Report Generation Tutorial

PyGSTi is able to construct polished report documents, which provide high-level summaries as well as detailed analyses of GST results. Reports are intended to be quick and easy way of analyzing a GST estimate, and pyGSTi's report generation functions are specifically designed to interact with its high-level driver functions (see the high-level algorithms tutorial). Currently there is only a single report generation function, pygsti.report.create_general_report, which takes one or more Results objects as input and produces an HTML file as output. The HTML format allows the reports to include interactive plots and switches, making it easy to compare different types of analysis or data sets.

PyGSTi's "general" report creates a stand-alone HTML document which cannot run Python. Thus, all the results displayed in the report must be pre-computed (in Python). If you find yourself wanting to fiddle with things and feel that the general report is too static, please consider using a Workspace object (see following tutorials) within a Jupyter notebook, where you can intermix report tables/plots and Python. Internally, create_general_report is just a canned routine which uses a WorkSpace object to generate various tables and plots and then inserts them into a HTML template.

Note to veteran users: PyGSTi has recently transitioned to producing HTML (rather than LaTeX/PDF) reports. The way to generate such report is largely unchanged, with one important exception. Previously, the Results object had various report-generation methods included within it. We've found this is too restrictive, as we'd sometimes like to generate a report which utilizes the results from multiple runs of GST (to compare them, for instance). Thus, the Results class is now just a container for a DataSet and its related GateSets, GatestringStructures, etc. All of the report-generation capability is now housed in within separate report functions, which we now demonstrate.

Get some Results

We start by performing GST using do_long_sequence_gst, as usual, to create a Results object (we could also have just loaded one from file).

In [1]:
import pygsti
from pygsti.construction import std1Q_XYI

gs_target = std1Q_XYI.gs_target
fiducials = std1Q_XYI.fiducials
germs = std1Q_XYI.germs
maxLengths = [1,2,4,8,16]
ds = pygsti.io.load_dataset("tutorial_files/Example_Dataset.txt", cache=True)

#Run GST
gs_target.set_all_parameterizations("TP") #TP-constrained
results = pygsti.do_long_sequence_gst(ds, gs_target, fiducials, fiducials, germs,
                                      maxLengths, verbosity=3)
Loading from cache file: tutorial_files/Example_Dataset.txt.cache
--- Gate Sequence Creation ---
   1282 sequences created
   Dataset has 3382 entries: 1282 utilized, 0 requested sequences were missing
--- LGST ---
  Singular values of I_tilde (truncating to first 4 of 6) = 
  4.245030583357432
  1.1797105733752995
  0.9564978918311123
  0.9423535266759971
  0.04708902142849772
  0.015314932955168468
  
  Singular values of target I_tilde (truncating to first 4 of 6) = 
  4.242640687119284
  1.4142135623730958
  1.4142135623730947
  1.4142135623730945
  2.31214371751647e-16
  4.4245234371241965e-17
  
    Resulting gate set:
    
    rho0 =    0.7071  -0.0302   0.0396   0.7480
    
    
    Mdefault = TP-POVM with effect vectors:
    0:
     0.73
       0
       0
     0.65
    
    1:
     0.69
       0
       0
    -0.65
    
    
    
    Gi = 
       1.0000        0        0        0
       0.0094   0.9238   0.0542  -0.0155
       0.0285  -0.0149   0.9021   0.0200
      -0.0142   0.0280   0.0009   0.9057
    
    
    Gx = 
       1.0000        0        0        0
       0.0064   0.9053   0.0281  -0.0044
      -0.0006   0.0215  -0.0471  -0.9983
      -0.0692  -0.0056   0.8095   0.0090
    
    
    Gy = 
       1.0000        0        0        0
      -0.0152  -0.0245   0.0379   0.9906
       0.0076  -0.0126   0.8876  -0.0257
      -0.0771  -0.8084  -0.0476   0.0210
    
    
    
    
--- Iterative MLGST: Iter 1 of 5  92 gate strings ---: 
  --- Minimum Chi^2 GST ---
    Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
     groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
    --- Outer Iter 0: norm_f = 62.7837, mu=0, |J|=1362.27
    --- Outer Iter 1: norm_f = 42.2737, mu=172.193, |J|=4003.22
    --- Outer Iter 2: norm_f = 41.0782, mu=57.3976, |J|=4002.28
    --- Outer Iter 3: norm_f = 41.0771, mu=19.1325, |J|=4002.28
    Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
  Sum of Chi^2 = 41.0771 (92 data params - 31 model params = expected mean of 61; p-value = 0.976519)
  Completed in 0.2s
  2*Delta(log(L)) = 41.2329
  Iteration 1 took 0.3s
  
--- Iterative MLGST: Iter 2 of 5  168 gate strings ---: 
  --- Minimum Chi^2 GST ---
    Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
     groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
    --- Outer Iter 0: norm_f = 175.573, mu=0, |J|=4118.16
    --- Outer Iter 1: norm_f = 124.232, mu=1805.46, |J|=4115.29
    --- Outer Iter 2: norm_f = 119.708, mu=601.819, |J|=4114.91
    --- Outer Iter 3: norm_f = 119.299, mu=200.606, |J|=4114.85
    --- Outer Iter 4: norm_f = 119.288, mu=66.8688, |J|=4114.84
    Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
  Sum of Chi^2 = 119.288 (168 data params - 31 model params = expected mean of 137; p-value = 0.859758)
  Completed in 0.3s
  2*Delta(log(L)) = 119.601
  Iteration 2 took 0.5s
  
