Merge branch 'beta' to become version 0.9.7.2

Author: Erik Nielsen

.gitignore

@@ -47,6 +47,7 @@ packages/pygsti.egg-info
 packages/pygsti/baseobjs/parsetab_string.py
 packages/pygsti/objects/fastgatecalc.cpp
 packages/pygsti/objects/fastreplib.cpp
+packages/pygsti/objects/fastopcalc.cpp
 packages/pygsti/tools/fastcalc.c
 
 

.travis.yml

@@ -11,49 +11,65 @@ branches:
     - beta
     - develop
 
+# Hold off testing feature & bugfix branches in case this competes w/other testing.
+#    - /^feature-.*$/
+#    - /^bugfix-.*$/
+
 env:
-    - Drivers=True    # First in build mx because long duration
-    - Algorithms=True 
-    - AlgorithmsB=True
-    - Default=True    # construction, io, objects, tools, optimize
-    - Report=True      
-    - ReportB=True
-    - MPI=True        
+  global:
+    - NOSEOPTS_DEFAULT="-v --with-timer --with-id --rednose"
+    - NOSEOPTS_PARALLEL="--processes=-1 --process-timeout=2400" # timeout after 40 minutes (travis job timeout is 50 minutes)
+    - NOSEOPTS_ALL="$NOSEOPTS_DEFAULT $NOSEOPTS_PARALLEL"
+  matrix:
+    - NOSETESTS="drivers objects" NOSEOPTS=$NOSEOPTS_ALL # first in build mx because long duration
+    - NOSETESTS="algorithms" NOSEOPTS=$NOSEOPTS_ALL
+    - NOSETESTS="algorithmsb" NOSEOPTS=$NOSEOPTS_ALL
+    - NOSETESTS="tools iotest optimize construction extras" NOSEOPTS=$NOSEOPTS_ALL
+    - NOSETESTS="report" NOSEOPTS=$NOSEOPTS_ALL
+    - NOSETESTS="reportb" NOSEOPTS=$NOSEOPTS_ALL
+    - NOSETESTS="mpi" NOSEOPTS=$NOSEOPTS_DEFAULT # cannot be run in parallel
 
+# Install native package dependencies & initialize build environment
 before_install:
-  - sudo apt-get -qq update
-  - sudo apt-get -qq install gfortran libblas-dev liblapack-dev
-  - sudo apt-get -qq install openmpi-bin openmpi-common openssh-client openssh-server libopenmpi1.6 libopenmpi1.6-dbg libopenmpi-dev > /dev/null
-  - sudo bash CI/install.sh # travis_wait 30 "sudo bash CI/install.sh" # > /dev/null 2>&1"
-  - "export DISPLAY=:99.0"
-  - "sh -e /etc/init.d/xvfb start" #some tests require a display
-  - sleep 3 # give xvfb some time to start
+  - >
+    sudo apt-get update -qq -y &&
+    sudo apt-get install -qq -y
+    gfortran libblas-dev liblapack-dev openmpi-bin openmpi-common openssh-client
+    openssh-server libopenmpi1.6 libopenmpi1.6-dbg libopenmpi-dev
+    libsuitesparse-dev
+  - export DISPLAY=:99.0
+  - /bin/sh -e /etc/init.d/xvfb start && sleep 3 # some tests require a display
+  - cmake --version
+  - gcc --version
 
+# Install python package dependencies
 install:
-  - pip install zmq coverage >/dev/null
   - pip install -e .[travisci]
-  
+
+# Default `test' stage script
+script:
+  - cd test/test_packages
+  - nosetests $NOSEOPTS $NOSETESTS
+
+# Cache pip packages
 cache:
   pip: true
   timeout: 1000
 
-#Script For default "test" stage
-script:
-  - cd test
-  - python travisTests.py # just produce test_packages/runTravisTests.sh
-  - cd test_packages
-  - source runTravisTests.sh # no need for travis_wait (always outputs every 9m)
-  - cd ../..
+stages:
+  - test
+  - name: deploy # only run deployment for master & develop branches
+    if: branch IN (master, develop)
 
 jobs:
   include:
     - stage: deploy
       python: 3.5
-      env: Drivers=False Algorithms=False AlgorithmsB=False Default=False Report=False ReportB=False MPI=False
+      env: NOSETESTS=none NOSEOPTS=$NOSEOPTS_DEFAULT
+      install: skip
       script: python CI/deploy
 
 notifications:
-    email:
-        on_success: change
-        on_failure: always
-
+  email:
+    on_success: change
+    on_failure: always

CI/install.sh

@@ -16,6 +16,7 @@ include jupyter_notebooks/Tutorials/other/advanced/*.ipynb
 include jupyter_notebooks/Examples/*.ipynb
 include juptyer_notebooks/Tutorials/tutorial_files/.placeholder
 include jupyter_notebooks/Examples/example_files/.git_placeholder
+include scripts/update2v0.9.7.py
 include test/runTests.py
 include test/travisTests.py
 include test/__init__.py

MANIFEST.in

@@ -36,7 +36,7 @@ PyGSTi is designed with a modular structure so as to be highly customizable
 and easily integrated to new or existing python software.  It runs using
 python2.7 or python3.  To faclilitate integration with software for running
 cloud-QIP experiments, pyGSTi `Circuit` objects can be converted to IBM's
-**QASM** and Rigetti Quantum Computing's **QUIL** circuit description languages.
+**OpenQASM** and Rigetti Quantum Computing's **Quil** circuit description languages.
 
