I’ve been working on a binary classification problem, transitioning from the Iris dataset to one that’s non-linearly separable. Despite experimenting with the EfficientSU2 ansatz, I’m currently achieving only a 70 %classification accuracy. Could you assist in demonstrating how QML might reach ~100% accuracy for this not that hard case (for classical ML)?
Here is my notebook:quantumMind/PennyLane/notebooks/tut4c_binClassifier_non-linear.ipynb at main · balewski/quantumMind · GitHub
See the plot at the bottom, which shows clasiffication boundary
I realized I have not done my homework properly. After I 1 hour of work on manual HPO to adjusted
NesterovMomentumOptimizer(stepsize=0.05, momentum=0.95), increasing number of steps to 200, and added selection of the best train accuracy the resulting validation accuracy is now 91% - good enough for government work.
The code in git is updated and the plot on the bottom looks very believable.
1 Like