Why does Pastas recompute the simulation when tmin/tmax are provided in plot methods?

[dbrakenhoff]

@AnneDeGraaf raised an interesting question about why Pastas recomputes the simulation instead of just clipping the axes to the provided limits.

Quoting from #868:

These are plot_* methods. It seems to me these kinds of methods should only visualize results and not recompute anything. In other words, the tmin and tmax values should only work on the plotting functionality e.g. setting xlims through matplotlib. However, the way these changes are implented now, the tmin and tmax values are fed back to various computing methods which recompute some results which are then plotted. Notably in:
a. Model.simulate
b. Model.residuals
c. Model.noise
d. Model.get_contributions
I think these recomputations are only aesthetical, but the behavior is very obscure. Also, from an architecture standpoint, it's inconsistent: through these plot_* methods you give access to the above mentioned Model methods, but you're being selective about which kwarg can be changed and which can't. Would it be okay to change the other kwargs like name, freq, warmup etc too? And what would happen if in the future we were to make some changes to the Model.simulate method that involve tmin and tmax, could that lead to unexpected behavior when a using the plot_* methods?

[dbrakenhoff]

This is a really good question. We discussed this during a Pastas developers meeting, and it took us a while to remember that this is intentional :).

The short answer is: we want users to be able to create plots with simulations outside of the calibration period of the model.

Say we have data from 1980 to 2020, but we want to calibrate on the period 2000-2010

# we build some model, ml
ml.solve(tmin="2000", tmax="2010")

Now we can inspect the results of the calibration:

ml.plots.results()

But let's say we want to inspect how well our model is able to simulate the period after 2010. This requires recomputing the simulation.

ml.plots.results(tmin="2010", tmax="2020")

Similarly, the same goes if we want to test our model on the period before 2000

ml.plots.results(tmin="1990", tmax="2000")

So the upside is we can use our plotting routines to quickly inspect the fit of our model outside the model calibration period. The downside is that in rare cases, the warmup period is too short, and modifying the tmin/tmax causes the plot results to differ. A warning will be added in Pastas that warns the users of this after a solve call (#784).