--- Iterative MLGST: Iter 3 of 5  450 gate strings ---: 
  --- Minimum Chi^2 GST ---
    Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
     groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
    --- Outer Iter 0: norm_f = 498.161, mu=0, |J|=4505.2
    --- Outer Iter 1: norm_f = 416.334, mu=2013.07, |J|=4507.21
    --- Outer Iter 2: norm_f = 415.465, mu=671.024, |J|=4506.99
    --- Outer Iter 3: norm_f = 415.46, mu=223.675, |J|=4506.99
    Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
  Sum of Chi^2 = 415.46 (450 data params - 31 model params = expected mean of 419; p-value = 0.539658)
  Completed in 0.5s
  2*Delta(log(L)) = 415.96
  Iteration 3 took 0.8s
  
--- Iterative MLGST: Iter 4 of 5  862 gate strings ---: 
  --- Minimum Chi^2 GST ---
    Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
     groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
    --- Outer Iter 0: norm_f = 860.476, mu=0, |J|=5095.16
    --- Outer Iter 1: norm_f = 814.571, mu=2303.12, |J|=5078.86
    --- Outer Iter 2: norm_f = 814.34, mu=767.708, |J|=5078.57
    Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
  Sum of Chi^2 = 814.34 (862 data params - 31 model params = expected mean of 831; p-value = 0.65359)
  Completed in 0.6s
  2*Delta(log(L)) = 815.742
  Iteration 4 took 1.1s
  
--- Iterative MLGST: Iter 5 of 5  1282 gate strings ---: 
  --- Minimum Chi^2 GST ---
    Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
     groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
    --- Outer Iter 0: norm_f = 1265.58, mu=0, |J|=5735.19
    --- Outer Iter 1: norm_f = 1252.76, mu=2582.97, |J|=5736.35
    --- Outer Iter 2: norm_f = 1252.69, mu=860.99, |J|=5736.93
    Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
  Sum of Chi^2 = 1252.69 (1282 data params - 31 model params = expected mean of 1251; p-value = 0.481202)
  Completed in 0.9s
  2*Delta(log(L)) = 1254.49
  Iteration 5 took 1.7s
  
  Switching to ML objective (last iteration)
  --- MLGST ---
    Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
     groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
    --- Outer Iter 0: norm_f = 627.245, mu=0, |J|=3702.41
    --- Outer Iter 1: norm_f = 627.238, mu=3782.07, |J|=3378.33
    --- Outer Iter 2: norm_f = 627.234, mu=4.13103e+07, |J|=3299.12
    --- Outer Iter 3: norm_f = 627.225, mu=2.26458e+07, |J|=3086.64
    --- Outer Iter 4: norm_f = 627.218, mu=2.09848e+07, |J|=3021.71
    --- Outer Iter 5: norm_f = 627.217, mu=2.05917e+07, |J|=3128.49
    Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
    Maximum log(L) = 627.217 below upper bound of -2.13594e+06
      2*Delta(log(L)) = 1254.43 (1282 data params - 31 model params = expected mean of 1251; p-value = 0.467364)
    Completed in 1.8s
  2*Delta(log(L)) = 1254.43
  Final MLGST took 1.8s
  
Iterative MLGST Total Time: 6.1s
  -- Adding Gauge Optimized (go0) --

Make a report

Now that we have results, we use the create_standard_report method within pygsti.report.factory to generate a report. If the given filename ends in ".pdf" then a PDF-format report is generated; otherwise the file name specifies a folder that will be filled with HTML pages. To open a HTML-format report, you open the main.html file directly inside the report's folder. Setting auto_open=True makes the finished report open in your web browser automatically.

In [2]:
#HTML
pygsti.report.create_standard_report(results, "tutorial_files/exampleReport", 
                                     title="GST Example Report", verbosity=1, auto_open=True)

print("\n")

#PDF
pygsti.report.create_standard_report(results, "tutorial_files/exampleReport.pdf", 
                                     title="GST Example Report", verbosity=1, auto_open=True)
*** Creating workspace ***
*** Generating switchboard ***
Found standard clifford compilation from std1Q_XYI
*** Generating tables ***
*** Generating plots ***
*** Merging into template file ***
Output written to tutorial_files/exampleReport directory
Opening tutorial_files/exampleReport/main.html...
*** Report Generation Complete!  Total time 37.8488s ***


*** Creating workspace ***
*** Generating switchboard ***
Found standard clifford compilation from std1Q_XYI
*** Generating tables ***
*** Generating plots ***
*** Merging into template file ***
/usr/local/lib/python3.6/site-packages/matplotlib/cbook/deprecation.py:107: MatplotlibDeprecationWarning:

Passing one of 'on', 'true', 'off', 'false' as a boolean is deprecated; use an actual boolean (True/False) instead.

