DS Visualisation

Picture is worth a thousand words

Dominik Matula

Outline

Intro
Charts ZOO
Plotting tools
Grammar of graphics
Plotnine
Plotting geo data

Motivation

Motivation

Project usecase

Not so uncommon scenario:
- You have a binary target
- You have (engineered) tons of features
- You need to find out how much information do they contain
Examples:
- Dynamic pricing project (CZ) – see example report
- MCF Hackathon (UK)
- Next best offer (CZ)
- …

Charts ZOO

How to decide what chart to use

Sometimes it’s obvious. Sometimes it’s a hard problem…
Think about what story would you like to tell.
- What is the key message you need to continue the story with?
- Is there a standard plot type that matches your needs?
- …
Note:
- It’s like belles-lettres. The more you read, the better you know what’s realy good.
- The best way is to practice. (Data journalism, feedback from others, …)
- Of course, you need to know your tools.

I prefer to start with a pencil and a sheet of paper…

Graphs Gallery - link

Tip: Caveats section - link

Why is pie chart bad by definition?
Whether to cut y axis in barplots to emphasize changes?
How to deal with spaghetti in your linecharts?

Seaborn examples - link

Plotting tools intro

Optional part of the lecture

Most common plotting tools

matplotlib (2003)
seaborn (2013)
plotly (2013)
plotnine (2017)
seaborn.objects (2022)

Matplotlib

Library for making 2D plots in Python
Built on top of NumPy
By John D. Hunter (1968-2012), neurobiologist
- Build as a patch to IPython for enabling MATLAB-style plotting

import matplotlib as mpl
import matplotlib.pyplot as plt

Matplotlib: Figure x Axes

Axes (canvas) vs Figure (whole picture)
- Figure contains one or more axes
You can get them via plt.figure() (returns Figure), plt.subplots() (returns Figure and Axes tuple)
But we won’t be studying plotting in matplotlib
- Although, some details will be covered later.
- See the docs for more details..

Matplotlib figure - example

import matplotlib.pyplot as plt

fig, ax = plt.subplots()

fig, axes = plt.subplots(nrows=2, ncols=2)

fig = plt.figure()
fig.add_axes([0,0,0.5,0.5])    # xmin, ymin, dx, dy
fig.add_axes([0.5,0.5,0.5,0.5]) 

axes = fig.axes
axes[0].set_title("1st")
axes[1].set_title("2nd")

fig.show()

Seaborn 101

Seaborn is Python library for making statistical graphics
- Built on top of matplotlib
- Closely related to pandas DataFrames
- De facto standard tool for Python DS visualizations.
Three figure-level functions:
- relplot - x~y relation plots
  (scatterplot, linechart, ..)
- displot - distribution plots
  (histogram, KDE, ..)
- catplot - categorical plots
  (barplot, boxplot, ..)
A lot of params – docs is your friend!
- Tip: kind = change plot type, col = create subplots (facets)

Seaborn 101 - examples

import seaborn as sns

tips = sns.load_dataset("tips")
tips.head()

	total_bill	tip	sex	smoker	day	time	size
0	16.99	1.01	Female	No	Sun	Dinner	2
1	10.34	1.66	Male	No	Sun	Dinner	3
2	21.01	3.50	Male	No	Sun	Dinner	3
3	23.68	3.31	Male	No	Sun	Dinner	2
4	24.59	3.61	Female	No	Sun	Dinner	4

sns.displot(data = tips, x="total_bill")

sns.displot(data = tips, x="total_bill", kind="kde")

sns.relplot(data = tips, x="total_bill", y="tip", color="black")

sns.relplot(data = tips, x="total_bill", y="tip", hue="sex")

sns.catplot(data = tips, x="sex", y="tip", kind="box")

sns.displot(data = tips, x="tip", hue="time", col="sex", kind="kde")

