end simplify_network() immediately if already meets criteria

[jGaboardi]

We need some logic to end sgeop.simplify_network() immediately if the input roads dataset already meets our simplififed criteria. This can be determined within the call to sgeop.nodes.fix_topology() and simply return the result there.

The check should be after induce_nodes() is called where we are checking if roads and roads_w_nodes are equivalent. If they are, no further simplification is needed, as far as I understand.

Thoughts @martinfleis @anastassiavybornova ?

[martinfleis]

If they are, no further simplification is needed, as far as I understand

wrong. Topological correctness does not mean that there are no artifacts.

There is nothing to do if the two data frames are equal and the artifacts dataframe is empty.

[jGaboardi]

So that would be here after the first loop is completed, correct? (if that artifacts df is empty)

[jGaboardi]

When this condition is met, should it raise an error, or can we actually the input simplified enough and return? It seems like this could be true on both the first and second loops?

[martinfleis]

or can we actually the input simplified enough and return

We can't. Threshold detection occasionally fails in certain contexts due to the distribution of block shapes. It should raise as it does now.

[jGaboardi]

OK, I'm wondering if it would be prudent to cut a patch release of momepy after pysal/momepy#666 so at least we can fail gracefully when an skeleton is passed in that can't be polygonized?

[jGaboardi]

After that I can look back to try to determine better logic this.

[martinfleis]

We'd need to backport that. But if you wait a bit, I think that Streetscape will be ready for 0.9.

[jGaboardi]

Yeah, I think this can surely wait. Just wanted to keep it on the radar.

[jGaboardi]

So I have been playing around with some ideas and thinking on this, and (unless I am missing something obvious) the logic gets a bit subjective. We may simply want to make it clear to the user that the algorithm we employ does not decide if a network dataset is "simplified enough" and the user themself needs to determine that.

As a case in point, below is some code and plots of the original Apalachicola, FL testing data (SI-0) and three successive rounds of simplification (SI-{1,2,3}). SI-{1,2} are both clean results, with SI-2 being slightly over-simplified. However, SI-3 leads to dramatic over-simplification while not being "wrong." It should be noted that a fourth round of simplification fails with our current logic

My goal with this was to find some simple & clear criterion for popping out of the process, but this is seems to be more difficult than I had imagined These are some relevant changes on the @GH68_early_terminate_if_all_good branch in:

• simplify.simplify_network()
• artifacts.get_artifacts()

import geopandas, matplotlib.pyplot, sgeop

f = "sgeop/tests/data/apalachicola_original.parquet"
raw_original = geopandas.read_parquet(f, columns=["geometry"]).reset_index(drop=True)

iterations = 3
scene = "Simplification Iteration {}"
scenes = {scene.format(0): raw_original}

for i in range(iterations):
    scenes[scene.format(i+1)] = sgeop.simplify_network(scenes[scene.format(i)].copy())

panels = [[scene.format(0), scene.format(1)], [scene.format(2), scene.format(3)]]
fig, axd = matplotlib.pyplot.subplot_mosaic(panels, figsize=(9, 9), layout="constrained")
for k, ax in axd.items():
    scenes[k].plot(ax=ax)
    ax.axis("off")
    ax.set_title(k)

Perhaps we should leave this issue open for now, but I'll not spend more time on it at this point.

---