimally, when using GateSetTomography
you should set the parameterization of the initial model. This can be viewed as setting the constraints on the optimization. For instance, when the gates in the model are parameterized as trace-preserving (TP) maps, the optimization will be constrained to trying gate sets with TP gates (because every set of parameters corresponds to a set of TP gates). In the cell below, we constrain the optimization to TP gate sets by using .target_model("TP")
, which returns a version of the target model where all the gates are TP-parameterized, the state preparation has trace = 1, and the POVM effects always add to the identity. This could also be done by calling set_all_parameterizations("TP")
on the fully-parameterized target model returned by .target_model()
. See the tutorial on explicit models for more information on setting a model's parameterization.
from pygsti.modelpacks import smq1Q_XYI
target_model_TP = smq1Q_XYI.target_model("TP")
proto = pygsti.protocols.GateSetTomography(target_model_TP)
results_TP = proto.run(data)
A summary of what's inside a Results object is obtained by printing it (for more examples of how to use a Results object, see the Results tutorial).
print(results_TP)
The gaugeopt_suite
argument specifies a set of gauge optimizations to be performed on the final GST estimate. It is a dictionary whose keys are gauge-optimization names (these can be whatever you want) and whose values are dictionaries of arguments ultimately to be passed to the gaugeopt_to_target
function (which provides full documentation). (For example, by specifying itemWeights
we can set the ratio of the state preparation and measurement (SPAM) weighting to the gate weighting when performing a gauge optimization.) In lieu of a dictionary of gaugeopt_to_target
arguments, the elements of gaugeopt_suite
may also be strings which name a built-in set of gauge optimizations (e.g. "stdgaugeopt"
is the name of the standard gauge optimization).
If gaugeopt_suite
is set to a string, this is the same as passing a dictionary with a single key-value pair where both key and value are equal to the string. Thus, the default "stdgaugeopt"
is equivalent to specifying the dictionary {"stdgaugeopt": "stdgagueopt"}
.
The example below performs a customized gauge-optimization where the gate parameters are weighted 1000 times more relative to the SPAM parameters. Mathematically this corresponds to a multiplicative factor of 0.001 preceding the sum-of-squared-difference terms corresponding to SPAM elements in the model. Typically it is good to weight the gates parameters more heavily since GST amplifies gate parameter errors via long operation sequences but cannot amplify SPAM parameter errors. For more details on the arguments of gaugeopt_to_target
, see the previous tutorial on low-level algorithms. For more infomation, see the gauge optimization tutorial.
The cell below also illustrates how you can create a TP target model by calling set_all_parameterizations
explicitly instead of using the equivalent and more condensed .target_model("TP")
.
target_model_TP2 = smq1Q_XYI.target_model() # a "fully parameterized" (unconstrained) model
target_model_TP2.set_all_parameterizations("TP") # change parameterization to TP gates
proto = pygsti.protocols.GateSetTomography(
target_model_TP2, name="GSTwithMyGO",
gaugeopt_suite={'my_gauge_opt': {'itemWeights': {'gates': 1.0, 'spam': 0.001}}}
)
results_TP2 = proto.run(data)
print(results_TP2.estimates['default'].goparameters.keys()) # names of all the gauge opts that were done
custom_gauge_opt_model = results_TP2.estimates['default'].models['my_gauge_opt']
So far we've giving the GateSetTomography.run
method an "standard" experiment design containing circuits chosen to amplify all of a standard TP (or CPTP) model's parameters (see the StandardGSTExpermentDesign
used in the DataSet tutorial). A GateSetTomography
protocol can be run on more general experiment designs, namely those that specify the circuits to use as either a list of lists of Circuit
objects or a list of or single CircuitStructure
object(s). A CircuitStructure
is preferable as it allows the structured plotting of the sequences in report figures. In this example, we'll just generate a standard set of circuit structures, but with some of the sequences randomly dropped (see the tutorial on GST circuit reduction.