Seaborn 102

There are other functions to produce plots in Seaborn
- sns.scatterplot, sns.relplot, …
- You plot them directly on a canvas (ax)
  - Last active is used in case you do not provide any (ax=ax)
  - You get a new one, if you don’t have any
Plus you can customize default outputs using matplotlib (add lines, texts, etc.)
Usual approach
- Create canvas (figure, ax/es)
- Plot charts on ax/es
- Customize (add labels etc.)
- And .show the figure (if needed)

Seaborn 102 - example

import seaborn as sns
import matplotlib.pyplot as plt

fig, axes = plt.subplots(nrows=1, ncols=2)
sns.scatterplot(data=tips, x="total_bill", y="tip", hue="sex", ax=axes[0])
sns.scatterplot(data=tips, x="total_bill", y="tip", hue="smoker", ax=axes[1])

fig.suptitle("Tipping")
for ax in axes:
    ax.set_xlabel("Total bill [$]")
    ax.set_ylabel("Tip [$]")

fig.show()

Seaborn 103

We can use matplotlib to customize plot
- ax’s methods to draw some extra stuff
- Plus there are other ways (e.g, plt itself)
Some of the most handy artefacts to add
- ax.axvline, ax.axhline – vertical/horizontal line
- ax.axline – arbitrary line (xy1, xy2 or slope)
- ax.annotate – add arbitrary text (incl. arrow if needed)
- ax.arrow – add only the arrow
- …
You can customize labels etc. as well
- ax.set_title
- ax.set_xlabel, ax.set_ylabel
- …

Seaborn 103 - example

fig, ax = plt.subplots(1, 1)
sns.scatterplot(data=tips, x="total_bill", y="tip", ax=ax)
ax.set_title("Nadpis")
ax.axhline(y=2, color="red")
ax.text(x=35, y=1.6, s="Kritická hranice", color="red")
ax.annotate(
    text="Pepa",
    xy=[7.3, 5.5],
    xytext=[7.3, 7],
    arrowprops=dict(facecolor="black"),
    horizontalalignment="center",
)

fig.show()

Seaborn Help - link

Imperative x Declarative
plotting

Leonardo’s Task

Imperative plotting

Providing detailed instructions step by step
- Draw a yellow circle in the middle.
- Draw two small black circles side by side..
- (…)
- Paint a small white dot on position x=37, y=59.
Pros:
- Full control.
- Get whatever you want
Cons:
- Tedious.
- Leonardos are rare.
matplotlib (to some extent)

Declarative plotting

Asking a friend/artist to draw a picture according to your needs
- Can you draw me a young lady with dark hair sitting alone and having a misterious smile on her face?
Pros
- Nicely looking results out-of-the box.
Cons
- Lower level of control.
- Need to know language to express your needs.
Functional API: seaborn
Object API: plotnine, seaborn.objects

Bonus: Declarative + AI

Sorcery plotting

Like Declarative, but instead of explaining your needs you just wave your magic wand and scream a magic formula..
- Mona Lisa!
Pros
- Negligible effort.
- Masterpiece?
Cons:
- Need to know (a lot of) magic words.
- Might not correspond to your needs.
- It usually needs to be customized afterwards.
  But sometimes it’s not possible…
Databricks, Excel (and seaborn..)

Ex: Imperative - `matplotlib`

import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
tips = sns.load_dataset("tips")

# Create a figure and axes:
fig, ax = plt.subplots()

# Plot the data:
ax.scatter(x=tips['total_bill'], y=tips['tip'])

# Fit and plot a linear regression line:
m, b = np.polyfit(x=tips['total_bill'], y=tips['tip'], deg=1)
ax.plot(tips['total_bill'], m*tips['total_bill'] + b, color='red')

# Set the labels, title, and legend:
ax.set_xlabel('Total bill')
ax.set_ylabel('Tip')
ax.set_title('Imperative Example - Matplotlib (fig, ax)')
ax.legend(['Data', 'Linear fit'])

Ex: Declarative - `seaborn`

import matplotlib.pyplot as plt
import seaborn as sns
tips = sns.load_dataset("tips")

# Ddeclare varibles -> elements relationship
sns.regplot(data=tips, x='total_bill', y='tip', line_kws={'label': 'Linear Fit', 'color': 'red'})

