Re: medical data
by krykaczewski - Monday, May 18, 2015, 13:47:32

Dear Konrad,

There was no meaning to use the same time interval to count statistics on it, as these features arrived in different times. For example, ecg changed instantaneously, while rr was changing very slowly (like once per minute). Therefore, we used whole 1.8s interval for ECG and for other qualities we used time delay:

    hr/br = 40s
    rr = 50s
    temp = 30s

In this way, for temperature this interval looked like [t - 1.8s - 30s, t]. Otherwise, training data would be almost constant in wide periods.

I hope this will clarify a little bit.