I'm using pi-pulse pulses on qubit Armonk for determining 0 and 1 by Machine Learning. But when I run the code from https://qiskit.org/textbook/ch-quantum-hardware/calibrating-qubits-pulse.html. I get an incorrect distribution of states $|0\rangle$ and $|1\rangle$.
For example, in the Qiskit textbook, this image looks like:
About 3 months ago, the image that I got by using this textbook was correct, but now it's not. When I use Machine Learning for old states, I get excellent results, but now my algorithms have failed on this task.
Please, help me with my problem. I don't know, how to fix it. Why did this happen? That is the problem?