Latex file(s) successfully generated.  Attempting to compile with pdflatex...
------------
FileNotFoundErrorTraceback (most recent call last)
<ipython-input-2-3cc7d2a54f4e> in <module>()
      7 #PDF
      8 pygsti.report.create_standard_report(results, "tutorial_files/exampleReport.pdf", 
----> 9                                      title="GST Example Report", verbosity=1, auto_open=True)

~/Projects/pyGSTi/packages/pygsti/report/factory.py in create_standard_report(results, filename, title, confidenceLevel, comm, ws, auto_open, link_to, brevity, advancedOptions, verbosity)
   1042                 assert(cmd), "Cannot render PDF documents: no `latex_cmd` render option."
   1043                 printer.log("Latex file(s) successfully generated.  Attempting to compile with %s..." % cmd)
-> 1044                 _merge.compile_latex_report(base, [cmd] + flags, printer, auto_open)
   1045             else:
   1046                 raise ValueError("Unrecognized format: %s" % fmt)

~/Projects/pyGSTi/packages/pygsti/report/merge_helpers.py in compile_latex_report(report_filename, latex_call, printer, auto_open)
    804     try:
    805         #Run latex
--> 806         stdout, stderr, returncode = process_call(call)
    807         evaluate_call(call, stdout, stderr, returncode, printer)
    808         printer.log("Initial output PDF %s successfully generated." %

~/Projects/pyGSTi/packages/pygsti/report/merge_helpers.py in process_call(call)
    675     """
    676     process = _subprocess.Popen(call, stdout=_subprocess.PIPE,
--> 677                                 stderr=_subprocess.PIPE)
    678     stdout, stderr = process.communicate()
    679     return stdout, stderr, process.returncode

/usr/local/Cellar/python/3.6.5/Frameworks/Python.framework/Versions/3.6/lib/python3.6/subprocess.py in __init__(self, args, bufsize, executable, stdin, stdout, stderr, preexec_fn, close_fds, shell, cwd, env, universal_newlines, startupinfo, creationflags, restore_signals, start_new_session, pass_fds, encoding, errors)
    707                                 c2pread, c2pwrite,
    708                                 errread, errwrite,
--> 709                                 restore_signals, start_new_session)
    710         except:
    711             # Cleanup if the child failed starting.

/usr/local/Cellar/python/3.6.5/Frameworks/Python.framework/Versions/3.6/lib/python3.6/subprocess.py in _execute_child(self, args, executable, preexec_fn, close_fds, pass_fds, cwd, env, startupinfo, creationflags, shell, p2cread, p2cwrite, c2pread, c2pwrite, errread, errwrite, restore_signals, start_new_session)
   1342                         if errno_num == errno.ENOENT:
   1343                             err_msg += ': ' + repr(err_filename)
-> 1344                     raise child_exception_type(errno_num, err_msg, err_filename)
   1345                 raise child_exception_type(err_msg)
   1346 

FileNotFoundError: [Errno 2] No such file or directory: 'pdflatex': 'pdflatex'

There are several remarks about these reports worth noting:

  1. The HTML reports are the primary report type in pyGSTi, and are much more flexible. The PDF reports are more limited (they can only display a single estimate and gauge optimization), and essentially contain a subset of the information and descriptive text of a HTML report. So, if you can, use the HTML reports. The PDF report's strength is its portability: PDFs are easily displayed by many devices, and they embed all that they need neatly into a single file. If you need to generate a PDF report from Results objects that have multiple estimates and/or gauge optimizations, consider using the Results object's view method to single out the estimate and gauge optimization you're after.
  2. It's best to use Firefox when opening the HTML reports. (If there's a problem with your brower's capabilities it will be shown on the screen when you try to load the report.)
  3. You'll need pdflatex on your system to compile PDF reports.
  4. To familiarize yourself with the layout of an HTML report, click on the gray "Help" link on the black sidebar.

Multiple estimates in a single report

Next, let's analyze the same data two different ways: with and without the TP-constraint (i.e. whether the gates must be trace-preserving) and furthermore gauge optmimize each case using several different SPAM-weights. In each case we'll call do_long_sequence_gst with gaugeOptParams=False, so that no gauge optimization is done, and then perform several gauge optimizations separately and add these to the Results object via its add_gaugeoptimized function.

In [ ]:
#Case1: TP-constrained GST
tpTarget = gs_target.copy()
tpTarget.set_all_parameterizations("TP")
results_tp = pygsti.do_long_sequence_gst(ds, tpTarget, fiducials, fiducials, germs,
                                      maxLengths, gaugeOptParams=False, verbosity=1)

#Gauge optimize
est = results_tp.estimates['default']
gsFinal = est.gatesets['final iteration estimate']
gsTarget = est.gatesets['target']
for spamWt in [1e-4,1e-2,1.0]:
    gs = pygsti.gaugeopt_to_target(gsFinal,gsTarget,{'gates':1, 'spam':spamWt})
    est.add_gaugeoptimized({'itemWeights': {'gates':1, 'spam':spamWt}}, gs, "Spam %g" % spamWt)
In [ ]:
#Case2: "Full" GST
fullTarget = gs_target.copy()
fullTarget.set_all_parameterizations("full")
results_full = pygsti.do_long_sequence_gst(ds, fullTarget, fiducials, fiducials, germs,
                                           maxLengths, gaugeOptParams=False, verbosity=1)

#Gauge optimize
est = results_full.estimates['default']
gsFinal = est.gatesets['final iteration estimate']
gsTarget = est.gatesets['target']
for spamWt in [1e-4,1e-2,1.0]:
    gs = pygsti.gaugeopt_to_target(gsFinal,gsTarget,{'gates':1, 'spam':spamWt})
    est.add_gaugeoptimized({'itemWeights': {'gates':1, 'spam':spamWt}}, gs, "Spam %g" % spamWt)

We'll now call the same create_standard_report function but this time instead of passing a single Results object as the first argument we'll pass a dictionary of them. This will result in a HTML report that includes switches to select which case ("TP" or "Full") as well as which gauge optimization to display output quantities for. PDF reports cannot support this interactivity, and so if you try to generate a PDF report you'll get an error.

