I have 165000 observations. Each observation is 180 component vector. One component in this vector is an integer in range from 0 to 200.
Every vector is constructed from sensor readings taken at 180 consecutive times. Readings may fail at some unpredictable times, thus some vector components may be undefined. 40% - 60% of random vector components may be undefined in every vector. I am trying to find an approximation function that will allow me to predict sensor reading at any time. There may be different correlations between some of these vectors. In fact these sensors read traffic speed on different streets of a big city during a period of a month. What other methods for prediction with incomplete data can be used for the task I described? Thanks! >From Ted Dunning <ted.dunn...@gmail.com> >Subject Re: Function approximation in Mahout? >Date Wed, 21 Apr 2010 14:34:58 GMT > >Mahout does not have a lot of regression capabilities at this time, other >than various forms of binomial regression (SVM, logistic regression, >decision forests) but other forms of regression are relatively lacking. > >Commons math has some capabilities, but not in a particularly scalable form. > >What size is your problem? > >On Tue, Apr 20, 2010 at 2:07 PM, Dmitri O.Kondratiev <doko...@gmail.com >wrote: > >> Hello, >> Does Mahout support any function approximation frameworks, such as greedy >> function approximation with gradient boosting (TreeNet)? >> http://en.wikipedia.org/wiki/TreeNet#Names >> >> Thanks! >>
