问题
In ggplot2
, it's easy to create a faceted plot with facets that span both rows and columns. Is there a "slick" way to do this in altair
? facet documentation
It's possible to have facets plot in a single column,
import altair as alt
from vega_datasets import data
iris = data.iris
chart = alt.Chart(iris).mark_point().encode(
x='petalLength:Q',
y='petalWidth:Q',
color='species:N'
).properties(
width=180,
height=180
).facet(
row='species:N'
)
and in a single row,
chart = alt.Chart(iris).mark_point().encode(
x='petalLength:Q',
y='petalWidth:Q',
color='species:N'
).properties(
width=180,
height=180
).facet(
column='species:N'
)
but often, I just want to plot them in a grid using more than one column/row, i.e. those that line up in a single column/row don't mean anything in particular.
For example, see facet_wrap
from ggplot2
: http://www.cookbook-r.com/Graphs/Facets_(ggplot2)/#facetwrap
回答1:
In Altair version 3.1 or newer (released June 2019), wrapped facets are supported directly within the Altair API. Modifying your iris example, you can wrap your facets at two columns like this:
import altair as alt
from vega_datasets import data
iris = data.iris()
alt.Chart(iris).mark_point().encode(
x='petalLength:Q',
y='petalWidth:Q',
color='species:N'
).properties(
width=180,
height=180
).facet(
facet='species:N',
columns=2
)
Alternatively, the same chart can be specified with the facet as an encoding:
alt.Chart(iris).mark_point().encode(
x='petalLength:Q',
y='petalWidth:Q',
color='species:N',
facet=alt.Facet('species:N', columns=2)
).properties(
width=180,
height=180,
)
The columns argument can be similarly specified for concatenated charts in alt.concat()
and repeated charts alt.Chart.repeat()
.
回答2:
You can do this by specifying .repeat()
and the row
and column
list of variables. This is closer to ggplot's facet_grid()
than facet_wrap()
but the API is very elegant. (See discussion here.) The API is here
iris = data.iris()
alt.Chart(iris).mark_circle().encode(
alt.X(alt.repeat("column"), type='quantitative'),
alt.Y(alt.repeat("row"), type='quantitative'),
color='species:N'
).properties(
width=250,
height=250
).repeat(
row=['petalLength', 'petalWidth'],
column=['sepalLength', 'sepalWidth']
).interactive()
Which produces:
Note that the entire set is interactive in tandem (zoom-in, zoom-out).
Be sure to check out RepeatedCharts and FacetedCharts in the Documentation.
Creating a facet_wrap()
style grid of plots
If you want a ribbon of charts laid out one after another (not necessarily mapping a column or row to variables in your data frame) you can do that by wrapping a combination of hconcat()
and vconcat()
over a list of Altair plots.
I am sure there are more elegant ways, but this is how I did it.
Logic used in the code below:
- First, create a
base
Altair chart - Use
transform_filter()
to filter your data into multiple subplots - Decide on the number of plots in one row and slice up that list
- Loop through the list of lists, laying down one row at a time.