In [ ]:
ws = pygsti.report.create_standard_report({'TP': results_tp, "Full": results_full},
                                         "tutorial_files/exampleMultiEstimateReport",
                                         title="Example Multi-Estimate Report", 
                                         verbosity=2, auto_open=True)

In the above call we capture the return value in the variable ws - a Workspace object. PyGSTi's Workspace objects function as both a factory for figures and tables as well as a smart cache for computed values. Within create_standard_report a Workspace object is created and used to create all the figures in the report. As an intended side effect, each of these figures is cached, along with some of the intermediate results used to create it. As we'll see below, a Workspace can also be specified as input to create_standard_report, allowing it to utilize previously cached quantities.

Note to veteran users: Other report formats such as beamer-class PDF presentation and Powerpoint presentation have been dropped from pyGSTi. These presentation formats were rarely used and moreover we feel that the HTML format is able to provide all of the functionality that was present in these discontinued formats.

Another way: Because both results_tp and results_full above used the same dataset and gate sequences, we could have combined them as two estimates in a single Results object (see the previous tutorial on pyGSTi's Results object). This can be done by renaming at least one of the "default"-named estimates in results_tp or results_full (below we rename both) and then adding the estimate within results_full to the estimates already contained in results_tp:

In [ ]:
results_tp.rename_estimate('default','TP')
results_full.rename_estimate('default','Full')
results_both = results_tp.copy() #copy just for neatness
results_both.add_estimates(results_full, estimatesToAdd=['Full'])

Creating a report using results_both will result in the same report we just generated. We'll demonstrate this anyway, but in addition we'll supply create_standard_report a ws argument, which tells it to use any cached values contained in a given input Workspace to expedite report generation. Since our workspace object has the exact quantities we need cached in it, you'll notice a significant speedup. Finally, note that even though there's just a single Results object, you still can't generate a PDF report from it because it contains multiple estimates.

In [ ]:
pygsti.report.create_standard_report(results_both,
                                     "tutorial_files/exampleMultiEstimateReport2",
                                     title="Example Multi-Estimate Report (v2)", 
                                     verbosity=2, auto_open=True, ws=ws)

Multiple estimates and do_stdpractice_gst

It's no coincidence that a Results object containing multiple estimates using the same data is precisely what's returned from do_stdpractice_gst (see docstring for information on its arguments). This allows one to run GST multiple times, creating several different "standard" estimates and gauge optimizations, and plot them all in a single (HTML) report.

In [4]:
results_std = pygsti.do_stdpractice_gst(ds, gs_target, fiducials, fiducials, germs,
                                        maxLengths, verbosity=4, modes="TP,CPTP,Target",
                                        gaugeOptSuite=('single','toggleValidSpam'))

# Generate a report with "TP", "CPTP", and "Target" estimates
pygsti.report.create_standard_report(results_std, "tutorial_files/exampleStdReport", 
                                     title="Post StdPractice Report", auto_open=True,
                                     verbosity=1)
-- Std Practice:  Iter 1 of 3  (TP) --: 
  --- Gate Sequence Creation ---
     1282 sequences created
     Dataset has 3382 entries: 1282 utilized, 0 requested sequences were missing
  --- LGST ---
    Singular values of I_tilde (truncating to first 4 of 6) = 
    4.245030583357432
    1.1797105733752995
    0.9564978918311123
    0.9423535266759971
    0.04708902142849772
    0.015314932955168468
    
    Singular values of target I_tilde (truncating to first 4 of 6) = 
    4.242640687119284
    1.4142135623730958
    1.4142135623730947
    1.4142135623730945
    2.31214371751647e-16
    4.4245234371241965e-17
    
      Resulting gate set:
      
      rho0 =    0.7071  -0.0302   0.0396   0.7480
      
      
      Mdefault = TP-POVM with effect vectors:
      0:
       0.73
         0
         0
       0.65
      
      1:
       0.69
         0
         0
      -0.65
      
      
      
      Gi = 
         1.0000        0        0        0
         0.0094   0.9238   0.0542  -0.0155
         0.0285  -0.0149   0.9021   0.0200
        -0.0142   0.0280   0.0009   0.9057
      
      
      Gx = 
         1.0000        0        0        0
         0.0064   0.9053   0.0281  -0.0044
        -0.0006   0.0215  -0.0471  -0.9983
        -0.0692  -0.0056   0.8095   0.0090
      
      
      Gy = 
         1.0000        0        0        0
        -0.0152  -0.0245   0.0379   0.9906
         0.0076  -0.0126   0.8876  -0.0257
        -0.0771  -0.8084  -0.0476   0.0210
      
      
      
      
  --- Iterative MLGST: Iter 1 of 5  92 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 62.7837, mu=0, |J|=1362.27
      --- Outer Iter 1: norm_f = 42.2737, mu=172.193, |J|=4003.22
      --- Outer Iter 2: norm_f = 41.0782, mu=57.3976, |J|=4002.28
      --- Outer Iter 3: norm_f = 41.0771, mu=19.1325, |J|=4002.28
      Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
    Sum of Chi^2 = 41.0771 (92 data params - 31 model params = expected mean of 61; p-value = 0.976519)
    Completed in 0.2s
    2*Delta(log(L)) = 41.2329
    Iteration 1 took 0.3s
    
