ICSS-master.zip

# ICSS Matlab implementation of the [ICSS algorithm of Inclan and Tiao](http://www.jstor.org/stable/2290916) ("Use of cumulative sums of squares for retrospective detection of changes of variance"). ## How does it work? Load to the `matlab` directory and run `demo`, or `demo_accelermeter_data` for an application of x-axis value of a recorded activity series with a smartphone. The result, for the paper provided data is:  For accelerometer data:  For any vector of values, run `ICSS(data)` to obtain the change points. ## Available datasets There are a couple of predefined datasets availble. These can be generated using `data = ProvideDataBatch(size, type)`. The types are: * alternating: Generate alternating variances of `1` and `5` with mean `0` * paper: use the dataset as defined in the paper (changepoints at `391` and `518`, with variances `1`, `0.365` and `1.033`) * homogeneous: homogeneous dataset with mean 0 and variance 1 * single: create a single breakpoint at half or the data. Variance goes there from `1` to `2`.