harmonize returned values for lambda

[wpetry]

I find it confusing that lambda returns λ when type_lambda = 'all' and type_lambda = 'last', but returns log(λ) for type_lambda = 'stochastic'. I suggest adding a logical argument log that allows the user to toggle between these scales. Given the function name lambda, I think it would be most natural to set log = FALSE as the default.

[levisc8]

Hi @wpetry - I thought it was pretty standard to use log lambda for stochastic models. Is that not the case?

Would mean(n_t_1/n_t) as the default for stochastic lambda make more sense then?

[wpetry]

You're right about standard usage of the log scale. It just surprised me that the same function returned values a different scale by default. I think I'm tying the function name, lambda, to the linear scale (vs. log_lambda). Moreover, it's not always easy to immediately identify the scale based on the returned value (e.g., log(2.5)=0.92).

I see three options:

1. prioritize standard usage: this is the current behaviour
2. prioritize consistency: always report lambda on the linear scale (here, log_lambda could optionally be added as a function that wraps lambda)
3. prioritize explicit choice: add a boolean argument for log that defaults to TRUE if stochastic model, else FALSE (i.e., mirroring the current behaviour). However, when log = TRUE by default, return an informative warning: log lambda is returned for stochastic models; set 'log = FALSE' for lambda on linear scale. This preserves the existing behaviour, but gives the user more information about how to interpret the output. The warning could be suppressed simply by explicitly setting the log argument.

I like option 3.