-
import altair as alt
from vega_datasets import data
from altair.expr import datum
iris = data.iris()
base = alt.Chart(iris).mark_point().encode(
x='petalLength:Q',
y='petalWidth:Q',
color='species:N'
).properties(
width=60,
height=60
)
#create a list of subplots
subplts = []
for pw in iris['petalWidth'].unique():
subplts.append(base.transform_filter(datum.petalWidth == pw))
def facet_wrap(subplts, plots_per_row):
rows = [subplts[i:i+plots_per_row] for i in range(0, len(subplts), plots_per_row)]
compound_chart = alt.hconcat()
for r in rows:
rowplot = alt.vconcat() #start a new row
for item in r:
rowplot |= item #add suplot to current row as a new column
compound_chart &= rowplot # add the entire row of plots as a new row
return compound_chart
compound_chart = facet_wrap(subplts, plots_per_row=6)
compound_chart
to produce:
回答3:
Starting from Ram's answer, and using a more functional approach, you could also try:
import altair as alt
from vega_datasets import data
from altair.expr import datum
iris = data.iris()
base = alt.Chart(iris).mark_point().encode(
x='petalLength:Q',
y='petalWidth:Q',
color='species:N'
)
# chart factory
def make_chart(base_chart, pw, options):
title = 'Petal Width {:.2f}'.format(pw)
chart = base_chart\
.transform_filter(datum.petalWidth == pw)\
.properties(width=options['width'], height=options['height'], title=title)
return chart
# create all charts
options = {'width': 50, 'height': 60}
charts = [make_chart(base, pw, options) for pw in sorted(iris['petalWidth'].unique())]
# make a single row
def make_hcc(row_of_charts):
hconcat = [chart for chart in row_of_charts]
hcc = alt.HConcatChart(hconcat=hconcat)
return hcc
# take an array of charts and produce a facet grid
def facet_wrap(charts, charts_per_row):
rows_of_charts = [
charts[i:i+charts_per_row]
for i in range(0, len(charts), charts_per_row)]
vconcat = [make_hcc(r) for r in rows_of_charts]
vcc = alt.VConcatChart(vconcat=vconcat)\
.configure_axisX(grid=True)\
.configure_axisY(grid=True)
return vcc
# assemble the facet grid
compound_chart = facet_wrap(charts, charts_per_row=6)
compound_chart.properties(title='My Facet grid')
This way it should be easy to tweak the code and pass some configuration options to all of your plots (e.g. show/hide ticks, set the same bottom/top limits for all the plots, etc).
回答4:
Here's a general solution that has a spot to add layers. The DataFrame in this case has three columns and is in long form.
numcols=3 # specify the number of columns you want
all_categories=df['Category_Column'].unique() # array of strings to use as your filters and titles
rows=alt.vconcat(data=df)
numrows=int(np.ceil(len(all_categories) / numcols))
pointer=0
for _ in range(numrows):
row=all_categories[pointer:pointer+numcols]
cols=alt.hconcat()
for a_chart in row:
# add your layers here
# line chart
line=alt.Chart().mark_line(point=True).encode(
x='variable',
y='value'
).transform_filter(datum.Category_Column == a_chart).properties(
title=a_chart, height=200, width=200)
# text labels
text=alt.Chart().mark_text().encode(
x='variable',
y='value'
).transform_filter(datum.Category_Column == a_chart)
both = line + text
cols |= both
rows &= cols
pointer += numcols
rows
回答5:
I found that doing a concatenation of length greater than two in either direction caused the data to become distorted and fall out of the window. I solved this by recursively breaking up the subplot array into quadrants and doing alternating row and column concatenations. If you don't have this problem, good for you: you can use one of the simpler implementations already posted. But, if you do, I hope this helps.
def facet_wrap(subplots, plots_per_row):
# base cases
if len(subplots) == 0 or plots_per_row == 0:
return None
if len(subplots) == 1:
return subplots[0]
# split subplots list into quadrants
# we always fill top and left first
quadrants = [[], [], [], []] # tl, tr, bl, br
for subplot_index, subplot in enumerate(subplots):
right_half = (subplot_index % plots_per_row) >= plots_per_row // 2
lower_half = subplot_index >= len(subplots) / 2
quadrants[2 * lower_half + right_half].append(subplot)
# recurse on each quadrant
# we want a single chart or None in place of each quadrant
m = plots_per_row % 2 # if plots_per_row is odd then we need to split it unevenly
quadplots = [
facet_wrap(q, plots_per_row // 2 + m * (0 == (i % 2))) \
for i, q in enumerate(quadrants)
]
# join the quadrants
rows = [quadplots[:2], quadplots[2:]]
colplot = alt.hconcat()
for row in rows:
rowplot = alt.vconcat()
for item in row:
if item != None:
rowplot = rowplot | item
colplot &= rowplot
return colplot
来源:https://stackoverflow.com/questions/50164001/multiple-column-row-facet-wrap-in-altair