Exploratory (Spatial) Data Analysis

Jon Reades - j.reades@ucl.ac.uk

1st October 2026

For this one, you really need to read the docs.

Getting Spatial (with Boroughs)

import geopandas as gpd
url = 'https://bit.ly/3neINBV'
boros = gpd.read_file(url, driver='GPKG')
boros.plot(color='none', edgecolor='red', figsize=(8,4.5));

Convex Hull

boros['hulls'] = boros.geometry.convex_hull
boros.set_geometry('hulls').plot(column='NAME', categorical=True, alpha=0.5, figsize=(8,4.5));

Dissolve

boros['region'] = 'London'
ldn   = boros.dissolve(by='region') 
ax = ldn.plot(figsize=(8,4.5)) 
boros.plot(ax=ax, color='none', edgecolor='red')

Buffer & Simplify

ldn.buffer(1000).simplify(1000).plot(figsize=(8,4.5))

Difference

And some nice chaining…

ldn.buffer(3000).simplify(2500).difference(ldn.geometry).plot(figsize=(8,4.5))

Legendgrams

Code begins on next slide.

Implementing Legendgrams

import pysal as ps
# https://github.com/pysal/mapclassify
import mapclassify as mc
# https://jiffyclub.github.io/palettable/
import palettable.matplotlib as palmpl
from legendgram import legendgram

f,ax = plt.subplots(figsize=(10,8))
gdf.plot(column='price', scheme='Quantiles', cmap='magma', k=5, ax=ax)
q = mc.Quantiles(gdf.price.array, k=5)

# https://github.com/pysal/legendgram/blob/master/legendgram/legendgram.py
legendgram(f, ax, 
               gdf.price, q.bins, pal=palmpl.Magma_5,
               legend_size=(.4,.2), # legend size in fractions of the axis
               loc = 'upper left', # mpl-style legend loc
               clip = (0,500), # clip range of the histogram
               frameon=True)

Note that the number of colours need to match k, which is 5 in this case.

It should be possible to set up the colormap and bins such that they can be passed to both GeoPandas and Legendgram.

KNN Weights

Implementing KNN

from pysal.lib import weights
w = weights.KNN.from_dataframe(gdf, k=3)
gdf['w_price'] = weights.lag_spatial(w, gdf.price)
gdf[['name','price','w_price']].sample(5, random_state=42)

	name	price	w_price
83	Southfields Home	85.0	263.0
53	Flat in Islington, Central London	55.0	190.0
70	3bedroom Family Home minutes from Kensington Tube	221.0	470.0
453	Bed, 20 min to Liverpool st, EAST LONDON	110.0	186.0
44	Avni Kensington Hotel	430.0	821.0

Spatial Lag of Distance Band

Implementing Lags

w2 = weights.DistanceBand.from_dataframe(gdf, threshold=2000, alpha=-0.25)
gdf['price_std'] = (gdf.price - gdf.price.mean()) / gdf.price.std()
gdf['w_price_std'] = weights.lag_spatial(w2, gdf.price_std)
gdf[['name','price_std','w_price_std']].sample(5, random_state=42)

	name	price_std	w_price_std
83	Southfields Home	-0.27	0.00
53	Flat in Islington, Central London	-0.51	-0.58
70	3bedroom Family Home minutes from Kensington Tube	0.83	0.46
453	Bed, 20 min to Liverpool st, EAST LONDON	-0.07	-0.82
44	Avni Kensington Hotel	2.52	3.25

Moran’s I

mi = esda.Moran(gdf['price'], w)
print(f"{mi.I:0.4f}")
print(f"{mi.p_sim:0.4f}")
moran_scatterplot(mi)

Local Moran’s I

Implementing Local Moran’s I

lisa = esda.Moran_Local(gdf.price, w)
# Break observations into significant or not
gdf['sig'] = lisa.p_sim < 0.05
# Store the quadrant they belong to
gdf['quad'] = lisa.q
gdf[['name','price','sig','quad']].sample(5, random_state=42)

	name	price	sig	quad
83	Southfields Home	85.0	False	3
53	Flat in Islington, Central London	55.0	False	3
70	3bedroom Family Home minutes from Kensington Tube	221.0	False	1
453	Bed, 20 min to Liverpool st, EAST LONDON	110.0	False	3
44	Avni Kensington Hotel	430.0	False	1

Full LISA

plot_local_autocorrelation(lisa, gdf, 'price')

Additional Resources

There’s so much more to find, but:

• Pandas Reference
• EDA with Pandas on Kaggle
• EDA Visualisation using Pandas
• Python EDA Analysis Tutorial
• Better EDA with Pandas Profiling [Requires module installation]
• Visualising Missing Data
• Choosing Map Colours

Thank You

References