Skip to content

Sampler behavior changes in 1.0.0

Jump to bottom
Radomir Stevanovic edited this page Sep 25, 2020 · 1 revision

Between versions 0.9.x and 1.x.x, the samplers in dwave.system.samplers.* slightly modified their solver selection behavior.

The changes are introduced in the specialized samplers PR and supported by instance-level Client config defaults (released as part of dwave-cloud-client==0.8.0).

DWaveSampler / DWaveCliqueSampler

  • client set to "qpu", can be overridden by explicit keyword argument.

  • solver preference set as client-instance default (solver hint). Can be overridden by config file, environment variable, explicit keyword argument.

Local install

  1. In a clean environment (default, first time install):
sampler = DWaveSampler()

selects a QPU with the highest number of qubits, i.e. an Advantage QPU.

  1. If the user already has a config file or environment variable selecting a QPU as the default solver, the default instantiation:
sampler = DWaveSampler()

will select exactly that QPU the user externally set.

  1. If the user already has a config file or environment variable selecting a hybrid or software solver as the default solver, the default instantiation:
sampler = DWaveSampler()

will fail with:

SolverNotFoundError: Solver with the requested features not available

That's because DWaveSampler tries to use a QPU client for an externally/user-set non-QPU solver. dwave.cloud.qpu.Client filters out non-QPU solvers.

There are two main resolutions in this scenario:

i) Modify user preference for solver by modifying solver in config file or the DWAVE_API_SOLVER environment variable. When running a program from the command-line, a simple override is possible with: DWAVE_API_SOLVER= python program.py. That sets the process environment variable DWAVE_API_SOLVER to an empty string.

ii) Explicitly set solver when calling DWaveSampler. For example:

sampler = DWaveSampler(solver=dict(qpu=True))
# or
sampler = DWaveSampler(solver='<solver-name>')
  1. When solver is specified explicitly in the constructor (as some code examples and docs snippets do):
sampler = DWaveSampler(solver=dict(qpu=True))

the solver hint to use an Advantage system is overridden, and the behavior is identical to the one from 0.9.x. For the code above, a QPU solver with the least average load is selected (D-Wave 2000Q or Advantage).

LeapIDE

New users

Users registered after 2020-09-23 will not have DWAVE_API_SOLVER environment variable set in their profile. With Ocean SDK 3.0.0 pre-installed since 2020-09-25, the behavior falls under (1) or (4) above.

Existing users

Existing users have their DWAVE_API_SOLVER (re-)set to '{"qpu": true}', resulting in a behavior described in (2) above.

Hybrid Samplers

Minor changes.

TODO: give more details.

Clone this wiki locally