# Create the same sequences but drop 50% of them randomly for each repeated-germ block.
# and only go out to a max-length of 8
orig_design = data.edesign # the original StandardGSTDesign
custom_maxlengths = [1, 2, 4, 8] # a subset of orig_design.maxlengths
circuit_structs = pygsti.construction.make_lsgst_structs(
orig_design.target_model, orig_design.prep_fiducials, orig_design.meas_fiducials,
orig_design.germs, custom_maxlengths, keepFraction=0.5, keepSeed=2020)
reduced_exp_design = pygsti.protocols.StructuredGSTDesign(orig_design.target_model, circuit_structs)
reduced_data = pygsti.protocols.ProtocolData(reduced_exp_design, data.dataset)
proto = pygsti.protocols.GateSetTomography(target_model_TP2, name="GSTwithReducedData")
results_reduced = proto.run(reduced_data)
StandardGST
¶The protocol embodies a standard set of GST protocols to be run on a set of data. It essentially runs multiple GateSetTomography
protocols on the given data which use different parameterizations of an ExplicitOpModel
(the StandardGST
protocol doesn't work with other types of Model
objects, e.g. implicit models, which don't implement set_all_parameterizations
). The modes
argument is a comma-separated list of the parameterization types that should be run (e.g. "TP,CPTP"
will compute a Trace-Preserving estimate and a Completely-Positive & Trace-Preserving estimate). The currently available modes are:
Gauge optimization(s) are controlled by the gaugeopt_suite
and gaugeopt_target
arguments, jsut as in GateSetTomography
. The gaugeopt_target
argument may be set to a Model
that is used as the target for gauge optimization, overriding the (typically ideal) target gates given by within the experiment design.
results_stdprac = pygsti.protocols.StandardGST().run(data)
print("Estimates: ", ", ".join(results_stdprac.estimates.keys()))
print("TP Estimate's gauge optimized models: ", ", ".join(results_stdprac.estimates["TP"].goparameters.keys()))
Next, we'll perform the same analysis but with a non-default standard suite of gauge optimizations - this one toggles the SPAM penalty in addition to varying the spam weight (the default suite just varies the spam weight without any SPAM penalty). See the gauge optimization tutorial for more details on gauge optmization "suites".
proto = pygsti.protocols.StandardGST(gaugeopt_suite="varySpam", name="StdGST_varySpam")
results_stdprac_nondefaultgo = proto.run(data)
print("Estimates: ", ", ".join(results_stdprac_nondefaultgo.estimates.keys()))
print("TP Estimate's gauge optimized models: ", ", ".join(results_stdprac_nondefaultgo.estimates["TP"].goparameters.keys()))
Finally, we'll demonstrate how to specify a fully custom set of gauge optimization parameters and how to use a separately-specified target model for gauge optimization. You can get a more intuitive gauge-optimized Model
when by placing as much expected noise as possible into the gauge-optimization target, as this essentially tells the algorithm "this is what I think the estimated model should look like". If you just use the perfect or ideal model for this (the default), then the gauge optimizer may make tradeoffs which don't reflect the expected physics (remember, all gauge-equivalent models product the same observables!). For example, it may spread error across all your gate operations when you expect just the 2-qubit operations are noisy.
my_goparams = { 'itemWeights': {'gates': 1.0, 'spam': 0.001} }
my_gaugeOptTarget = smq1Q_XYI.target_model().depolarize(
op_noise=0.005, spam_noise=0.01) # a guess at what estimate should be
proto = pygsti.protocols.StandardGST(gaugeopt_suite={ 'myGO': my_goparams },
gaugeopt_target=my_gaugeOptTarget,
name="StdGST_myGO")
results_stdprac_nondefaultgo = proto.run(data)
print("Estimates: ", ", ".join(results_stdprac_nondefaultgo.estimates.keys()))
print("TP Estimate's gauge optimized models: ", ", ".join(results_stdprac_nondefaultgo.estimates["TP"].goparameters.keys()))
To finish up, we'll write the results for processing in other tutorials. We do this by calling .write
on the results objects, optionally specifying the root diretory under which the results should be written. This is the same root directory that the experiment design and data are written to, as subdirectories beneath this directory separate these quantities.
Two remarks are in order:
When results are from running a protocol on data that was loaded with the load_data_from_dir
method (see the beginning of this notebook), then knowledge of this directory is remembered and you don't need to give a directory to write
(this is the case for all except results_reduced
, which created a new experiment design containing less experiments).
Notice how the name=
arguments given to protocols above are used as sub-directory names, e.g. under the "tutorial_files/Example_GST_Data/results" parent directory.
results_TP.write() # uses "../tutorial_files/Example_GST_Data" (where data was loaded from)
results_TP2.write() # ditto
results_stdprac.write() # ditto
results_reduced.write("../tutorial_files/Example_Reduced_GST_Data") # choose a different dir
While it is possible to pickle a results object, this method of serialization is not recommended for long-term storage since pickles are relatively fragile to changes in pyGSTi or other python libraries.
#Not recommended:
# import pickle
# pickle.dump(results_TP, open('../tutorial_files/exampleResults_TP.pkl',"wb"))
# pickle.dump(results_TP2, open('../tutorial_files/exampleResults_TP2.pkl',"wb"))
# pickle.dump(results_reduced, open('../tutorial_files/exampleResults_reduced.pkl',"wb"))
# pickle.dump(results_stdprac, open('../tutorial_files/exampleResults_stdprac.pkl',"wb"))