# Add non-data ink
plt.title('Declarative Example - Seaborn')
plt.legend(labels=['Data', 'Linear Fit']);

Ex: Declarative - `seaborn.objects`

import seaborn.objects as so
import seaborn as sns
tips = sns.load_dataset("tips")

(
    tips
    .pipe(so.Plot, x="total_bill", y="tip", color="smoker")
    .add(so.Dot())
    .add(so.Line(color="red"), so.PolyFit())
    .label(
        title="Declarative using seaborn.objects"
    )
)

Ex: Sorcery - Databricks

Grammar of graphics

How to communicate our plotting desires?

What is it?

Grammar is the system and structure of a language.

. . .

Grammar of Graphics is a framework that allows us to explain the components of any visual in a straightforward manner.

Layered Grammar of Graphics

By Hadley Wickham, 2010
Journal of Computational and Graphical Statistics
Extensions and refinements developed while working on ggplot2

A Layered Grammar of graphics, available online

7 Components of GG

Theme = Adds all non-data ink
Coordinates = How do we position the visual?
Statistics = How we preprocess the data?
Facets = Do we split the visual into subplots?
Geometric objects = What marks are we using?
Aesthetics = How do we show it?
Data = What do we want to show?

Important

This is not how seaborn works. Its author implements this in seaborn.objects module. (seaborn >= 0.12)
Nevertheless, it worths to know how to think about plots. We’re covering just basics relevant for seaborn.

GG in Python

seaborn.objects
- Object API on top of seaborn
- Since late 0.11.x version (preview)
- Big step forward (wrt. seaborn), but still limited & slow development
plotnine
- Implementation of Grammar of graphics in Python by Hassan Kibirige
- Based on ggplot2 (R) (Layered Grammar of Graphics, Hadley Wickham)
- Built with: matplotlib, pandas, mizani (scales), statsmodels, scipy
Follows unusuall ggplot’s syntax: gluing layers by + operand

import plotnine as gg  # p9, pn, ...

gg.ggplot().show()

1) Data

This makes the difference - Painting vs Plotting
Data format
- wide vs. long vs. tidy data
- Use pandas .pivot, .melt, .long_to_wide to switch formats

Ex: data - Hazard

Adjusted Hazard - log-ins
- sessions clipped to 2.5h (cut end)
- overlaps resolved (cut first)
- deduplicated
Added few characteristics
- Hour / Day of week / Day of month / Week of year / Month
- Place characteristics
- Riskness of the player based on our model

	df_v_konto_key	df_v_misto_key	df_v_herni_pozice_key	prihlasenicas	odhlasenicas	doba_hrani	doba_pauzy	typmisto	sidlokodruian	jtsk_y	...	obec	psc	ulice	cp	hodina	den_v_tydnu	den_v_mesici	mesic	tyden
0	31347675069	127018397	156812556	2023-03-31 17:27:24	2023-03-31 17:31:33	249.0	NaN	H	18167829.0	464086.17	...	Lysá nad Labem	28922	Smetanova	789	17	Friday	31	March	13
1	31347675069	127018397	156812550	2023-03-31 17:34:21	2023-03-31 17:27:24	-417.0	0.0	H	18167829.0	464086.17	...	Lysá nad Labem	28922	Smetanova	789	17	Friday	31	March	13
2	31347675069	127018397	156812571	2023-03-31 18:09:20	2023-03-31 17:34:21	-2099.0	0.0	H	18167829.0	464086.17	...	Lysá nad Labem	28922	Smetanova	789	18	Friday	31	March	13
3	31347675069	127018397	156812571	2023-03-31 18:09:37	2023-03-31 18:09:20	-17.0	0.0	H	18167829.0	464086.17	...	Lysá nad Labem	28922	Smetanova	789	18	Friday	31	March	13
4	31347675069	127018397	160811749	2023-03-31 18:29:13	2023-03-31 18:09:37	-1176.0	0.0	H	18167829.0	464086.17	...	Lysá nad Labem	28922	Smetanova	789	18	Friday	31	March	13

