data transformation (either with log1p or log) and raw values

[kenichi-shimada]

We had a discussion about it earlier, but maybe we should switch our data transformation method from log1p to log. Given that csv files can take negative values, we'll have NaNs anyway even if we use log1p, and it was pointed out that log1p might not be appropriate for smaller values (I remember log1p=log(1+p) is like a posterior distribution where 1 is a prior).

On a related note, there are at least a few people including myself who like to work on raw values in the downstream process even though they want to work on the log-transformed data esp to apply the Gaussian mixture model, so it'd be great if we can download raw values for gating parameters (possibly by allowing users to choose between raw and log values). I hope you can consider this.

Thanks!