I work on comparing QSVM and Classic SVM (SKlearnSVM) with using Qiskit. I have to show quantum supremacy at 400000-500000 samples but I don't get good results. I have problem with long time training and using big RAM when I use big training samples (In both cases). If I use big training samples in quantum algorithm, I can't get results of tests in short time (such as 1-4 days is acceptable). For example, I try do a test with 1000000 samples when I train quantum kernel on 20 samples and I got a bug in running state for some work (It was already above 10000 works; The work hanged). The reason of bug is unknown. For tests I use 2 quants (maybe it is little?) and last version libraries. Dataset is my generate (with perceptible differents). I try normalize dataset but I got a code mistake of numpy library (can't reshape array ant e.t.c). For begining I used example from github tutorial, next I used other methods of class (that I find in qiskit - .train(),.test()) but I don't see differents here (if I use .run() and .predict() with rebooting of kernel).
And I have question - So can I to show this comparing generally with using Qiskit? So how? Maybe can I show quantum supremacy on other size of samples? Or can I use Qiskit library wrong?