  --- Iterative MLGST: Iter 2 of 5  168 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 175.573, mu=0, |J|=4118.16
      --- Outer Iter 1: norm_f = 124.232, mu=1805.46, |J|=4115.29
      --- Outer Iter 2: norm_f = 119.708, mu=601.819, |J|=4114.91
      --- Outer Iter 3: norm_f = 119.299, mu=200.606, |J|=4114.85
      --- Outer Iter 4: norm_f = 119.288, mu=66.8688, |J|=4114.84
      Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
    Sum of Chi^2 = 119.288 (168 data params - 31 model params = expected mean of 137; p-value = 0.859758)
    Completed in 0.3s
    2*Delta(log(L)) = 119.601
    Iteration 2 took 0.4s
    
  --- Iterative MLGST: Iter 3 of 5  450 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 498.161, mu=0, |J|=4505.2
      --- Outer Iter 1: norm_f = 416.334, mu=2013.07, |J|=4507.21
      --- Outer Iter 2: norm_f = 415.465, mu=671.024, |J|=4506.99
      --- Outer Iter 3: norm_f = 415.46, mu=223.675, |J|=4506.99
      Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
    Sum of Chi^2 = 415.46 (450 data params - 31 model params = expected mean of 419; p-value = 0.539658)
    Completed in 0.5s
    2*Delta(log(L)) = 415.96
    Iteration 3 took 0.8s
    
  --- Iterative MLGST: Iter 4 of 5  862 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 860.476, mu=0, |J|=5095.16
      --- Outer Iter 1: norm_f = 814.571, mu=2303.12, |J|=5078.86
      --- Outer Iter 2: norm_f = 814.34, mu=767.708, |J|=5078.57
      Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
    Sum of Chi^2 = 814.34 (862 data params - 31 model params = expected mean of 831; p-value = 0.65359)
    Completed in 0.7s
    2*Delta(log(L)) = 815.742
    Iteration 4 took 1.2s
    
  --- Iterative MLGST: Iter 5 of 5  1282 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 1265.58, mu=0, |J|=5735.19
      --- Outer Iter 1: norm_f = 1252.76, mu=2582.97, |J|=5736.35
      --- Outer Iter 2: norm_f = 1252.69, mu=860.99, |J|=5736.93
      Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
    Sum of Chi^2 = 1252.69 (1282 data params - 31 model params = expected mean of 1251; p-value = 0.481202)
    Completed in 1.0s
    2*Delta(log(L)) = 1254.49
    Iteration 5 took 1.8s
    
    Switching to ML objective (last iteration)
    --- MLGST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 627.245, mu=0, |J|=3702.41
      --- Outer Iter 1: norm_f = 627.238, mu=3782.07, |J|=3378.33
      --- Outer Iter 2: norm_f = 627.234, mu=4.13103e+07, |J|=3299.12
      --- Outer Iter 3: norm_f = 627.225, mu=2.26458e+07, |J|=3086.64
      --- Outer Iter 4: norm_f = 627.218, mu=2.09848e+07, |J|=3021.71
      --- Outer Iter 5: norm_f = 627.217, mu=2.05917e+07, |J|=3128.49
      Least squares message = Both actual and predicted relative reductions in the sum of squares are at most 1e-06
      Maximum log(L) = 627.217 below upper bound of -2.13594e+06
        2*Delta(log(L)) = 1254.43 (1282 data params - 31 model params = expected mean of 1251; p-value = 0.467364)
      Completed in 1.9s
    2*Delta(log(L)) = 1254.43
    Final MLGST took 1.9s
    