5 rows × 22 columns

Ex: `plotnine`

Goal: What is the distribution of log-ins in time? (hodina)
- Note: you can specify data globally in ggplot() and locally in given layer (stay tuned)

gg.ggplot(data = prihl).show()

2) Aesthetics

sns.scatterplot(data=tips, x="total_bill", y="tip", hue="smoker")  # using mapping

2) Aesthetics

Mapping of variable values to chart properties
- sns.relplot(data=tips, x="total_bill", y="tip")
- It’s called Aesthetics (ggplot2) x (Mark) properties (seaborn.objects)
How to:
- use mapping = gg.aes() to assign column names to aesthetics
  - you can set it globally (ggplot() param) or localy (on geometry-level, stay tuned)
  - e.g.: ggplot(data=tips, mapping=gg.aes(x = "total_bill", y="tip", color="sex"))
- use scale_<scalename>_<scaletype>() to customize aesthetics mapping
  - e.g.: scale_x_continuous(breaks = np.arange(0, 100, 10))
  - e.g.: scale_color_manual(values = {"Male": "blue", "Female": "orangered"})

Important

Reminder: not fully integrated concept in bare seaborn!

Non-aesthetics way

pocty_her = (
    prihl
    .groupby(["den_v_tydnu"], as_index=False)
    .agg(n = ("df_v_konto_key", "count"))
)

x_vals = pocty_her["den_v_tydnu"]
y_vals = pocty_her["n"]

sns.barplot(x=y_vals, y=x_vals)

Common aesthetics

Position: x, y
Size: pointsize, linewidth
Color: hue (single color provided by color)
Transparency: alpha
Style: marker, fill, linestyle, edgestyle
Text properties: text, halign, valign, fontsize

Ex: `plotnine`

Goal: What is the distribution of log-ins in time? (hodina)
How do we visualize variable distribution? histogram
What aesthetics does the plot have? x

gg.ggplot(data = prihl, mapping = gg.aes(x="hodina"))

(
    gg.ggplot(data = prihl, mapping = gg.aes(x="hodina"))
    + gg.scale_x_continuous(breaks = np.arange(0, 24, 2))
)

3) Statistics

Sometimes it’s necessary to aggregate data before mapping
- Barplot (count)
- Histogram, KDE (cut & count)
- Regression line (model fit)
- Boxplot (quantiles, IQR)
Either part of the plot or aggregated beforehand

Ex: `plotnine`

Goal: What is the distribution of log-ins in time? (hodina)
How do we visualize variable distribution? histogram
What aesthetics does the plot have? x
What statistic do we need? count

(
    gg.ggplot(data = prihl, mapping = gg.aes(x="hodina"))
    + gg.stat_count()
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
)

4) Geometric objects

How to actually visualize the data (data ink)
- Relies on aesthetics
- List of aesthetics differ among geoms
Basic ones
- points, lines, bars, tiles, text, …
Custom and compounded
- histogram, boxplot, area, contour, …

Ex: `plotnine`

Goal: What is the distribution of log-ins in time? (hodina)
How do we visualize variable distribution? histogram
What aesthetics does the plot have? x
What statistic do we need? count
What geom will we use to display the statistic?

(
    gg.ggplot(data = prihl, mapping = gg.aes(x="hodina"))
    + gg.stat_count(geom="bar")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
)

(
    gg.ggplot(data = prihl, mapping = gg.aes(x="hodina"))
    + gg.stat_count(geom="line")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
)

Ex: `plotnine`

Goal: What is the distribution of log-ins in time? (hodina)
How do we visualize variable distribution? histogram
What aesthetics does the plot have? x, group
What statistic do we need? count
What geom will we use to display the statistic? line

(
    gg.ggplot(data = prihl, mapping = gg.aes(x="hodina", group="tyden"))
    + gg.stat_count(geom="line")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
)

(
    gg.ggplot(data = prihl, mapping = gg.aes(x="hodina", group="tyden"))
    + gg.geom_line(stat="count", alpha=.3)
    + gg.geom_vline(xintercept=[1, 8], color="firebrick", linetype="dashed")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
)

