Expected result - spatial weight

[Arch4060]

I've a general question about the calculation of the spatial weight.

I've a smiple grid 3x3 made in qgis with a field as uID and one field for testing simply named value.

With this grid in a gpkg I run some operations with momepy - here is a sample snippet:

import pandas as pd
import numpy as np
import geopandas as gpd

import pyproj
#pyproj.datadir.set_data_dir('C:/Users/glass/anaconda3/envs/notebook/Library/share/proj')
from shapely.geometry import MultiPolygon, Polygon

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec

import libpysal

import momepy

def load_gdf(GRID):
    PATH = 'output/'
    GPKG = PATH + 'simcity.gpkg'
    LAYER = 'Grid_' + GRID
    gdf = gpd.read_file(GPKG, layer=LAYER, engine='pyogrio', use_arrow=True)

    return gdf

GRID_gdf = load_gdf('3x3')
GRID_gdf = GRID_gdf.rename(columns={'id':'uID'})
GRID_gdf = GRID_gdf.explode()  
GRID_gdf['geometry'] = GRID_gdf['geometry'].make_valid()
GRID_gdf['uID'] = GRID_gdf['uID'].astype(str)
GRID_df = GRID_gdf[['uID', 'value']]

limit = momepy.buffered_limit(GRID_gdf, buffer=0.00)
tessellation = momepy.Tessellation(GRID_gdf, unique_id='uID', limit=limit).tessellation

sw0 = momepy.sw_high(k=0, gdf=tessellation, ids='uID')
sw1 = momepy.sw_high(k=1, gdf=tessellation, ids='uID')
sw2 = momepy.sw_high(k=2, gdf=tessellation, ids='uID')
sw3 = momepy.sw_high(k=3, gdf=tessellation, ids='uID')
sw4 = momepy.sw_high(k=4, gdf=tessellation, ids='uID')

sw = sw1

tessellation['neighbours'] = momepy.Neighbors(tessellation, sw,'uID').series
tessellation = tessellation.merge(GRID_df, how='left', on='uID')

tessellation['value_sw'] = momepy.AverageCharacter(tessellation, values='value', spatial_weights=sw, unique_id='uID').series
tessellation['value_sw'] = round(tessellation['value_sw'],2)

f, ax = plt.subplots(figsize=(7, 7), facecolor='ivory')
tessellation.plot(ax=ax, column='value_sw', legend=True, scheme='natural_breaks', k=10, cmap='Blues', edgecolor="black", linewidth=0.25)
tessellation.apply(lambda x: ax.annotate(text=x['value_sw'], xy=x.geometry.centroid.coords[0], ha='center'), axis=1);

ax.set_axis_off()
plt.title('SimCity', fontsize=14, color='black', loc='center')
plt.show()

GPKG
simcity.zip

I get the following result:

I would expect by the given values, all corners would have the same values ?

[martinfleis]

Can you please provide a reproducible example that I can copy&paste to run the code?

[Arch4060]

I've added the full code and the gpkg as zip.file (updated at 10:30 - there were wrong values in the grid)/ The PATHof the gpkg must be adapted to your workspace in def load_gdf(GRID)

[Arch4060]

Copy & Paste Code with WKT

import pandas as pd
import geopandas as gpd

from shapely.geometry import MultiPolygon, Polygon
from shapely import wkt

import matplotlib.pyplot as plt

import libpysal
import momepy

df = pd.DataFrame(
    {
        "uID": ["8", "1", "5", "9", "3", "2", "4", "7", "6"],
        "value": [0.5, 0.5, 1.0, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5],
        "geometry": [
            "POLYGON ((68154.0965 349573.9678, 68164.0965 349573.9678, 68164.0965 349563.9678, 68154.0965 349563.9678, 68154.0965 349573.9678))",
            "POLYGON ((68144.0965 349593.9678, 68154.0965 349593.9678, 68154.0965 349583.9678, 68144.0965 349583.9678, 68144.0965 349593.9678))",
            "POLYGON ((68154.0965 349583.9678, 68164.0965 349583.9678, 68164.0965 349573.9678, 68154.0965 349573.9678, 68154.0965 349583.9678))",         
            "POLYGON ((68164.0965 349573.9678, 68174.0965 349573.9678, 68174.0965 349563.9678, 68164.0965 349563.9678, 68164.0965 349573.9678))",
            "POLYGON ((68164.0965 349593.9678, 68174.0965 349593.9678, 68174.0965 349583.9678, 68164.0965 349583.9678, 68164.0965 349593.9678))",
            "POLYGON ((68154.0965 349593.9678, 68164.0965 349593.9678, 68164.0965 349583.9678, 68154.0965 349583.9678, 68154.0965 349593.9678))",
            "POLYGON ((68144.0965 349583.9678, 68154.0965 349583.9678, 68154.0965 349573.9678, 68144.0965 349573.9678, 68144.0965 349583.9678))",
            "POLYGON ((68144.0965 349573.9678, 68154.0965 349573.9678, 68154.0965 349563.9678, 68144.0965 349563.9678, 68144.0965 349573.9678))",
            "POLYGON ((68164.0965 349583.9678, 68174.0965 349583.9678, 68174.0965 349573.9678, 68164.0965 349573.9678, 68164.0965 349583.9678))",
        ],
    }
)

df["geometry"] = gpd.GeoSeries.from_wkt(df["geometry"])
GRID_gdf = gpd.GeoDataFrame(df, geometry="geometry")
GRID_df = GRID_gdf[['uID', 'value']]

limit = momepy.buffered_limit(GRID_gdf, buffer=0.00)
tessellation = momepy.Tessellation(GRID_gdf, unique_id='uID', limit=limit).tessellation

sw0 = momepy.sw_high(k=0, gdf=tessellation, ids='uID')
sw1 = momepy.sw_high(k=1, gdf=tessellation, ids='uID')
sw2 = momepy.sw_high(k=2, gdf=tessellation, ids='uID')
sw3 = momepy.sw_high(k=3, gdf=tessellation, ids='uID')
sw4 = momepy.sw_high(k=4, gdf=tessellation, ids='uID')

sw = sw1

tessellation['neighbours'] = momepy.Neighbors(tessellation, sw,'uID').series
tessellation = tessellation.merge(GRID_df, how='left', on='uID')


tessellation['value_sw'] = momepy.AverageCharacter(tessellation, values='value', spatial_weights=sw, unique_id='uID').series
tessellation['value_sw'] = round(tessellation['value_sw'],2)

f, ax = plt.subplots(figsize=(7, 7), facecolor='ivory')
tessellation.plot(ax=ax, column='value_sw', legend=True, scheme='natural_breaks', k=10, cmap='Blues', edgecolor="black", linewidth=0.25)
tessellation.apply(lambda x: ax.annotate(text=x['value_sw'], xy=x.geometry.centroid.coords[0], ha='center'), axis=1)

ax.set_axis_off()

plt.title('SimCity', fontsize=14, color='black', loc='center')
plt.show()

I'm not sure if I do something wrong but it seems to be something with the numbers of the neighbours wrong, or I don't understand it?

[martinfleis]

You got me confused a lot...

If you do this based on the original GRID_gdf rather than it works as expected. If you do it based on tessellation, it also works as expected because there is an artefact caused by discretisation in the corners making some polygons not neighbours. You can't really assume that a 4-way queen contiguity on the corners will be preserved in tessellation. That algo is not designed like that.

You essentially created an artificial corner case.