Add QuantumKernelTrainer and KernelLoss classes #244
Merged
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
…o accept QKT as input.
caleb-johnson
requested review from
adekusar-drl,
manoelmarques,
pbark,
stefan-woerner and
woodsp-ibm
as code owners
Comment on lines
73
to
77
| def test_qkt(self): | ||
| """Test QuantumKernelTrainer""" | ||
| self.setUp() | ||
| with self.subTest("check default fit"): | ||
| qkt = QuantumKernelTrainer(optimizer=self.optimizer) |
There was a problem hiding this comment.
Could you please add a test where QuantumKernelTrainer is instantiated with all default values?
woodsp-ibm
reviewed
Nov 5, 2021
woodsp-ibm
reviewed
Nov 5, 2021
woodsp-ibm
reviewed
Nov 5, 2021
woodsp-ibm
reviewed
Nov 5, 2021
woodsp-ibm
reviewed
Nov 5, 2021
|
@calebj15 Are there any updates on this PR? |
…to quantum-kernel-trainer
… QuantumKernel.assign_user_parameters more robust and add user tests to test parameter expression assignment. Test QKT outputs in tests.
…to quantum-kernel-trainer
…achine-learning into quantum-kernel-trainer
adekusar-drl
reviewed
Dec 8, 2021
| super().setUp() | ||
| algorithm_globals.random_seed = 10598 | ||
| self.optimizer = COBYLA(maxiter=25) | ||
| self.backend = Aer.get_backend("statevector_simulator") |
There was a problem hiding this comment.
Just noticed, please replace statevector_simulator with aer_simulator_statevector, otherwise there's a deprecation warning.
Suggested change
| self.backend = Aer.get_backend("statevector_simulator") | |
| self.backend = Aer.get_backend("aer_simulator_statevector") |
| # Class fields | ||
| self._quantum_kernel = quantum_kernel | ||
| self._initial_point = initial_point | ||
| self._optimizer = optimizer if optimizer else SPSA() |
There was a problem hiding this comment.
Can be simplified:
Suggested change
| self._optimizer = optimizer if optimizer else SPSA() | |
| self._optimizer = optimizer or SPSA() |
Comment on lines
126
to
128
| self._loss = loss | ||
| else: | ||
| raise ValueError(f"Unknown loss {loss}!") |
There was a problem hiding this comment.
So if I pass a callable Callable[[Sequence[float]], float], I get an exception.
Could you also add a test with a custom loss to check this scenario?
There was a problem hiding this comment.
Yes I believe we decided against this usage, so I just need to update type hint. Thanks
Comment on lines
143
to
146
| return self._loss | ||
|
|
||
| @loss.setter | ||
| def loss(self, loss: KernelLoss) -> None: |
There was a problem hiding this comment.
Setter type should be the same as in the constructor.
Comment on lines
69
to
71
| def test_qkt(self): | ||
| """Test QuantumKernelTrainer""" | ||
| self.setUp() |
There was a problem hiding this comment.
self.setUp() is not required. Please remove.
Comment on lines
84
to
86
|
|
||
| with self.subTest("check fit with params"): | ||
| self.setUp() |
There was a problem hiding this comment.
self.setUp() is not required. Please remove.
|
@calebj15 you have a conflict in |
Comment on lines
1
to
3
| { | ||
| "cells": [ | ||
| { |
There was a problem hiding this comment.
Please rename tutorial to 08. There are already two new tutorials 06 Torch runtime, 07 PegasosQSVC.
Comment on lines
64
to
68
| " x[0]: number of function evaluations\n", | ||
| " x[1]: the parameters\n", | ||
| " x[2]: the function value\n", | ||
| " x[3]: the stepsize\n", | ||
| " x[4]: whether the step was accepted\n", |
There was a problem hiding this comment.
Please reformat as a proper docstring. By the way, starting last week, black should applied on the doc folder as well.
| "source": [ | ||
| "### Prepare the Dataset\n", | ||
| "\n", | ||
| "QKT only provides an advantage over classical counterparts for quantum kernels which are hard to compute classically. Finding quantum kernels which provide an advantage on real world data is a challenge in itself and is outside the scope of this tutorial.\n", |
There was a problem hiding this comment.
