LoadH5ADobs Missing Labels

[sylvia-science]

Hi,

I'm not very familiar with Anndata objects so I don't know if this is a common problem, but the one I was working with had missing observable values labeled with -1. Because the LoadH5ADobs tried to convert the column to a factor but is missing a label for these -1 values, the function can throw this error:

Error in factor(x = obs_value_i, labels = hfile_cate[[obs.i]][]) :
invalid 'labels'; length 24 should be 1 or 25

Which says that I have 24 labels for the 25 values with an extra -1 value.

My simple fix for this was checking if the number of unique values obs_value_i matches the length of the labels. Obviously this will fail if there is more than one missing value or if it's not -1.

if ('__categories' %in% names(x = hfile_obs)) {
  hfile_cate <- hfile_obs[['__categories']]
  for (i in seq_along(obs_groups)) {
    obs.i <- obs_groups[i]
    obs_value_i <- hfile_obs[[obs.i]][]
    if (obs.i %in% names(x = hfile_cate)){
      # If the observable has missing values, they're labeled as -1.
      # Then you add to add -1 to the categories
      if (length(unique(obs_value_i)) == length(hfile_cate[[obs.i]][])){
        obs_value_i <- factor(x = obs_value_i, labels =  hfile_cate[[obs.i]][])
      }else{
        labels = c('-1', hfile_cate[[obs.i]][])
        obs_value_i <- factor(x = obs_value_i, labels =  labels)
      }
    }
    matrix[,i] <- obs_value_i
  }
}