concept jerk magnitude in category machine learning

This is an excerpt from Manning's book Machine Learning with TensorFlow, Second Edition.
To cluster the time-series data, you’ll need to summarize the list of accelerometer vectors into a concise feature vector. One way is to generate a histogram of differences between consecutive magnitudes of the acceleration. The derivative of acceleration is called jerk, and you can apply the same operation to obtain a histogram outlining differences in jerk magnitudes.
In this chapter instead of creating feature vectors by turning sound into numbers using a Fast Fourier Transformation, you already have numbers this time and you simply need to convert those numbers – positions in (X, Y, Z) space - to jerk magnitudes. You can leverage Pandas and its nifty dataframe application programing interface (API) to read the participant CSV files into a big table structure in memory, and then you can get NumPy arrays out of that structure. Using NumPy arrays as I’ve shown you in prior chapters just makes dealing with TensorFlow and machine learning that much easier since you can then just combine the 22 participants and their positional data into a TensorFlow-compatible NumPy matrix. The workflow can be generalized to the following form shown visually in figure 8.3.