Suggested change
| "QKT only provides an advantage over classical counterparts for quantum kernels which are hard to compute classically. Finding quantum kernels which provide an advantage on real world data is a challenge in itself and is outside the scope of this tutorial.\n", | |
| "`QuantumKernelTrainer` only provides an advantage over classical counterparts for quantum kernels which are hard to compute classically. Finding quantum kernels which provide an advantage on real world data is a challenge in itself and is outside the scope of this tutorial.\n", |
| "\n", | ||
| "QKT only provides an advantage over classical counterparts for quantum kernels which are hard to compute classically. Finding quantum kernels which provide an advantage on real world data is a challenge in itself and is outside the scope of this tutorial.\n", | ||
| "\n", | ||
| "In this guide, we will use Qiskit Machine Learning's `ad_hoc.py` dataset located [here](../../qiskit_machine_learning/datasets/). Quantum kernel training does not provide a classification advantage on this dataset, but we will use it to demonstrate the kernel training process." |
There was a problem hiding this comment.
Suggested change
| "In this guide, we will use Qiskit Machine Learning's `ad_hoc.py` dataset located [here](../../qiskit_machine_learning/datasets/). Quantum kernel training does not provide a classification advantage on this dataset, but we will use it to demonstrate the kernel training process." | |
| "In this guide, we will use Qiskit Machine Learning's `ad_hoc` dataset, see the documentation [here](https://qiskit.org/documentation/machine-learning/stubs/qiskit_machine_learning.datasets.ad_hoc_data.html). Quantum kernel training does not provide a classification advantage on this dataset, but we will use it to demonstrate the kernel training process." |
Comment on lines
201
to
202
| "\n", | ||
| "Additionally, we select SPSA as the optimizer and initialize the trainable parameter to 0.1 with `initial_point`.\n", |
There was a problem hiding this comment.
In the code initial_point set as initial_point=[0.0], while in the text it is mentioned 0.1. Please correct.
Comment on lines
230
to
232
| "# To use the quantum-kernel-trainer runtime client, users should replace\n", | ||
| "# QuantumKernelTrainer with the\n", | ||
| "# qiskit_machine_learning.runtime.QuantumKernelTrainerClient class.\n", |
There was a problem hiding this comment.
I'd rather remove this comment about the runtime. Once runtime is out, the tutorial can be extended.
Comment on lines
263
to
268
| "{ 'optimal_parameters': {ParameterVectorElement(θ[0]): -0.0717475136739174},\n", | ||
| " 'optimal_point': array([-0.07174751]),\n", | ||
| " 'optimal_value': 6.825533666231239,\n", | ||
| " 'optimizer_evals': 30,\n", | ||
| " 'optimizer_time': None,\n", | ||
| " 'quantum_kernel': <qiskit_machine_learning.kernels.quantum_kernel.QuantumKernel object at 0x7fdda0701b50>}\n" |
There was a problem hiding this comment.
Could you please add a description of what you get in qka_results?
Comment on lines
325
to
327
| "From the callback data, we can plot how the loss evolves during the training process. We see it converges rapidly and reaches 100% test accuracy on this dataset with our choice of inputs.\n", | ||
| "\n", | ||
| "We can also display the final kernel matrix evaluated on the training data." |
There was a problem hiding this comment.
Could you please a description of what is shown on the kernel matrix plot in various colors?
Comment on lines
92
to
99
| loss: Union[str, Callable[[Sequence[float]], float]] = "svc_loss", | ||
| optimizer: Optional[Optimizer] = None, | ||
| initial_point: Optional[Sequence[float]] = None, | ||
| ): | ||
| """ | ||
| Args: | ||
| quantum_kernel: QuantumKernel to be trained | ||
| loss (Callable[[Sequence[float]], float] or str): |
There was a problem hiding this comment.
Can you please update loss type hints to be the same across the code? I guess it should be Union[str, KernelLoss] now.
| "source": [ | ||
| "### Train the Quantum Kernel\n", | ||
| "\n", | ||
| "To train the quantum kernel on the dataset (samples and labels), we call the `fit_kernel` method of `QuantumKernelTrainer`.\n", |
There was a problem hiding this comment.
