Running through the readme sequentially results in a warning

[tmcw]

Currently, if you run through the label-maker readme example - with the togo country config.json, and running in order:

label-maker download
label-maker labels
label-maker preview -n 10
label-maker images
label-maker package

The last step produces a warning for a missing file:

Saving QA tiles to data/togo.mbtiles
   100%     26.6 MiB       1.9 MiB/s            0:00:00 ETA
Retiling QA Tiles to zoom level 12 (takes a bit)
174704 features, 10870707 bytes of geometry, 4 bytes of separate metadata, 2597853 bytes of string pool
  99.9%  12/2060/1976
Determining labels for each tile
---
Roads: 11 tiles
Buildings: 9 tiles
Total tiles: 12
Writing out labels to data/labels.npz
Writing example images to data/examples
Downloading at most 10 tiles for class Roads
Downloading at most 10 tiles for class Buildings
Downloading 11 tiles to data/tiles
Couldn't open data/tiles/2063-1978-12.jpg, skipping
Saving packaged file to data/data.npz

Admittedly I haven't done enough digging yet, but it would look like there's possibly an off-by-one error here.

[tmcw]

Looks like this is #63

[jreiberkyle]

Hmm.. I wonder if this is truly related to #63? I get that skipping error as well on my slightly altered run through of the README here. I chocked it up to being due to tiles that don't contain either of the classes not being downloaded. Could you check with previewing just 9 or 8 tiles (making sure you clear out the preview directories)?

[drewbo]

@jreiberkyle yup, this is due to some lazy packaging scripts (it searches for an image for each tile even though, depending upon the config parameters, we don't download an image for each tile). Because the image download in the previous step is partially randomized (for background tiles), I think one of two approaches makes the most sense:

1. Keep the code as is, drop the warning
2. Glob the images directory and iterate over that instead of the full label/tile set

Any preferences? @jreiberkyle @tmcw @wronk

[wronk]

Do we want to support users to directly manipulate images directory (by adding or removing examples)? It seems like we might want (2) if that's the case. I'd also lean towards always alerting users to any mismatches.

Counter question to myself on (2): are there any risks of /images not matching up with /labels? Would that confuse users if we use a file list directly from /images but they expect something else based on their input to the labels command?

[drewbo]

@wronk I think both options will create the same output in most cases, even if users manually edit the /images directory (in the first case, we skip the image, in the second, we never glob it)

[wronk]

Maybe the second one makes more sense then? I really don't have any strong opinions here, so whatever you think will make it easier for the user is okay with me.

[jreiberkyle]

(it searches for an image for each tile even though, depending upon the config parameters, we don't download an image for each tile).

@drewbo in which instances are image tiles not downloaded? Is this only the case for background tiles?

[drewbo]

@jreiberkyle yup, either all background tiles or just a random portion of them (depending upon the background_ratio property).

I'm going to implement option 1 (drop the warning) and assume that everything in the images folder is what people want (if there were download issues, it will surface while downloading). If there are corrupt files (i.e. OSError as opposed to FileNotFoundError) then we can show a warning and skip (and probably resolve #43 in the process)

[jreiberkyle]

Sounds great @drewbo !