Hi everyone, we’re excited to announce PennyLane version 0.5.0.
This is a major release, that contains some great new features and improvements. These include:
Better optimization
Adds support for a new optimizer, qml.QNGOptimizer, which optimizes QNodes using quantum natural gradient descent. See [1909.02108] Quantum Natural Gradient for more details.
A new QNode method, QNode.metric_tensor(), has also been added, which returns the block-diagonal approximation to the Fubini-Study metric tensor evaluated on the attached device.
New QNode outputs
QNodes can now return a specified number of samples from a given observable via the top-level pennylane.sample() function. To support this on plugin devices, there is a new Device.sample method.
New operations
Added controlled rotation gates to PennyLane operations and default.qubit plugin.
Faster default.qubit
The default.qubit plugin is now slightly faster, as rotation gates are implemented directly instead of using scipy.linalg.expm.
Better tutorials
Tutorials are now collated into the top-level examples/ folder, and all are executable Python scripts. They are also now automatically built and included in the documentation, check them out here: https://pennylane.readthedocs.io/en/latest/tutorials.html
Links are now also provided at the top of every tutorial page for downloading the tutorial as an executable python script, or downloading the tutorial as a Jupyter notebook.
Breaking changes
Note that there has also been one breaking change for the CV operations. The following CV observables were renamed to comply with the new Operation/Observable scheme:
-
MeanPhotontoNumberOperator -
HomodynetoQuadOperator -
NumberStatetoFockStateProjector
See the full release notes for more details on the latest PennyLane release.
As always, this release would not have been possible without all the help from our contributors:
Simon Cross, @AroosaIjaz, @nathan , @johannesjmeyer , Rohit Midha, Nicolás Quesada, @Maria_Schuld , Antal Száva, @rooler.