Ex: `plotnine`

Goal: What is the distribution of log-ins in time? (hodina)
How do we visualize variable distribution? histogram
What aesthetics does the plot have? x, group
What statistic do we need? count
What geom will we use to display the statistic? line, vline
Will we use any facets? (mesic)

(
    gg.ggplot(data = prihl, mapping = gg.aes(x="hodina", group="tyden"))
    + gg.geom_line(stat="count", alpha=.3)
    + gg.geom_vline(xintercept=[1, 8], color="firebrick", linetype="dashed", alpha=.1)
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 6))
    + gg.facet_wrap("mesic")
)

6) Coordinates

A way the graph is projected to screen
Common coordinate systems
- Cartesian
- Polar
Adjust point-of-view
- Zooming-in
- Flipping
- Fixing aspect ratio
- …

Example:

Ex: `plotnine` - `coord_flip`

(
    prihl
    .assign(mesic = lambda d: pd.Categorical(d["mesic"], categories = list_mesicu, ordered=True))
    .pipe(gg.ggplot, mapping = gg.aes(x="mesic"))
    + gg.geom_histogram(stat="count")
)

(
    prihl
    .assign(mesic = lambda d: pd.Categorical(d["mesic"], categories = list_mesicu[::-1], ordered=True))
    .pipe(gg.ggplot, mapping = gg.aes(x="mesic"))
    + gg.geom_histogram(stat="count")
    + gg.coord_flip()
)

7) Theme

Represents all non-data ink
- Labels, titles, …
- Comments
- Grid lines
- Higlights
- Background
- …
Same Data->Aes->Stats->Geoms…, different feeling

Tip

Postprocessing might involve 3rd party tools as well.

Ex: `plotnine` - resolving night time

vsechny_casy = pd.DataFrame(
    [(tyden, hodina) for tyden in range(0, 52) for hodina in range(0, 24)], 
    columns=["tyden", "hodina"]
)

pocty_prihlaseni = (
    prihl
    .groupby(["hodina", "tyden"], as_index=False)
    .agg(n = ("df_v_konto_key", "count"))
    .merge(vsechny_casy, on=["hodina", "tyden"], how="right")
    .fillna({"n": 0})
)

(
    gg.ggplot(pocty_prihlaseni, mapping = gg.aes(x="hodina", y="n", group="tyden"))
    + gg.geom_line(stat="identity", alpha=.3)
    + gg.geom_vline(xintercept=[1, 8], color="firebrick", linetype="dashed")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
)

Ex: `plotnine` - labels & comments

(
    gg.ggplot(pocty_prihlaseni, mapping = gg.aes(x="hodina", y="n", group="tyden"))
    + gg.geom_line(stat="identity", alpha=.3)
    + gg.geom_rect(data = lambda d: d.head(3), xmin=1, xmax=8, ymin=0, ymax=600, alpha=.1)
    + gg.geom_vline(xintercept=[1, 8], color="firebrick", linetype="dashed")
    + gg.annotate(label="Povinna vecerka", x=1.5, y=400, geom="text", angle=90, size=7, color="firebrick")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
    + gg.labs(
        x = "Hodina",
        y = "Pocet her",
        title = "Rozdělení týdenních počtů her",
        subtitle = "",
        caption = f"Hazard dataset, {prihl.shape[0]} zaznamu"
    )
)

Ex: `plotnine` - themes

Use theme_... to setup predefined styles
- e.g., theme_light(), theme_dark(), theme_void(), ..
- Tip: _ = gg.theme_set(gg.theme_light()) to set it globally
Use theme() to customize plot appearance
- Syntax: <param> = gg.element_...()
- Example: plot_title = gg.element_text(...)

