A significant update, which primarily updates the package’s API to become more consistent, uniform, and following PEP8 style guidance. This unfortunately breaks backward compatibility with previous releases, but brings the package substantially closer to the future version 1.0 when the API will be stabilized. Most of the changes needed to update existing scripts should be straightforward, as names themselves are similar (though their location within the package may have changed) and now conform, for the most part, to the PEP8 style. Updated documentation and tutorials can also provide help to users seeking to update code that uses version 0.9.9.3 and lower. Note that pickle files are not compatible between any prior version of pyGSTi and this one (though most dataset files are). In addition to the many name and sub-package location updates, additional functionality has been added.
Added
- Adds a new interface for external circuit simulators. This should allow pyGSTi to provide access to more powerful circuit simulation algorithms in the future. In this release, the interface has been used to incorporate the CHP Clifford-circuit simulator as an option. This update was made possible through an extensive refactoring and development “evolution types” in pyGSTi, which are also simpler to use now.
- Adds support for distributed and shared memory.. This more efficiently splits many computations among multiple processors, avoiding the need to gather memory to a single processor and thereby reducing the memory requirement by a factor of up to the processor count. This update also allows pyGSTi to take advantage of the fact that multiple cores on a single machine (node) can share memory. The new implementation automatically detects when processors can share memory and utilizes this feature by default, avoiding copying in favor of access to shared memory.
- Adds ability to serialize (read from & write to disk) most of pyGSTi’s object types using a robust and JSON-compatible format. Most objects now possess read and write methods for this purpose. This, by design, largely obviates the need to store data in Python pickle files, which are both fragile and pose security risks.
- Adds a new flexible type of data set (FreeformDataset) and simulator (FreeformSimulator) objects that allow arbitrary data to be computed for and associated with circuits.
- Adds a means of directly running experiments on IBMQ hardware when provided with proper credentials (see the pygsti.extras.ibmq package).
- Adds initial support interpolating the results of computationally intensive physics simulations to create and optimize “physical models”. This functionality exists in pygsti.extras.interpygate.
- Add the ability to place simple bounds on a model’s parameters when it is optimized. This is particularly relevant for restricting the aforementioned “physical models” to the region where their interpolation is valid.
- Adds additional support for fitting data to time-dependent models (time dependent gate set tomography).
Fixed
- Fixes many bugs and addresses or resolves many issues.
Changed
- Enhancements to model creation routines. ProcessorSpec objects now play a central role, as they define the API of a quantum processor. Processor specification objects are used to construct models and experiment designs. Model construction routines have been revamped, and allow noise to be more easily added to models. The construction of different model types (explicit, local, or cloud-noise) is much more unified by construction routines having similar argument lists and behaviors. The use of “implied idle” gates in models and circuits cleaner and simpler to understand.
- Model parameterization is more transparent. Known how a model is parameterized is easier thanks to the ability to label parameters and print them out more straightforwardly. More sophisticated parameterization is possible by “collecting” parameters or by constructing models from the ground up. The parameter allocation framework is updated to be more robust. (For example model members don’t need to be added to models before their number of parameters can be queried.) When testing models, the number of parameters of the model can be specified manually, using the new num_modeltest_params attribute.
- Improves randomized and volumetric benchmarks. Increased incorporation of Pandas dataframes makes analysis of data from multiple benchmarking experiments possible.
- Improves and expands the interface with Google’s Cirq package. The new DataSet.add_cirq_trial_result method can be used to import Cirq.TrialResult objects, and supports using cirq.WaitGate within the conversion method Circuit.convert_to_cirq.
- Updates wildcard error computations, adding options and more robust optimization methods.
- The package’s deployment process has updated from providing only a source distribution (requiring users to compile the C-extensions on their own if they wanted to use them) to building wheels for several common platforms and python versions. This update should save many users time and effort installing pyGSTi.