Prioritize transformers for implementation or interfacing about sktime HOT 7 CLOSED

1. AutoCorrelationFunction: Java version is here
   https://github.com/goastler/uea-tsc/blob/master/src/main/java/timeseriesweka/filters/ACF.java
   public double[] fitAutoCorrelations(double[] data)
   is the actualy implementation, although happy to use an off the shelf one
   single parameter is maximum lag, which I think I default to max of 100 or seriesLength/4
   Needed for classifier RISE

from sktime.

2. FastFourierTransform: FFT
   https://github.com/goastler/uea-tsc/blob/master/src/main/java/timeseriesweka/filters/FFT.java
   set to either do an FFT or DFT
   Classifier: RISE and BOSS

from sktime.

3. PowerSpectrum:
   https://github.com/goastler/uea-tsc/blob/master/src/main/java/timeseriesweka/filters/PowerSpectrum.java
   public static double[] powerSpectrum(double[] d)

basically take the DFT and square the terms. Issues here are whether to use FFT and padding zeros (O(mlogm, but padding may skew things, especially if the data is not normalised) or use the exact DFT (O(m^2)). Only keep have of the terms, second half of power spectrum is a duplicate of the first
Needed for classifier RISE

from sktime.

4. SFA
   all of the boss transforms we have are encapsulated within boss. We should probably change this
   https://github.com/goastler/uea-tsc/blob/master/src/main/java/timeseriesweka/classifiers/BOSS.java
   however, the previously sourced boss implementation can be used here
   Needed for classifier BOSS

from sktime.

4. Piecewise Aggregate Approximation (PAA)
   https://github.com/goastler/uea-tsc/blob/master/src/main/java/timeseriesweka/filters/PAA.java
   Simple transform that forms the basis of many others.

from sktime.

5. SAX
   composed of PAA then a simple discretisation transform

from sktime.

For available Python implementation, see #6

from sktime.