(
    gg.ggplot(pocty_prihlaseni, mapping = gg.aes(x="hodina", y="n", group="tyden"))
    + gg.geom_line(stat="identity", alpha=.3)
    + gg.geom_rect(data = lambda d: d.head(3), xmin=1, xmax=8, ymin=0, ymax=600, alpha=.1)
    + gg.geom_vline(xintercept=[1, 8], color="firebrick", linetype="dashed")
    + gg.annotate(label="Povinna vecerka", x=1.5, y=400, geom="text", angle=90, size=7, color="firebrick")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 2))
    + gg.labs(
        x = "Hodina",
        y = "Pocet her",
        title = "Rozdělení týdenních počtů her",
        subtitle = "",
        caption = f"Hazard dataset, {prihl.shape[0]} zaznamu"
    )
    + gg.theme_light()
    + gg.theme(
        plot_title=gg.element_text(face="bold"),
        plot_caption=gg.element_text(color="gray", size=7),
        panel_border=gg.element_rect(color="white"),
        panel_grid_minor_x= gg.element_blank()
    )
)

Ex: `plotnine` - final polishing

(
    gg.ggplot(pocty_prihlaseni, mapping = gg.aes(x="hodina", y="n", group="tyden"))
    + gg.geom_smooth(mapping=gg.aes(group=1), method="gpr", color="firebrick")
    + gg.geom_rect(data = lambda d: d.head(3), xmin=1, xmax=8, ymin=0, ymax=600, alpha=.05)
    + gg.geom_line(stat="identity", alpha=.2)
    + gg.annotate(label="Povinna vecerka", x=4.5, y=500, geom="text", size=7, color="gray")
    + gg.scale_x_continuous(breaks=np.arange(0, 24, 1))
    + gg.labs(
        x = "Hodina začátku hry [UTC]",
        y = "Počet her v daném týdnu",
        title = "Rozdělení týdenních počtů her",
        subtitle = "",
        caption = f"Hazard dataset - {prihl.shape[0]:,.0f} zaznamu".replace(",", " ")
    )
    + gg.theme_light()
    + gg.theme(
        plot_title=gg.element_text(face="bold"),
        plot_caption=gg.element_text(color="gray", size=7),
        panel_border=gg.element_rect(color="white"),
        panel_grid_minor_x= gg.element_blank()
    )
)

Ex: Theme - MS Excel

Plotnine recap - (reference)

ggplot() - constructor creating a new ggplot object
aes() - controls aesthetics mapping on global / layer level
- scale_...() - controls aes appearance (e.g., transformation, breaks, labels)
  - e.g., scale_x_continuous, scale_color_manual
geom_...() - geometric layer
- e.g., geom_point, geom_line, geom_histogram, …
stat_...() - controls variable -> aesthetic aggregation
- e.g., stat_identity, stat_count, stat_density, …
facet_...() - variable -> facets (subplots)
- e.g., facet_grid, facet_wrap, ..
theme() - controls overall appearance
- theme_light(), theme_dark(), … - predefined themes
- theme_set() - global setup
- use element_...() to setup given element
coord_...() - controls plotting coordinates

Reading GEO data

gpd.read_file()
- ShapeFile: .shp
- GeoJson: .geojson
- GeoPackage: .gpkg
- Geo databases engine
Input filters
- bbox – bounding box (read only elements in a box)
- mask – intersection filter (read only CZ elements)
- rows – read only specific rows (e.g., rows=10, rows=slice(5,10))
- columns – read only specific columns
- ignore_geometry
Note: can handle various projections (.crs), too.

Reading GEO data – example

Where to get CZ GEO data?
- CUZK ⬅️ we’ll be using this one.
- OpenStreetMaps via GeoFrabrik.de [link] - POIs etc.
- Other sources …

import geopandas as gpd

kraje = gpd.read_file(kraje_shp_file).set_index("NAZEV")
kraje.head(3)

	KOD	REGS_KOD	NUTS3_KOD	geometry
NAZEV
Hlavní město Praha	19	19	CZ010	POLYGON ((-746760.67 -1058142.71, -746757.65 -...
Středočeský kraj	27	27	CZ020	POLYGON ((-700345.18 -989088.81, -700328.64 -9...
Jihočeský kraj	35	35	CZ031	POLYGON ((-711379.92 -1136266.41, -711363.97 -...