  Iterative MLGST Total Time: 6.3s
  -- Performing 'single' gauge optimization on TP estimate --
      -- Adding Gauge Optimized (single) --
  -- Performing 'Spam 0.001' gauge optimization on TP estimate --
      -- Adding Gauge Optimized (Spam 0.001) --
  -- Performing 'Spam 0.001+v' gauge optimization on TP estimate --
      -- Adding Gauge Optimized (Spam 0.001+v) --
-- Std Practice:  Iter 2 of 3  (CPTP) --: 
  --- Gate Sequence Creation ---
     1282 sequences created
     Dataset has 3382 entries: 1282 utilized, 0 requested sequences were missing
  --- Iterative MLGST: Iter 1 of 5  92 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 1.10824e+07, mu=0, |J|=21595.7
      --- Outer Iter 1: norm_f = 1055.69, mu=51712.5, |J|=1014.07
      --- Outer Iter 2: norm_f = 906.343, mu=17237.5, |J|=958.022
      --- Outer Iter 3: norm_f = 890.808, mu=5745.83, |J|=941.716
      --- Outer Iter 4: norm_f = 887.155, mu=1915.28, |J|=940.07
      --- Outer Iter 5: norm_f = 886.12, mu=638.426, |J|=940.129
      --- Outer Iter 6: norm_f = 886.031, mu=212.809, |J|=940.16
      --- Outer Iter 7: norm_f = 885.532, mu=70.9362, |J|=940.053
      --- Outer Iter 8: norm_f = 829.351, mu=189.163, |J|=926.197
      --- Outer Iter 9: norm_f = 639.664, mu=504.435, |J|=905.287
      --- Outer Iter 10: norm_f = 468.911, mu=1020.24, |J|=863.757
      --- Outer Iter 11: norm_f = 144.284, mu=340.081, |J|=1014.76
      --- Outer Iter 12: norm_f = 134.688, mu=113.36, |J|=982.031
      --- Outer Iter 13: norm_f = 133.055, mu=113.81, |J|=968.439
      --- Outer Iter 14: norm_f = 127.127, mu=95.4437, |J|=966.489
      --- Outer Iter 15: norm_f = 101.018, mu=63.6291, |J|=951.557
      --- Outer Iter 16: norm_f = 79.1423, mu=57.8456, |J|=949.601
      --- Outer Iter 17: norm_f = 55.2669, mu=19744.6, |J|=3971.7
      --- Outer Iter 18: norm_f = 53.4873, mu=8123.49, |J|=3974.72
      --- Outer Iter 19: norm_f = 52.6336, mu=5158.47, |J|=3975.96
      --- Outer Iter 20: norm_f = 51.6616, mu=2895.56, |J|=3977.28
      --- Outer Iter 21: norm_f = 50.1915, mu=965.188, |J|=3979
      --- Outer Iter 22: norm_f = 47.4829, mu=321.729, |J|=3981.8
      --- Outer Iter 23: norm_f = 44.9263, mu=107.243, |J|=3985.99
      --- Outer Iter 24: norm_f = 43.5176, mu=35.7477, |J|=3992.24
      --- Outer Iter 25: norm_f = 43.1156, mu=31.473, |J|=1802.81
      --- Outer Iter 26: norm_f = 41.9408, mu=10.491, |J|=4017.9
      --- Outer Iter 27: norm_f = 41.2585, mu=8.25944, |J|=4022.01
      --- Outer Iter 28: norm_f = 41.0862, mu=2.75315, |J|=4023.48
      --- Outer Iter 29: norm_f = 41.0859, mu=4.98069, |J|=4025.25
      --- Outer Iter 30: norm_f = 41.0858, mu=9.8871, |J|=4027.21
      Least squares message = Relative change in |x| is at most 1e-08
    Sum of Chi^2 = 41.1114 (92 data params - 31 model params = expected mean of 61; p-value = 0.976293)
    Completed in 0.9s
    2*Delta(log(L)) = 44.6096
    Iteration 1 took 0.9s
    
  --- Iterative MLGST: Iter 2 of 5  168 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 179.784, mu=0, |J|=4102.96
      --- Outer Iter 1: norm_f = 138.536, mu=2285.72, |J|=4101.71
      --- Outer Iter 2: norm_f = 125.246, mu=761.907, |J|=4104.28
      --- Outer Iter 3: norm_f = 120.441, mu=253.969, |J|=4105.63
      --- Outer Iter 4: norm_f = 119.465, mu=84.6563, |J|=4106.02
      --- Outer Iter 5: norm_f = 119.322, mu=62.5071, |J|=4106.74
      --- Outer Iter 6: norm_f = 119.317, mu=81.3216, |J|=4107.73
      --- Outer Iter 7: norm_f = 119.315, mu=139.502, |J|=4108.97
      --- Outer Iter 8: norm_f = 119.314, mu=242.128, |J|=4110.48
      --- Outer Iter 9: norm_f = 119.313, mu=426.252, |J|=4112.26
      --- Outer Iter 10: norm_f = 119.312, mu=771.941, |J|=4114.3
      --- Outer Iter 11: norm_f = 119.312, mu=1478.24, |J|=4116.6
      Least squares message = Relative change in |x| is at most 1e-08
    Sum of Chi^2 = 119.375 (168 data params - 31 model params = expected mean of 137; p-value = 0.858461)
    Completed in 0.6s
    2*Delta(log(L)) = 122.407
    Iteration 2 took 0.7s
    
  --- Iterative MLGST: Iter 3 of 5  450 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 520.731, mu=0, |J|=4388.01
      --- Outer Iter 1: norm_f = 425.747, mu=2348.99, |J|=4387.99
      --- Outer Iter 2: norm_f = 418.289, mu=782.998, |J|=4387.4
      --- Outer Iter 3: norm_f = 416.05, mu=365.617, |J|=4387.75
      --- Outer Iter 4: norm_f = 415.687, mu=355.573, |J|=4389.64
      --- Outer Iter 5: norm_f = 415.658, mu=503.681, |J|=4392.13
      --- Outer Iter 6: norm_f = 415.643, mu=802.345, |J|=4395.23
      --- Outer Iter 7: norm_f = 415.629, mu=1291.1, |J|=4399.08
      --- Outer Iter 8: norm_f = 415.617, mu=2101.57, |J|=4403.66
      --- Outer Iter 9: norm_f = 415.607, mu=3490.92, |J|=4408.96
      --- Outer Iter 10: norm_f = 415.599, mu=6019.34, |J|=4414.94
      --- Outer Iter 11: norm_f = 415.596, mu=11138.8, |J|=4421.59
      Least squares message = Relative change in |x| is at most 1e-08
    Sum of Chi^2 = 415.944 (450 data params - 31 model params = expected mean of 419; p-value = 0.532983)
    Completed in 1.0s
    2*Delta(log(L)) = 431.785
    Iteration 3 took 1.3s
    
