Plot training curve of VQC and MDM with SPSA

In this example, we will show how to plot the training curve of quantum neural network VQC and MDM, using an SPSA optimizer.

# Author: Gregoire Cattan
# License: BSD (3-clause)

import matplotlib.pyplot as plt
from pyriemann.estimation import Shrinkage

from pyriemann_qiskit.classification import QuanticVQC
from pyriemann_qiskit.pipelines import QuantumMDMWithRiemannianPipeline
from pyriemann_qiskit.utils.dataset import (
    generate_linearly_separable_dataset,
    get_mne_sample,
)
from pyriemann_qiskit.utils.hyper_params_factory import get_spsa

print(__doc__)

Setup

# Define the plot area
fig, axes = plt.subplots(1, 2)
fig.suptitle("Training curves")

# Generate vectors for VQC
Xv, yv = generate_linearly_separable_dataset(n_samples=20)

# ... and matrices for MDM
Xm, ym = get_mne_sample(n_trials=100)

Instantiate the pipelines

# Create the SPSA optimizer
optimizer = get_spsa(max_trials=100)

# Instantiate VQC
vqc = QuanticVQC(optimizer=optimizer)

# ... and the quantum MDM
# This used QAOA under the hood, which is a quantum parametric circuit
# analog to neural network
mdm = QuantumMDMWithRiemannianPipeline(
    metric={"mean": "qlogeuclid", "distance": "logeuclid"},
    quantum=True,
    regularization=Shrinkage(shrinkage=0.9),
    shots=1024,
    seed=696288,
    qaoa_optimizer=optimizer,
)

[QClass]  Initializing Quantum Classifier
[QuantumMDMWithRiemannianPipeline]  Running QMDM with metric {'mean': 'qlogeuclid', 'distance': 'logeuclid'}
[QClass]  Initializing Quantum Classifier

Fit and plot learning curve for VQC

vqc.fit(Xv, yv)
evaluated_values = vqc.evaluated_values_

# Note: The optimizer converge:
# VQC+SPSA is appropriate for the toy dataset.
axe = axes[0]
axe.plot(evaluated_values)
axe.set_ylabel("Evaluated values (VQC)")
axe.set_xlabel("Evaluations")

[QClass]  Quantum simulation will be performed
GPU optimization disabled. No device found.
[QClass]  Fitting:  (20, 2)
[QClass]  Feature dimension =  2
[QClass]  Quantum backend =  AerSimulator('aer_simulator_statevector')
[QClass]  seed =  319794
[QClass]  VQC training...
[QClass]  Training...

Text(0.5, 23.633333333333333, 'Evaluations')

Fit and plot learning curve for MDM with QAOA

mdm.fit(Xm, ym)
evaluated_values = mdm._pipe[2]._optimizer.evaluated_values_

# Note: The optimizer doesn't converge:
# QAOA+SPSA is not appropriate for the MNE dataset.
axe = axes[1]
axe.plot(evaluated_values)
axe.set_ylabel("Evaluated values (MDM)")
axe.set_xlabel("Evaluations")

plt.show()

[QClass]  Quantum simulation will be performed
GPU optimization disabled. No device found.
[QClass]  Fitting:  (100, 2, 2)
[QClass]  Feature dimension =  2
[QClass]  Quantum backend =  AerSimulator('aer_simulator_statevector')
[QClass]  seed =  696288
[QClass]  Quantic MDM initiating algorithm
[QClass]  Using NaiveQAOAOptimizer
[QClass]  Training...