`geopandas`’s cool tools

Work with gpd.DataFrame as you work with pd.DataFrame
Beside, geometry can provide (gdf = gpd.DataFrame()):
- Area measurements – gdf.area
- Centroid location – gdf.centroid
- Boundary polygon – gdf.boundary
- Distance measurements – gdf.distance(other_gdf.iloc[0])
- Checking geometry intersections – `gdf.intersects(other_gdf)
- Get convex hull geometry – gdf.convex_hull
Note, it works on active geometry
- gdf.set_geometry(col_name)

`geopandas`’s cool tools – examples

kraje.centroid.head()

NAZEV
Hlavní město Praha    POINT (-739952.022 -1045726.481)
Středočeský kraj       POINT (-734798.177 -1053881.92)
Jihočeský kraj        POINT (-757095.136 -1152712.059)
Plzeňský kraj         POINT (-835551.971 -1084574.549)
Karlovarský kraj      POINT (-859350.657 -1015909.505)
dtype: geometry

kraje["rozloha"] = kraje.area/1e6
kraje["rozloha"]

NAZEV
Hlavní město Praha        496.176193
Středočeský kraj        10928.395151
Jihočeský kraj          10058.126861
Plzeňský kraj            7648.609450
Karlovarský kraj         3310.111472
Ústecký kraj             5338.749064
Liberecký kraj           3163.580958
Královéhradecký kraj     4759.092545
Pardubický kraj          4519.556466
Kraj Vysočina            6795.087814
Jihomoravský kraj        7185.859183
Olomoucký kraj           5271.469023
Moravskoslezský kraj     5430.528592
Zlínský kraj             3961.499796
Name: rozloha, dtype: float64

Plotting

Alas, Seaborn can’t work with geodata directly 😞
Fortunately, GeoPandas provides direct way to plot geometry
Use gdf.plot() (same as pandas’s df.plot())
- Note: kind="geo" plots the map. There are other options as well.
- See Docs for more details

kraje.plot()

Plotting

Map any variable to color via column

Warning

Watch out, it’s not hue anymore!

kraje.plot(column="rozloha")

Adjusting plots via `matplotlib`

You can plot on matplotlib’s canvas via ax param:

fig, axes = plt.subplots(1, 2, figsize=(15, 35))

kraje.plot(ax=axes[0])
axes[0].set_axis_off()
axes[0].set_title("Česká republika")

kraje.plot(column="rozloha", ax=axes[1])
axes[1].set_axis_off()
axes[1].set_title("Rozhloha krajů ČR")

fig.show()

Adding content on geo plots

Just add other layers on the ax
Note: use zorder to control layers order!

centroids = kraje.centroid
borders = kraje.boundary

fig, ax = plt.subplots(1, 1)

# geometry plot
kraje.plot(
    column="rozloha",
    alpha=0.3,
    cmap="OrRd",
    ax=ax,
    legend=True,
    legend_kwds={"label": "Area"},
    zorder=5,
)

# extra geometries
centroids.plot(ax=ax, color="red", markersize=2, zorder=2)
borders.plot(ax=ax, color="red", linewidth=0.1, zorder=3)
ax.set_axis_off()
ax.set_title("Kraje a jejich centra")

# extra matplotlib elements
prague_x = centroids["Hlavní město Praha"].x
prague_y = centroids["Hlavní město Praha"].y
ax.axvline(prague_x, linestyle="dotted", color="black")
ax.axhline(prague_y, linestyle="dotted", color="black")

fig.show()

Dealing with missings

Use missing_kwd dict param to set color, edgecolor, label etc.¨

fig, axes = plt.subplots(1, 2, figsize=(15, 35))

kraje_wo_vysocina = kraje.assign(rozloha = lambda d: np.where(d.index == 'Kraj Vysočina', np.nan, d["rozloha"]))
kraje_wo_vysocina.plot(ax=axes[0])
kraje_wo_vysocina.plot(column="rozloha", missing_kwds=dict(color="red"), ax=axes[1])

for ax in axes:
    ax.set_axis_off()

fig.show()