  --- Iterative MLGST: Iter 4 of 5  862 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 893.164, mu=0, |J|=4750.25
      --- Outer Iter 1: norm_f = 815.33, mu=2372.71, |J|=4738.51
      --- Outer Iter 2: norm_f = 814.675, mu=2184.19, |J|=4739.75
      --- Outer Iter 3: norm_f = 814.612, mu=2654.57, |J|=4742.24
      --- Outer Iter 4: norm_f = 814.583, mu=3911.8, |J|=4746.52
      --- Outer Iter 5: norm_f = 814.565, mu=6296.89, |J|=4752.54
      --- Outer Iter 6: norm_f = 814.552, mu=10677.9, |J|=4760.2
      --- Outer Iter 7: norm_f = 814.545, mu=19253, |J|=4769.42
      Least squares message = Relative change in |x| is at most 1e-08
    Sum of Chi^2 = 815.074 (862 data params - 31 model params = expected mean of 831; p-value = 0.646841)
    Completed in 1.3s
    2*Delta(log(L)) = 823.722
    Iteration 4 took 1.8s
    
  --- Iterative MLGST: Iter 5 of 5  1282 gate strings ---: 
    --- Minimum Chi^2 GST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 1289.5, mu=0, |J|=5010.07
      --- Outer Iter 1: norm_f = 1254.01, mu=2582.86, |J|=5006.93
      --- Outer Iter 2: norm_f = 1253.18, mu=2427.07, |J|=5010.14
      --- Outer Iter 3: norm_f = 1253.05, mu=2664.24, |J|=5016.29
      --- Outer Iter 4: norm_f = 1253.01, mu=3781.49, |J|=5025.33
      --- Outer Iter 5: norm_f = 1252.98, mu=5981.45, |J|=5037.22
      --- Outer Iter 6: norm_f = 1252.96, mu=9875.24, |J|=5051.82
      --- Outer Iter 7: norm_f = 1252.95, mu=16863.4, |J|=5068.99
      --- Outer Iter 8: norm_f = 1252.94, mu=30266.9, |J|=5088.58
      --- Outer Iter 9: norm_f = 1252.94, mu=59685.2, |J|=5110.48
      Least squares message = Relative change in |x| is at most 1e-08
    Sum of Chi^2 = 1253.54 (1282 data params - 31 model params = expected mean of 1251; p-value = 0.47442)
    Completed in 2.2s
    2*Delta(log(L)) = 1270.99
    Iteration 5 took 3.0s
    
    Switching to ML objective (last iteration)
    --- MLGST ---
      Created evaluation tree with 1 subtrees.  Will divide 1 procs into 1 (subtree-processing)
       groups of ~1 procs each, to distribute over 43 params (taken as 1 param groups of ~43 params).
      --- Outer Iter 0: norm_f = 635.496, mu=0, |J|=3112.36
      --- Outer Iter 1: norm_f = 627.361, mu=979.324, |J|=2703.11
      Least squares message = Relative change in |x| is at most 1e-08
      Maximum log(L) = 627.761 below upper bound of -2.13594e+06
        2*Delta(log(L)) = 1255.52 (1282 data params - 31 model params = expected mean of 1251; p-value = 0.458742)
      Completed in 1.3s
    2*Delta(log(L)) = 1255.52
    Final MLGST took 1.3s
    
  Iterative MLGST Total Time: 8.9s
  -- Performing 'single' gauge optimization on CPTP estimate --
      -- Adding Gauge Optimized (single) --
  -- Performing 'Spam 0.001' gauge optimization on CPTP estimate --
      -- Adding Gauge Optimized (Spam 0.001) --
  -- Performing 'Spam 0.001+v' gauge optimization on CPTP estimate --
      -- Adding Gauge Optimized (Spam 0.001+v) --
-- Std Practice:  Iter 3 of 3  (Target) --: 
  --- Gate Sequence Creation ---
     1282 sequences created
     Dataset has 3382 entries: 1282 utilized, 0 requested sequences were missing
  -- Performing 'single' gauge optimization on Target estimate --
      -- Adding Gauge Optimized (single) --
  -- Performing 'Spam 0.001' gauge optimization on Target estimate --
      -- Adding Gauge Optimized (Spam 0.001) --
  -- Performing 'Spam 0.001+v' gauge optimization on Target estimate --
      -- Adding Gauge Optimized (Spam 0.001+v) --
*** Creating workspace ***
*** Generating switchboard ***
Found standard clifford compilation from std1Q_XYI
Found standard clifford compilation from std1Q_XYI
Found standard clifford compilation from std1Q_XYI
*** Generating tables ***
*** Generating plots ***
*** Merging into template file ***
Output written to tutorial_files/exampleStdReport directory
Opening tutorial_files/exampleStdReport/main.html...
*** Report Generation Complete!  Total time 122.583s ***
Out[4]:
<pygsti.report.workspace.Workspace at 0x11ce6a0b8>

Reports with confidence regions

To display confidence intervals for reported quantities, you must do two things:

  1. you must specify the confidenceLevel argument to create_standard_report.
  2. the estimate(s) being reported must have a valid confidence-region-factory.

Constructing a factory often means computing a Hessian, which can be time consuming, and so this is not done automatically. Here we demonstrate how to construct a valid factory for the "Spam 0.001" gauge-optimization of the "CPTP" estimate by computing and then projecting the Hessian of the likelihood function.

