Additional data
Figure: maximum (cross) and minimum (circles) deviations for x1 and x2 with median profiles (solid and dashed lines, respectively).
We have performed optimization with up to ten data sets with 5 features and 300 samples. Each DATA scatter was drawn from a standard normal distribution Ν(0,1). MODEL function generates data for each data set (again, 5 features and 300 samples) by varying two parameters of the standard normal distribution, i.e. N(x1,x2). We fix all other configurations of the optimization. The results are shown in the figure above.
Since we know the true values of x1 and x2 (0.0, and 1.0, respectively), we can assess the optimization performance as a function of number of data sets by calculating max and min deviations from the true values across the data sets in a series of similar optimization runs (here, 6 runs per case). One can see from the figure that the median profile for the maximum deviation increases with the number of data sets.
Note such increase in the estimate deviations is for the given optimization configuration. For example, we have verified that if we allow for larger number of iterations, we will not observe the increase (data not shown) as a function of number of data sets, since the optimization for any number of data sets between 1 and 10 is equally good given the longer search time. To observe the increase one would need to increase the number of data sets even more. Thus, it is purely a characteristic of the optimization routine.
Some morphological clones obtained during Espoo maple tree studies (see Potapov et al., GigaScience)
Figure: The Espoo maple is in the middle, the best-fit SSMs are named by date, three clones per best-fit SSM.