Suggested change
| "To train the quantum kernel on the dataset (samples and labels), we call the `fit_kernel` method of `QuantumKernelTrainer`.\n", | |
| "To train the quantum kernel on the dataset (samples and labels), we call the `fit` method of `QuantumKernelTrainer`.\n", |
Comment on lines
118
to
122
| loss = loss.lower() | ||
| if loss == "svc_loss": | ||
| self._loss = self._str_to_loss(loss) | ||
| else: | ||
| raise ValueError(f"Unknown loss {loss}!") |
There was a problem hiding this comment.
I guess a string to loss conversion is done in the _str_to_loss, so these checks are unnecessary as you mentioned below.
|
@stefan-woerner Please review as well as you left a change request. |
stefan-woerner
previously approved these changes
Dec 14, 2021
oscar-wallis
pushed a commit
that referenced
this pull request
Feb 16, 2024
* Add QuantumKernelTrainer class, KernelLoss class, and refactor QSVC to accept QKT as input. * Refactor to unconfuse mypy * Add runtime client for QKA * Remove qka program info from __init__ * Remove vqe language from qka client * Retract changes to QSVC * Address peer review comments. Remove kernels.algorithms package. Make QuantumKernel.assign_user_parameters more robust and add user tests to test parameter expression assignment. Test QKT outputs in tests. * Change SVCAlignment class to SVCLoss and update KernelLoss doc strings. * Fix equation in SVCLoss docstring * Docstring improvements * Add quantum kernel trainer runtime client and unit tests. * Fix spelling errors * Fix spelling errors * Clean up docstring for SVCLoss * Clean up docstring for SVCLoss * Fix spelling errors * Fix equation formatting in docstring * Fix equation formatting in docstring * New spelling errors * New spelling errors * Fix spelling errors. Add 'runtime' to pylint dict * Fix mypy errors * Fix mypy errors * style error * style error * spelling error in docstring * Fix circular imports due to typing. * Fix circular imports due to typing. * Make QuantumKernel class serializable by runtime. * Update quantum_kernel_trainer.py * Update quantum_kernel_trainer.py * Update quantum_kernel_trainer.py * Make KernelLoss.get_variational_callable a base method * Make SVCLoss more efficient by only calculating kernel once * Create kernels.algorithms package for QuantumKernelTrainer. Fix all CI/CD errors * spelling errors * Clean up kernel loss docstring * Create kernels.algorithms package for quantum kernel trainer. Clean up CI/CD errors. Address kernel loss feedback. * spelling errors * spelling errors * spelling errors * Create separate file for kernel loss classes. Fix spelling errors. * add runtime to pylint dict * add vqe to pylint dict * Address peer feedback * html errors * Address peer review comments * Fix mypy errors * Add a quantum kernel training tutorial * Print results of kernel fitting in tutorial * Rename QKT tutorial. Fix bad assertion in qkt test. * Make QKT tests deterministic. Add to pylintdict * Address minor comments in qkt tutorial * Fix spelling errors in QKT tutorial * Clean up docstring for qkt * Fix comment in qkt tutorial. Remove unnecessary type checks from QKT module * Clean up docstring in QKT * Fix mypy error in qkt * Fix mypy errors in qkt * Fix mypy errors in qkt Co-authored-by: Manoel Marques <[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.
Summary
Closes #227
Variational quantum algorithms generally aim to train some quantum circuit parameters using a classical optimizer. To implement this, parameters of a
QuantumCircuitmust be bound and re-bound many times during optimization; however, theQuantumCircuitAPI "forgets" the parameter after it is bound, making consecutive binds of aQuantumCircuitparameter a cumbersome process.The introduction of a
QuantumKernelTrainer (QKT)class simplifies this process by providing afit_kernelmethod, which will return some optimizedQuantumKernel.feature_mapparameter values which can be used to generate an optimized kernel operator.Changes from this PR:
Addition of a
qiskit_machine_learning.kernels.algorithms.QuantumKernelTrainer (QKT)class which facilitates optimizing a set of a parameters within aQuantumKernelfeature map with regard to a given dataset.Addition of a new
qiskit_machine_learning.utils.loss_functionsbase class,KernelLoss.KernelLossfunctions take a set of circuit parameters to optimize, aQuantumKernelobject, and a labeled dataset as input and provide a loss value as output.Addition of a new
KernelLosschild class,SVCLoss.