Merge branch 'master' into evolution_operator

doc/introduction/data.rst

@@ -14,6 +14,13 @@ where the quantum dataset is a collection of `quantum data` obtained from variou
     The packages ``zstd`` and ``dill`` are required to use the :mod:`~pennylane.data` module. 
     These can be installed with ``pip install zstd dill``.
 
+.. warning::
+
+        PennyLane datasets use the ``dill`` module to compress, store, and read data. Since ``dill``
+        is built on the ``pickle`` module, we reproduce an important warning from the ``pickle``
+        module: it is possible to construct malicious pickle data which will execute arbitrary code
+        during unpickling. Never unpickle data that could have come from an untrusted source, or
+        that could have been tampered with.
 
 Loading Datasets in PennyLane
 -----------------------------

doc/releases/changelog-dev.md

@@ -327,6 +327,9 @@
   Replaces `qml.transforms.make_tape` with `make_qscript`.
   [(#3429)](https://github.com/PennyLaneAI/pennylane/pull/3429)
 
+* Extended the functionality of `qml.matrix` to qutrits.
+  [(#3460)](https://github.com/PennyLaneAI/pennylane/pull/3460)
+
 * File `qcut.py` in `qml.transforms` reorganized into multiple files in `qml.transforms.qcut`
   [3413](https://github.com/PennyLaneAI/pennylane/pull/3413)
 
@@ -356,7 +359,7 @@
   [(#3399)](https://github.com/PennyLaneAI/pennylane/pull/3399)
 
 * Improved the performance of executing circuits under the `jax.vmap` transformation, which can now leverage the batch-execution capabilities of some devices. [(#3452)](https://github.com/PennyLaneAI/pennylane/pull/3452)
-  
+
 <h4>Return types project</h4>
 
 * The autograd interface for the new return types now supports devices with shot vectors.
@@ -644,6 +647,9 @@ Deprecations cycles are tracked at [doc/developement/deprecations.rst](https://d
   $U^{\dagger} | \psi \rangle$ the actual state being measured in the $Z$-basis.
   [(#3409)](https://github.com/PennyLaneAI/pennylane/pull/3409)
 
+* Adds warnings about using ``dill`` to pickle and unpickle datasets. 
+  [#3505](https://github.com/PennyLaneAI/pennylane/pull/3505)
+
 <h3>Bug fixes</h3>
 
 * Fixed a bug where `hamiltonian_expand` didn't preserve the type of the inputted results in its output.
@@ -687,6 +693,7 @@ Deprecations cycles are tracked at [doc/developement/deprecations.rst](https://d
 
 This release contains contributions from (in alphabetical order):
 
+Guillermo Alonso
 Juan Miguel Arrazola
 Utkarsh Azad
 Samuel Banning

pennylane/data/data_manager.py

@@ -212,7 +212,9 @@ def _generate_folders(node, folders):
 def load(
     data_name, attributes=None, lazy=False, folder_path="", force=False, num_threads=50, **params
 ):
-    r"""Downloads the data if it is not already present in the directory and return it to user as a :class:`~pennylane.data.Dataset` object.
+    r"""Downloads the data if it is not already present in the directory and return it to user as a
+    :class:`~pennylane.data.Dataset` object. For the full list of available datasets, please see the
+    `datasets website <https://pennylane.ai/qml/datasets.html>`_.
 
     Args:
         data_name (str)   : A string representing the type of data required such as `qchem`, `qpsin`, etc.
@@ -226,6 +228,14 @@ def load(
 
     Returns:
         list[:class:`~pennylane.data.Dataset`]
+
+    .. warning::
+
+        PennyLane datasets use the ``dill`` module to compress, store, and read data. Since ``dill``
+        is built on the ``pickle`` module, we reproduce an important warning from the ``pickle``
+        module: it is possible to construct malicious pickle data which will execute arbitrary code
+        during unpickling. Never unpickle data that could have come from an untrusted source, or
+        that could have been tampered with.
     """
 
     _ = lazy

pennylane/ops/functions/matrix.py

@@ -138,9 +138,12 @@ def _matrix(tape, wire_order=None):
     wire_order = wire_order or tape.wires
 
     # initialize the unitary matrix
-    result = qml.math.eye(2 ** len(wire_order), like=interface)
+    if len(tape.operations) == 0:
+        result = qml.math.eye(2 ** len(wire_order), like=interface)
+    else:
+        result = matrix(tape.operations[0], wire_order=wire_order)
 
-    for op in tape.operations:
+    for op in tape.operations[1:]:
         U = matrix(op, wire_order=wire_order)
         # Coerce the matrices U and result and use matrix multiplication. Broadcasted axes
         # are handled correctly automatically by ``matmul`` (See e.g. NumPy documentation)