 Installation
 ------------
@@ -151,11 +151,15 @@ Tutorials and Examples
 ----------------------
 There are numerous tutorials (meant to be pedagogical) and examples (meant to be demonstrate
 how to do some particular thing) in the form of Jupyter notebooks beneath the `pyGSTi/jupyter_notebooks`
-directory.  The [START HERE](https://github.com/pyGSTio/pyGSTi/blob/master/jupyter_notebooks/START_HERE.ipynb)
-notebook will direct you where to go based on what you're most interested in learning about.  One possible
-direction is to view the [FAQ](https://github.com/pyGSTio/pyGSTi/blob/master/jupyter_notebooks/FAQ.ipynb), which
+directory.  The root "START HERE" notebook will direct you where to go based on what you're most
+interested in learning about.  You can view the
+[read-only GitHub version of this notebook](https://github.com/pyGSTio/pyGSTi/blob/master/jupyter_notebooks/START_HERE.ipynb) 
+or you can [explore the tutorials interactively](https://mybinder.org/v2/gh/pyGSTio/pyGSTi/master)
+using JupyterHub via Binder.  Note the existence of a
+[FAQ](https://github.com/pyGSTio/pyGSTi/blob/master/jupyter_notebooks/FAQ.ipynb), which
 addresses common issues.
 
+
 #### Running notebooks *locally*
 While it's possible to view the notebooks on GitHub using the links above, it's
 usually nicer to run them *locally* so you can mess around with the code as

README.md

@@ -4,4 +4,4 @@
 #    in the file "license.txt" in the top-level pyGSTi directory
 #*****************************************************************
 """ PyGSTi version number """
-__version__ = "0.9.7.1"
+__version__ = "0.9.7.2"

packages/pygsti/_version.py

@@ -179,7 +179,7 @@ def do_lgst(dataset, prepStrs, effectStrs, targetModel, opLabels=None, opLabelAl
     if svdTruncateTo is None or svdTruncateTo == targetModel.dim: #use target sslbls and basis
         lgstModel = _objs.ExplicitOpModel(targetModel.state_space_labels, targetModel.basis)
     else: # construct a default basis for the requested dimension 
-        dumb_basis = _objs.Basis('gm',[1]*svdTruncateTo) # - just act on diagonal density mx
+        dumb_basis = _objs.DirectSumBasis( [_objs.BuiltinBasis('gm',1)]*svdTruncateTo) # - just act on diagonal density mx
         lgstModel = _objs.ExplicitOpModel([('L%d'%i,) for i in range(svdTruncateTo)], dumb_basis)
 
     for opLabel in opLabelsToEstimate:
@@ -2437,6 +2437,7 @@ def _objective_func(vectorGS):
             v = _np.where( probs < min_p, v + S*(probs - min_p) + S2*(probs - min_p)**2, v) #quadratic extrapolation of logl at min_p for probabilities < min_p
             v = _np.where( minusCntVecMx == 0, totalCntVec * _np.where(probs >= a, probs, (-1.0/(3*a**2))*probs**3 + probs**2/a + a/3.0), v)
                     #special handling for f == 0 terms using quadratic rounding of function with minimum: max(0,(a-p)^2)/(2a) + p
+            #assert( _np.all(v >= 0) ), "LogL term is < 0! (This is usually caused by using a large #samples without reducing minProbClip)"
             v = _np.sqrt( v )
             v.shape = [KM] #reshape ensuring no copy is needed
             if cptp_penalty_factor != 0:
@@ -2534,6 +2535,7 @@ def _objective_func(vectorGS):
             v = _np.maximum(v,0)  #remove small negative elements due to roundoff error (above expression *cannot* really be negative)
             v = _np.where( probs < min_p, v + S*(probs - min_p) + S2*(probs - min_p)**2, v) #quadratic extrapolation of logl at min_p for probabilities < min_p
             v = _np.where( minusCntVecMx == 0, 0.0, v)
+            #assert( _np.all(v >= 0) ), "LogL term is < 0! (This is usually caused by using a large #samples without reducing minProbClip)"
             v = _np.sqrt( v )
             assert(v.shape == (KM,)) #reshape ensuring no copy is needed
 
@@ -2687,7 +2689,7 @@ def do_iterative_mlgst(dataset, startModel, circuitSetsToUseInEstimation,
                        verbosity=0, check=False, circuitWeightsDict=None,
                        opLabelAliases=None, memLimit=None,
                        profiler=None, comm=None, distributeMethod = "deriv",
-                       alwaysPerformMLE=False, evaltree_cache=None):
+                       alwaysPerformMLE=False, onlyPerformMLE=False, evaltree_cache=None):
     """
     Performs Iterative Maximum Likelihood Estimation Gate Set Tomography on the dataset.
 
