The data distribution for the inputs to the TFTis normally normalized, however it could be useful to look at the data distribution.
Initially, the target is the magnetic field, which has a similar distribution to the current, which has significantly more data points at flat bottom and at flat top, which is inherent to the signal itself. Maybe we can improve the distribution a bit…
The target field residual with the main field component removed significantly improves the data distribution input to the model.