allow user to specify the minimum sampling variance as a percentage of the prior variance

[carolyner]

The code in imabc.R, copied below, checks that the sampling variance does not drop below a minimum value, currently hard coded to be 10% of the prior variance (i.e. 0.01 in lines 1031, 1035, and 1037. Revising this to be an option (e.g., 0=no minimum up to 1, which would not allow any narrowing) will increase user flexibility and improve documentation. Suggest a default of 10%.

      # Deal with parameters that don't have any variance
      if (any(diag(sample_cov) <= 0.1*prior_sds^2)) {
        # This occurs when adding a new parameter: it is set to default for all prior draws
        # it may also occur when sampling becomes over concentrated around center points
        message("Potentially over-concentrated sampling: Parameters with sampling variance < 10% of prior variance")
        is_small <- (diag(sample_cov)<= 0.1*prior_sds^2)
        sample_sd <- sqrt(diag(sample_cov))
        sd_next <- sqrt(0.1)*prior_sds


        # limit variance to become no more than 10% of prior variance while retaining correlation structure
        sample_sd[is_small] <- sd_next[is_small]
        sample_cov <- increase_samplevar(var_data,sample_sd)
        }

Note that this code also refers to a new function, increase_samplevar() which may need some clean-up.

Finally, please improve the message above, to provide information about the triggering parameter and the corresponding variance (3 sig digits).