How to use Gramian Angular fields to improve performance on high frequency datasets
Traditional feature engineering methods like time aware aggregation and spectrograms can have limitations. Spectrograms cannot capture correlation between each segment of the signal with other segments of the signal. If we try to do this with tabular aggregates it becomes a high dimensionality problem.
Gramian Angular Field images of signal data can solve the above problem using a matrix which can be used with computer vision models easily without the limitations of dimensionality.
Prerequisites: PYTS library