@@ -2806,6 +2808,10 @@ def do_iterative_mlgst(dataset, startModel, circuitSetsToUseInEstimation,
         not just the final one.  When False, chi2 minimization is used for all
         except the final iteration (for improved numerical stability).
 
+    onlyPerformMLE : bool, optional
+        When True, `alwaysPerformMLE` must also be true, and in this case only 
+        a ML optimization is performed for each iteration.
+
     evaltree_cache : dict, optional
         An empty dictionary which gets filled with the *final* computed EvalTree
         (and supporting info) used in this computation.
@@ -2823,6 +2829,9 @@ def do_iterative_mlgst(dataset, startModel, circuitSetsToUseInEstimation,
         of the i-th iteration.
     """
 
+    if onlyPerformMLE:
+        assert(alwaysPerformMLE), "Must set `alwaysPerformMLE` to True whenever `onlyPerformMLE` is True."
+        
     printer = _objs.VerbosityPrinter.build_printer(verbosity, comm)
     if profiler is None: profiler = _dummy_profiler
 
@@ -2864,12 +2873,13 @@ def do_iterative_mlgst(dataset, startModel, circuitSetsToUseInEstimation,
             num_fd = fditer if (i == 0) else 0
 
             evt_cache = {} # get the eval tree that's created so we can reuse it
-            _, mleModel = do_mc2gst(dataset, mleModel, stringsToEstimate,
-                                      maxiter, maxfev, num_fd, tol, cptp_penalty_factor,
-                                      spam_penalty_factor, minProbClip, probClipInterval,
-                                      useFreqWeightedChiSq, 0,printer-1, check,
-                                      check, circuitWeights, opLabelAliases,
-                                      memLimit, comm, distributeMethod, profiler, evt_cache)
+            if not onlyPerformMLE:
+                _, mleModel = do_mc2gst(dataset, mleModel, stringsToEstimate,
+                                        maxiter, maxfev, num_fd, tol, cptp_penalty_factor,
+                                        spam_penalty_factor, minProbClip, probClipInterval,
+                                        useFreqWeightedChiSq, 0,printer-1, check,
+                                        check, circuitWeights, opLabelAliases,
+                                        memLimit, comm, distributeMethod, profiler, evt_cache)
 
             if alwaysPerformMLE:
                 _, mleModel = do_mlgst(dataset, mleModel, stringsToEstimate,
@@ -3054,7 +3064,7 @@ def _spam_penalty_jac_fill(spamPenaltyVecGradToFill, mdl, prefactor, opBasis):
     Helper function - jacobian of CPTP penalty (sum of tracenorms of gates)
     Returns a (real) array of shape ( _spam_penalty_size(mdl), nParams).
     """
-    BMxs = opBasis.get_composite_matrices() #shape [mdl.dim, dmDim, dmDim]
+    BMxs = opBasis.elements #shape [mdl.dim, dmDim, dmDim]
     ddenMxdV = dEMxdV = BMxs.conjugate() # b/c denMx = sum( spamvec[i] * Bmx[i] ) and "V" == spamvec
       #NOTE: conjugate() above is because ddenMxdV and dEMxdV will get *elementwise*
       # multiplied (einsum below) by another complex matrix (sgndm or sgnE) and summed

packages/pygsti/algorithms/core.py

@@ -685,7 +685,7 @@ def _spam_penalty_jac_fill(spamPenaltyVecGradToFill, mdl_pre, mdl_post,
     Helper function - jacobian of CPTP penalty (sum of tracenorms of gates)
     Returns a (real) array of shape (_spam_penalty_size(mdl), gaugeGroupEl.num_params()).
     """
-    BMxs = opBasis.get_composite_matrices() #shape [mdl.dim, dmDim, dmDim]
+    BMxs = opBasis.elements #shape [mdl.dim, dmDim, dmDim]
     ddenMxdV = dEMxdV = BMxs.conjugate() # b/c denMx = sum( spamvec[i] * Bmx[i] ) and "V" == spamvec
       #NOTE: conjugate() above is because ddenMxdV and dEMxdV will get *elementwise*
       # multiplied (einsum below) by another complex matrix (sgndm or sgnE) and summed

packages/pygsti/algorithms/gaugeopt.py

@@ -15,11 +15,10 @@
 from .protectedarray import ProtectedArray
 from .circuitparser import CircuitParser
 from .verbosityprinter import VerbosityPrinter
-from .label import Label
+from .label import Label, CircuitLabel
 
-from .basis import Basis
+from .basis import Basis,BuiltinBasis,ExplicitBasis,TensorProdBasis,DirectSumBasis
 from .parameterized import parameterized
-from .dim import Dim
 from .smartcache import SmartCache, CustomDigestError, smart_cached
 
 #Imported in tools instead, since this makes more logical sense