在 altair 中设置单个点标记的大小

Question

I've been trying to figure out how to set individual point sizes in Altair, but can't quite crack it. It's easy to do in matplotlib, but I'm trying to teach myself all the Altair things, and am stumped. Here's how I'd do it in matplotlib and my best attempt in altair:

import numpy as np
import pandas as pd
import random
import matplotlib.pyplot as plt
import altair as alt

def generate_data(x, y):
  data = pd.DataFrame(
      columns = ['x', 'y', 'point_size'],
      data = {
          'x': pd.Series(np.tile(range(1,x+1), y) ),
          'y': np.repeat(range(1,x+1),y),
          'point_size': [random.randint(0,3) for i in range(0, x*y)],
          },
      )
  return data

def plot_points(data):
  fig = data.plot(
      kind='scatter',
      x='x', 
      y='y', 
      marker='.', 
      s = data['point_size']*50, # setting the size factor
      )


def plot_points_altair(data):
  fig = alt.Chart(data[data.point_size>0]).mark_point(filled=True).encode(
      x=alt.X('x'),
      y=alt.Y('y'),
      size=alt.StrokeWidth('point_size:N'),
      )
  return fig

data = generate_data(10,10)
plot_points(data)
plot_points_altair(data)

I filter out the zero values to avoid plotting them, but Altair treats my point sizes like a categorical, and I'd like control the actual point size of each point in, well, points (or pixels or picas or what have you).

Answer 1

You can do this by using a size encoding with scale=None : this will tell Altair to use the raw data to control the values of the specified encoding, rather than implicitly generating a mapping between the data domain and the visual domain:

import numpy as np
import pandas as pd
import random
import altair as alt

def generate_data(x, y):
  data = pd.DataFrame(
      columns = ['x', 'y', 'point_size'],
      data = {
          'x': pd.Series(np.tile(range(1,x+1), y) ),
          'y': np.repeat(range(1,x+1),y),
          'point_size': [50 * random.randint(0,3) for i in range(0, x*y)],
          },
      )
  return data


def plot_points_altair(data):
  fig = alt.Chart(data[data.point_size>0]).mark_point(filled=True).encode(
      x=alt.X('x'),
      y=alt.Y('y'),
      size=alt.Size('point_size:N', scale=None),
      )
  return fig

data = generate_data(10,10)
plot_points_altair(data)