Clean up how interface is handled in QNode and qml.execute
#6225
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Hello. You may have forgotten to update the changelog!
|
albi3ro
reviewed
Sep 6, 2024
astralcai
changed the title
numpy as autogradinterface is handled in QNode and qml.execute
albi3ro
reviewed
Sep 10, 2024
albi3ro
reviewed
Sep 10, 2024
Co-authored-by: Christina Lee <[email protected]>
astralcai
commented
Sep 12, 2024
astralcai
commented
Sep 12, 2024
astralcai
commented
Sep 12, 2024
albi3ro
reviewed
Sep 13, 2024
albi3ro
reviewed
Sep 13, 2024
albi3ro
approved these changes
Sep 13, 2024
PietropaoloFrisoni
approved these changes
Sep 16, 2024
mudit2812
pushed a commit
that referenced
this pull request
Sep 16, 2024
Regarding `numpy` and `autograd`: - When the parameters are of the `numpy` interface, internally treat it as `interface=None`. - Does not change the behaviour of treating user specified `interface="numpy"` as using autograd. Regarding interfaces in general: - The set of canonical interface names in `INTERFACE_MAP` is expanded to include more specific names such as `jax-jit`, and `tf-autograph`. `_convert_to_interfaces` in `qnode.py` uses a separate `interface_conversion_map` to further map the specific interfaces to their corresponding general interface names that can be passed to the `like` argument of `qml.math.asarray` (e.g. "tf" to "tensorflow", "jax-jit" to "jax"). - In `QNode` and `qml.execute`, every time we get an interface from user input or `qml.math.get_interface`, we map it to a canonical interface name using `INTERFACE_MAP`. Aside from these two scenarios, we assume that the interface name is one of the canonical interface names everywhere else. `QNode.interface` is now assumed to be one of the canonical interface names. - User input of `interface=None` gets mapped to `numpy` immediately. Internally, `QNode.interface` will never be `None`. It'll be `numpy` for having no interface. - If `qml.math.get_interface` returns `numpy`, we do not map it to anything. We keep `numpy`. Collateral bug fix included as well: - Fixes a bug where a circuit of the `autograd` interfaces sometimes returns results that are not `autograd`. - Adds `compute_sparse_matrix` to `Hermitian` [sc-73144] --------- Co-authored-by: Christina Lee <[email protected]>
mudit2812
pushed a commit
that referenced
this pull request
Sep 16, 2024
Regarding `numpy` and `autograd`: - When the parameters are of the `numpy` interface, internally treat it as `interface=None`. - Does not change the behaviour of treating user specified `interface="numpy"` as using autograd. Regarding interfaces in general: - The set of canonical interface names in `INTERFACE_MAP` is expanded to include more specific names such as `jax-jit`, and `tf-autograph`. `_convert_to_interfaces` in `qnode.py` uses a separate `interface_conversion_map` to further map the specific interfaces to their corresponding general interface names that can be passed to the `like` argument of `qml.math.asarray` (e.g. "tf" to "tensorflow", "jax-jit" to "jax"). - In `QNode` and `qml.execute`, every time we get an interface from user input or `qml.math.get_interface`, we map it to a canonical interface name using `INTERFACE_MAP`. Aside from these two scenarios, we assume that the interface name is one of the canonical interface names everywhere else. `QNode.interface` is now assumed to be one of the canonical interface names. - User input of `interface=None` gets mapped to `numpy` immediately. Internally, `QNode.interface` will never be `None`. It'll be `numpy` for having no interface. - If `qml.math.get_interface` returns `numpy`, we do not map it to anything. We keep `numpy`. Collateral bug fix included as well: - Fixes a bug where a circuit of the `autograd` interfaces sometimes returns results that are not `autograd`. - Adds `compute_sparse_matrix` to `Hermitian` [sc-73144] --------- Co-authored-by: Christina Lee <[email protected]>
mudit2812
pushed a commit
that referenced
this pull request
Sep 18, 2024
Regarding `numpy` and `autograd`: - When the parameters are of the `numpy` interface, internally treat it as `interface=None`. - Does not change the behaviour of treating user specified `interface="numpy"` as using autograd. Regarding interfaces in general: - The set of canonical interface names in `INTERFACE_MAP` is expanded to include more specific names such as `jax-jit`, and `tf-autograph`. `_convert_to_interfaces` in `qnode.py` uses a separate `interface_conversion_map` to further map the specific interfaces to their corresponding general interface names that can be passed to the `like` argument of `qml.math.asarray` (e.g. "tf" to "tensorflow", "jax-jit" to "jax"). - In `QNode` and `qml.execute`, every time we get an interface from user input or `qml.math.get_interface`, we map it to a canonical interface name using `INTERFACE_MAP`. Aside from these two scenarios, we assume that the interface name is one of the canonical interface names everywhere else. `QNode.interface` is now assumed to be one of the canonical interface names. - User input of `interface=None` gets mapped to `numpy` immediately. Internally, `QNode.interface` will never be `None`. It'll be `numpy` for having no interface. - If `qml.math.get_interface` returns `numpy`, we do not map it to anything. We keep `numpy`. Collateral bug fix included as well: - Fixes a bug where a circuit of the `autograd` interfaces sometimes returns results that are not `autograd`. - Adds `compute_sparse_matrix` to `Hermitian` [sc-73144] --------- Co-authored-by: Christina Lee <[email protected]>
