Adding error models to program breaks measure_observables

measure_observables created new programs to calibrate readout errors (and can also symmetrize the errors). Unfortunately, if the program given to measure_observables contains error model information, that will not be captures by the calibration, and the results will not match the expected behavior from calibration.

[ecpeterson]

As a short term fix, you can replace _get_qvm_compiler_based_on_endpoint with QVMCompiler (as imported from pyquil.api._compiler). That method used to switch between QVMCompiler and LocalQVMCompiler as needed, but LocalQVMCompiler has since been removed.

[marcusps]

Following @ecp-rigetti 's comment, it seems like

from pyquil.api._compiler import QVMCompiler
from pyquil.api._base_connection import ForestConnection
from pyquil.api._qvm import QVM
from pyquil.noise import decoherence_noise_with_asymmetric_ro

from pyquil.device import NxDevice, gates_in_isa

import networkx as nx

nineq_square = nx.convert_node_labels_to_integers(nx.grid_2d_graph(3, 3))
nineq_device = NxDevice(topology=nineq_square)

name = '9q-generic-asym-noisy-readout-qvm'
connection = ForestConnection()
noisy_readout_qvm = QVM(connection=connection, 
                        noise_model=decoherence_noise_with_asymmetric_ro(gates=gates_in_isa(qc.get_isa()),
                                                                         p00=0.99,
                                                                         p11=0.85))
qvmn = QuantumComputer(name=name,
                       qam=noisy_readout_qvm,
                       device=nineq_device,
                       compiler=QVMCompiler(device=nineq_device,
                                            endpoint=connection.compiler_endpoint))

works – i.e., provides a QVM with an error model (even if the program does not introduce errors)