The first preprocessing step often consists in restricting the signal to a frequency range of interest through filtering. MNE-Python supports band-pass, low-pass, high-pass, band-stop, and notch filtering. Instances of Raw can be filtered using the filter method that supports fast Fourier transform (FFT) based finite impulse response (FIR) filters (optionally using the overlap-add technique to minimize computation time), as well as infinite impulse response (IIR) filters such as Butterworth filters implemented in SciPy. Several channels can be filtered in parallel, thanks to the standard multiprocessing Python module exposed via the Joblib package (http://pythonhosted.org/joblib/). The FFTs used to implement FIR filters can also be efficiently computed on the graphical processing unit (GPU) via CUDA and PyCUDA (Klöckner et al., 2012), further reducing the execution time.