mudit2812
pushed a commit
that referenced
this pull request
Sep 23, 2024
Regarding `numpy` and `autograd`: - When the parameters are of the `numpy` interface, internally treat it as `interface=None`. - Does not change the behaviour of treating user specified `interface="numpy"` as using autograd. Regarding interfaces in general: - The set of canonical interface names in `INTERFACE_MAP` is expanded to include more specific names such as `jax-jit`, and `tf-autograph`. `_convert_to_interfaces` in `qnode.py` uses a separate `interface_conversion_map` to further map the specific interfaces to their corresponding general interface names that can be passed to the `like` argument of `qml.math.asarray` (e.g. "tf" to "tensorflow", "jax-jit" to "jax"). - In `QNode` and `qml.execute`, every time we get an interface from user input or `qml.math.get_interface`, we map it to a canonical interface name using `INTERFACE_MAP`. Aside from these two scenarios, we assume that the interface name is one of the canonical interface names everywhere else. `QNode.interface` is now assumed to be one of the canonical interface names. - User input of `interface=None` gets mapped to `numpy` immediately. Internally, `QNode.interface` will never be `None`. It'll be `numpy` for having no interface. - If `qml.math.get_interface` returns `numpy`, we do not map it to anything. We keep `numpy`. Collateral bug fix included as well: - Fixes a bug where a circuit of the `autograd` interfaces sometimes returns results that are not `autograd`. - Adds `compute_sparse_matrix` to `Hermitian` [sc-73144] --------- Co-authored-by: Christina Lee <[email protected]>
PietropaoloFrisoni
added a commit
that referenced
this pull request
Nov 20, 2024
… an instance of `networkx` Graph object (#6600) **Context:** Issue #6585 indicates that we don't currently allow arguments with types defined in libraries not listed in the supported interfaces. In #6225, the convention of treating internally as `interface=None` parameters of the `numpy` interface has been introduced. The idea of this PR is to allow non-trainable `qnode` inputs to "just work" as arguments, without any special consideration or treatment. We want to rely on the default `numpy` interface for libraries we know nothing about. **Description of the Change:** By default, we set `None` as the value of the interface if the latter is not found in the list of supported interfaces. According to the convention introduced in #6225, this should automatically map the interface to `numpy`. **Benefits:** Now the `qnode` accepts arguments with types defined in libraries that are not necessarily in the list of supported interfaces, such as the `Graph` class defined in `networkx`. **Possible Drawbacks:** None that I can think of. **Related GitHub Issues:** #6585 **Related Shortcut Stories:** [sc-78344] --------- Co-authored-by: Christina Lee <[email protected]>
astralcai
pushed a commit
that referenced
this pull request
Nov 20, 2024
… an instance of `networkx` Graph object (#6600) **Context:** Issue #6585 indicates that we don't currently allow arguments with types defined in libraries not listed in the supported interfaces. In #6225, the convention of treating internally as `interface=None` parameters of the `numpy` interface has been introduced. The idea of this PR is to allow non-trainable `qnode` inputs to "just work" as arguments, without any special consideration or treatment. We want to rely on the default `numpy` interface for libraries we know nothing about. **Description of the Change:** By default, we set `None` as the value of the interface if the latter is not found in the list of supported interfaces. According to the convention introduced in #6225, this should automatically map the interface to `numpy`. **Benefits:** Now the `qnode` accepts arguments with types defined in libraries that are not necessarily in the list of supported interfaces, such as the `Graph` class defined in `networkx`. **Possible Drawbacks:** None that I can think of. **Related GitHub Issues:** #6585 **Related Shortcut Stories:** [sc-78344] --------- Co-authored-by: Christina Lee <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Regarding
numpyandautograd:numpyinterface, internally treat it asinterface=None.interface="numpy"as using autograd.Regarding interfaces in general:
INTERFACE_MAPis expanded to include more specific names such asjax-jit, andtf-autograph._convert_to_interfacesinqnode.pyuses a separateinterface_conversion_mapto further map the specific interfaces to their corresponding general interface names that can be passed to thelikeargument ofqml.math.asarray(e.g. "tf" to "tensorflow", "jax-jit" to "jax").QNodeandqml.execute, every time we get an interface from user input orqml.math.get_interface, we map it to a canonical interface name usingINTERFACE_MAP. Aside from these two scenarios, we assume that the interface name is one of the canonical interface names everywhere else.QNode.interfaceis now assumed to be one of the canonical interface names.interface=Nonegets mapped tonumpyimmediately. Internally,QNode.interfacewill never beNone. It'll benumpyfor having no interface.qml.math.get_interfacereturnsnumpy, we do not map it to anything. We keepnumpy.Collateral bug fix included as well:
autogradinterfaces sometimes returns results that are notautograd.compute_sparse_matrixtoHermitian[sc-73144]