geoshiny 0.0.3

GIS data renderer

Geoshiny

This is a tool to create customized maps using OpenStreetMap data.

NOTE: this software is under development, the interface and the functionalities can and probably will change.

Features

• Generate customized maps of cities and small regions
• Outputs:
  • SVG
  • PNG
  • many others
• Store a filtered intermediate representation in JSONL to easily generate images without a database

Input data

First you need a PostGIS database with the data to represent in flex format. To create that, get a PDB file of the region you are interested to represent (you can also import the whole planet.osm file) and use osm2pgsql with the flex output to run the pgosm-flex scripts.

If you have Docker, you can find a script that populates a PostGIS instance for you in script_utils/database_from_pbf.sh, run it with the pbf file name as an argument.

The connection string for this database has to be put in the environment variable PGIS_CONN_STR

export PGIS_CONN_STR=postgres://username:password@localhost:5432/osm_data

Installation

You will probably need to install the geos library on your system, on Ubuntu for example is the package libgeos-dev, on macOS using bres it's geos.

The default installation does not require GDAL and cannot generate geoTIFF. That functionality is in the geotiff extra and you will need the GDAL libraries installed on your computer, the procedure depends on the OS and can be tricky.

Then, requirements.txt dependencies need to be installed, for example with

python3 -m pip install --no-binary Shapely -r requirements.txt

there are a lot of bugs related to the way GDAL is installed, for example you may need to reinstall it after installing numpy, and avoid using the wheel but let it build (python3 -m pip install --no-binary GDAL GDAL).

Also, GDAL seems to need to see the gdal commands when building, or it fails at runtime. So if you are using a virtualenv you'll need to activate it first or alter you PATH if you are in a CI pipeline (check the Makefile for examples).

I still have to find reliable ways to install and import GDAL and Shapely without this import order problem, so for now I use only Shapely.

Usage

NOTE: this is under development, usage will change soon

The library expects two callbacks, representation and renderer.

The representation is called for each map feature in the given extent, receives the OSM tags as an input and returns None for features to be ignored or an arbitrary Python object (usually a dictionary) with the relevant information to represent. Can also alter the Shapely geometry.

The renderer will receive the output of the representation function and produce the matplotlib attributes like color and alpha.

So one takes care of deciding what to represent and the other of how to represent it. This decoupling allows to change representation and store intermediate values in a file. Using file_to_representation you can generate the representation once and render different extents with different styles easily without even running a database instance.

import asyncio

from geoshiny.database_extract import raw_data_from_extent
from geoshiny.types import (
    ExtentDegrees,
    ObjectStyle,
)

from geoshiny.draw_helpers import (
    data_to_representation,
    data_to_representation_file,
    file_to_representation,
    representation_to_figure,
)

def representation(osm_id: int, geom, tags: dict) -> Optional[dict]:
    if tags.get("bicycle") == "designated":
        return dict(path_type="bike")
    if "water" in tags:
        return dict(surface_type="water")

    if tags.get("landuse") == "grass":
        return dict(surface_type="grass")
    if tags.get("leisure") == "park":
        return dict(surface_type="grass")
    [...]

def renderer(osm_id: int, shape: BaseGeometry, d: dict):
    water_style = ObjectStyle(facecolor="blue", edgecolor="darkblue", linewidth=0.1)
    grass_style = ObjectStyle(facecolor="green", linewidth=0.1)

    if d.get("surface_type") == "water":
        return water_style
    [...]

# northern part of Rostock, Germany
extent = ExtentDegrees(
        latmin=54.0960,
        latmax=54.2046,
        lonmin=12.0029,
        lonmax=12.1989,
    )
# this directly renders to a file
generate_chart(
  'generated.png',
  extent,
  representation,
  renderer,
)
# but you can run the steps yourself
loop = asyncio.get_event_loop()
db_data = loop.run_until_complete(raw_data_from_extent(extent))
reprs = data_to_representation(db_data, entity_callback=representation)

db_img = representation_to_figure(reprs, extent, renderer, figsize=3000)

db_img.savefig("image.png")
db_img.savefig("image.svg")

data_to_representation_file(
        db_data,
        'somefile.jsonl',
        entity_callback=representation,
    )
img2 = representation_to_figure(
        file_to_representation('somefile.jsonl'),
        extent,
        renderer,
        figsize=3000,
    )
img2.savefig("image2.png")
img2.savefig("image2.svg")

Testing

NOTE: this will also probably change, I'm looking at ways to run the tests without git-lfs

To run the test you need git-lfs and docker installed. If you didn't have git-lfs enabled before cloning the repo, you have to do git lfs pull to download the SQL dump file first. Use make test-from-zero to perform a complete integration test, it will use a dump of a small postgis DB of around 200 MB to create a dockerized postgis instance and run scripts against that. The instance id deleted after the test or in case of errors.

TODO

• "Proper" automated tests with a reasonable data fixture
• Examine the possibility of removing GDAL or making it optional, it's a pain to install
• Add labeling/text
• XKCD style output (from matplotlib, should work out of the box)
• layers/pipelines to further process the output
• Offer both async and sync access if possible, hiding the loop to sync users
• Visual comparison of output images (may require opencv as a test dependency, is it worth it?)
• Helper to generate world files (https://en.wikipedia.org/wiki/World_file)
• 3D output (check QGIS formats / glTF)
• tileset output?
• Create and document helpers to make the usage simpler (once the interface is stabilized)
• Examples and screenshot gallery
• Spatialite support?
• Optional pipeline definition interface to combine processing steps?
• Graph functionalities (e.g. show road distance from a set of POI) ?
• MoviePy integration example
• Integrated webserver/UI? Maybe better off as a separate thing