Predicting the future of discrete sequences from fractal representations of the past

We propose a novel approach for building finite memory predictive models similar in spirit to variable memory length Markov models (VLMMs). The models are constructed by first transforming the n-block structure of the training sequence into a geometric structure of points in a unit hypercube, such that the longer is the common suffix shared by any two n-blocks, the closer lie their point representations. Such a transformation embodies a Markov assumption-n-blocks with long common suffixes are likely to produce similar continuations. Prediction contexts are found by detecting clusters in the geometric n-block representation of the training sequence via vector quantization. We compare our model with both the classical (fixed order) and variable memory length Markov models on five data sets with different memory and stochastic components. Fixed order Markov models (MMs) fail on three large data sets on which the advantage of allowing variable memory length can be exploited. On these data sets, our predictive models have a superior, or comparable performance to that of VLMMs, yet, their construction is fully automatic, which, is shown to be problematic in the case of VLMMs. On one data set, VLMMs are outperformed by the classical MMs. On this set, our models perform significantly better than MMs. On the remaining data set, classical MMs outperform the variable context length strategies.