Exploring

Even more, you can have interactive maps out-of-the-box!
How to:
- Install Python dependencies -folium and mapclassify in my case
- Use .explore() instead of .plot()

kraje.explore(column="rozloha")

Make this Notebook Trusted to load map: File -> Trust Notebook

Plotting with `plotnine`

It’s simple: gpd + geom_map()

(
    gg.ggplot(data=kraje) 
    + gg.geom_map(mapping=gg.aes(fill="rozloha")) 
    + gg.coord_equal()
    + gg.theme_light()
    + gg.theme_void()
    + gg.labs(title = "Kraje CR dle rozlohy")
)

Exercise

Your turn

Load hazard dataset provided by @jhucin
Prepare charts to explore & answer the following questions:
- Does the time distribution differ among week/weekend days?
- Who are 10 the most active users?
- Can we distinguish between regular and occasional players?
- Does the time distribution differ among occasional & regular players?
- How does individual [player/place] play time vary?
- How does typical play time relate to frequency of playings?
- How many places does typical player visit?
- What is the distribution of break times?
  - (Note: what about filtering out last break time in a session?)
[Optional] Extra tasks
- Where in CR are the players the most active?
- Does the play length differ among regions?
- Are the players taking mandatory break times?
- Are risky players more concentrated somewhere? (place/city/region)
- Do the players have any favourite place?
- Do the risky players behave differently? (play time, break time, frequency of playes, …)
- Riskness of a place based on riskness of regular players on that place – are there any suspicious places? (avg / quantile of high risky players / …)
- …

DS Visualisation

Outline

Motivation

Motivation

Project usecase

Charts ZOO

How to decide what chart to use

Graphs Gallery - link

Tip: Caveats section - link

Seaborn examples - link

Plotting tools intro

Most common plotting tools

Matplotlib

Matplotlib: Figure x Axes

Matplotlib figure - example

Seaborn 101

Seaborn 101 - examples

Seaborn 102

Seaborn 102 - example

Seaborn 103

Seaborn 103 - example

Seaborn Help - link

Imperative x Declarativeplotting

Leonardo’s Task

Imperative plotting

Declarative plotting

Bonus: Declarative + AI

Sorcery plotting

Ex: Imperative - matplotlib

Ex: Declarative - seaborn

Ex: Declarative - seaborn.objects

Ex: Sorcery - Databricks

Grammar of graphics

What is it?

Layered Grammar of Graphics

7 Components of GG

GG in Python

1) Data

Ex: data - Hazard

Ex: plotnine

2) Aesthetics

2) Aesthetics

Non-aesthetics way

Common aesthetics

Ex: plotnine

3) Statistics

Ex: plotnine

4) Geometric objects

Ex: plotnine

Ex: plotnine

5) Facets

Ex: plotnine

6) Coordinates

Ex: plotnine - coord_flip

7) Theme

Ex: plotnine - resolving night time

Ex: plotnine - labels & comments

Ex: plotnine - themes

Ex: plotnine - final polishing

Ex: Theme - MS Excel

Plotnine recap - (reference)

Reading GEO data

Reading GEO data – example

geopandas’s cool tools

geopandas’s cool tools – examples

Plotting

Plotting

Adjusting plots via matplotlib

Adding content on geo plots

Dealing with missings

Exploring

Plotting with plotnine

Exercise

Exercise

Imperative x Declarative
plotting

Ex: Imperative - `matplotlib`

Ex: Declarative - `seaborn`

Ex: Declarative - `seaborn.objects`

Ex: `plotnine`

Ex: `plotnine`

Ex: `plotnine`

Ex: `plotnine`

Ex: `plotnine`

Ex: `plotnine`

Ex: `plotnine` - `coord_flip`

Ex: `plotnine` - resolving night time

Ex: `plotnine` - labels & comments

Ex: `plotnine` - themes

Ex: `plotnine` - final polishing

`geopandas`’s cool tools

`geopandas`’s cool tools – examples

Adjusting plots via `matplotlib`

Plotting with `plotnine`