The article describing the details of the 2nd AnDi Challenge specifies that values of K and α are randomly drawn from Gaussians with means α ∈ (0,2) and K ∈ [10^-12 10^6]. However, no ranges are specified for the standard deviations of these Gaussians, or for any model-specific parameters. Being able to generate data reflective of the challenge dataset will be very useful for network training, but this isn't possible without knowing what values each parameter can take. Is this information available anywhere else online? I assume paths will be generated using randomly sampled parameters; if so, will the code you use to do this be made available?
Posted by: SolomonAsghar @ Feb. 5, 2024, 5:56 p.m.Thanks for your inquiry. In the manuscript, we specify that α ∈ (0,2) and K ∈ [10^-12, 10^6] are the supports of the distributions. We do not provide specific ranges for the averages or the standard deviations of the distributions. However, the values of these parameters must be such to allow their calculation from distributions of values bound in the given supports.
Exemplary values of the parameters are given in Table 2 of the manuscript and in the Submission tutorial, section 3: https://github.com/AnDiChallenge/andi_datasets/blob/master/source_nbs/tutorials/challenge_two_submission.ipynb
Ah okay thanks for the clarification. I mistook the line "[K1, K2, . . . , Kn]: average values of the (Gaussian) distribution of the generalized diffusion coefficient for each of the n diffusive states considered in a given experiment, with support [10^−12, 10^6]" to be specifying the support of the average values as opposed to the support of the distributions.
What about the model specific parameters (e.g. transition matrices M, trap radii r_t, number of traps N_t, etc.)? Is any further information about those available, or do we just generate trajectories using values similar to the exemplary values?
Posted by: SolomonAsghar @ Feb. 6, 2024, 9:35 a.m.Besides the constraints and limitations imposed by the simulations (FOV, trajectory length, ...), we would like to assess the methods' performance on a wide range of parameters, compatible with biological experiments.
Posted by: carlo.manzo @ Feb. 6, 2024, 9:48 a.m.