In [5]:
#Construct and initialize a "confidence region factory" for the CPTP estimate
crfact = results_std.estimates["CPTP"].add_confidence_region_factory('Spam 0.001', 'final')
crfact.compute_hessian(comm=None) #we could use more processors
crfact.project_hessian('intrinsic error')

pygsti.report.create_standard_report(results_std, "tutorial_files/exampleStdReport2", 
                                     title="Post StdPractice Report (w/CIs on CPTP)",
                                     confidenceLevel=95, auto_open=True, verbosity=1)
    
--- Hessian Projector Optimization from separate SPAM and Gate weighting ---
  Resulting intrinsic errors: 0.0119628 (gates), 0.000159745 (spam)
  Resulting sqrt(mean(gateCIs**2)): 0.0144373
  Resulting sqrt(mean(spamCIs**2)): 0.00181812
*** Creating workspace ***
*** Generating switchboard ***
Found standard clifford compilation from std1Q_XYI
Found standard clifford compilation from std1Q_XYI
Found standard clifford compilation from std1Q_XYI
*** Generating tables ***
*** Generating plots ***
*** Merging into template file ***
Output written to tutorial_files/exampleStdReport2 directory
Opening tutorial_files/exampleStdReport2/main.html...
*** Report Generation Complete!  Total time 131.073s ***
Out[5]:
<pygsti.report.workspace.Workspace at 0x1198c3cc0>

Reports with multiple different data sets

We've already seen above that create_standard_report can be given a dictionary of Results objects instead of a single one. This allows the creation of reports containing estimates for different DataSets (each Results object only holds estimates for a single DataSet). Furthermore, when the data sets have the same gate sequences, they will be compared within a tab of the HTML report.

Below, we generate a new data set with the same sequences as the one loaded at the beginning of this tutorial, proceed to run standard-practice GST on that dataset, and create a report of the results along with those of the original dataset. Look at the "Data Comparison" tab within the gauge-invariant error metrics category.

In [ ]:
#Make another dataset & estimates
depol_gateset = gs_target.depolarize(gate_noise=0.1)
datagen_gateset = depol_gateset.rotate((0.05,0,0.03))

#Compute the sequences needed to perform Long Sequence GST on 
# this GateSet with sequences up to lenth 512
gatestring_list = pygsti.construction.make_lsgst_experiment_list(
    std1Q_XYI.gs_target, std1Q_XYI.prepStrs, std1Q_XYI.effectStrs,
    std1Q_XYI.germs, [1,2,4,8,16,32,64,128,256,512])
ds2 = pygsti.construction.generate_fake_data(datagen_gateset, gatestring_list, nSamples=1000,
                                             sampleError='binomial', seed=2018)
results_std2 = pygsti.do_stdpractice_gst(ds2, gs_target, fiducials, fiducials, germs,
                                     maxLengths, verbosity=3, modes="TP,CPTP,Target",
                                     gaugeOptSuite=('single','toggleValidSpam'))

pygsti.report.create_standard_report({'DS1': results_std, 'DS2': results_std2},
                                    "tutorial_files/exampleMultiDataSetReport", 
                                    title="Example Multi-Dataset Report", 
                                    auto_open=True, verbosity=1)

Other cool create_standard_report options

Finally, let us highlight a few of the additional arguments one can supply to create_standard_report that allows further control over what gets reported.

  • Setting the link_to argument to a tuple of 'pkl', 'tex', and/or 'pdf' will create hyperlinks within the plots or below the tables of the HTML linking to Python pickle, LaTeX source, and PDF versions of the content, respectively. The Python pickle files for tables contain pickled pandas DataFrame objects, wheras those of plots contain ordinary Python dictionaries of the data that is plotted. Applies to HTML reports only.

  • Setting the brevity argument to an integer higher than $0$ (the default) will reduce the amount of information included in the report (for details on what is included for each value, see the doc string). Using brevity > 0 will reduce the time required to create, and later load, the report, as well as the output file/folder size. This applies to both HTML and PDF reports.

Below, we demonstrate both of these options in very brief (brevity=4) report with links to pickle and PDF files. Note that to generate the PDF files you must have pdflatex installed.

In [ ]:
pygsti.report.create_standard_report(results_std,
                                    "tutorial_files/exampleBriefReport", 
                                    title="Example Brief Report", 
                                    auto_open=True, verbosity=1,
                                    brevity=4, link_to=('pkl','pdf'))

Advanced Reports: create_report_notebook

In addition to the standard HTML-page reports demonstrated above, pyGSTi is able to generate a Jupyter notebook containing the Python commands to create the figures and tables within a general report. This is facilitated by Workspace objects, which are factories for figures and tables (see previous tutorials). By calling create_report_notebook, all of the relevant Workspace initialization and calls are dumped to a new notebook file, which can be run (either fully or partially) by the user at their convenience. Creating such "report notebooks" has the advantage that the user may insert Python code amidst the figure and table generation calls to inspect or modify what is display in a highly customizable fashion. The chief disadvantages of report notebooks is that they require the user to 1) have a Jupyter server up and running and 2) to run the notebook before any figures are displayed.

The line below demonstrates how to create a report notebook using create_report_notebook. Note that the argument list is very similar to create_general_report.

In [ ]:
pygsti.report.create_report_notebook(results, "tutorial_files/exampleReport.ipynb", 
                                     title="GST Example Report Notebook", confidenceLevel=None,
                                     auto_open=True, connected=False, verbosity=3)
In [ ]: