 Run in Google Colab
|
 View on GitHub
|
Lecture 10: 1D and Data Mapping¶
In this notebook, we will explore how to visualize data in a 1D context and how to map data to basic visual elements. We will start by showing some older plotting methods using matplotlib and pandas, just for context, and then we will move on to Altair.
The focus here is on taking a gradual approach to Altair. We'll begin with very simple examples, then we will progressively introduce more advanced features like alt.X() and alt.Y().
Dataset: We'll use the heart_data.csv dataset containing medical examination information. Let’s load it and inspect the first few rows.
import pandas as pd
import matplotlib.pyplot as plt
import altair as alt
# If you're running in a notebook environment that limits rows for interactive charts:
alt.data_transformers.disable_max_rows()
# Read the dataset
df = pd.read_csv("../../Datasets/heart_data.csv")
df.head()
The dataset columns are:
id: Patient identifier.
age: Age in days (we might convert it to years if needed).
gender: 1 (women), 2 (men).
height: Height in centimeters.
weight: Weight in kilograms.
ap_hi: Systolic blood pressure.
ap_lo: Diastolic blood pressure.
cholesterol: 1 (normal), 2 (above normal), 3 (well above normal).
gluc: Glucose level (1, 2, 3 with similar meaning as cholesterol).
smoke: Binary (0 if not smoking, 1 if smoking).
alco: Binary (0 if not an alcoholic, 1 if alcoholic).
active: Binary (0 if not physically active, 1 if physically active).
cardio: Binary (0 if no cardiovascular disease, 1 if it is present).
1. A Quick Look at Matplotlib and Pandas¶
Although our main focus is Altair, let's briefly look at some older methods to understand the evolution of plotting libraries in Python.
# Simple histogram of 'weight' using matplotlib
plt.figure(figsize=(8, 4))
plt.hist(df['weight'], bins=20, color='lightblue', edgecolor='black')
plt.title('Weight Distribution (Matplotlib)')
plt.xlabel('Weight (kg)')
plt.ylabel('Frequency')
plt.show()
Pandas can also do quick visualizations since it integrates with matplotlib under the hood.
df['height'].plot(kind='box', title='Height Box Plot (Pandas)', ylabel='Height (cm)')
plt.show()
As you can see, matplotlib and pandas are powerful but can sometimes be verbose or less straightforward for interactive or layered visualizations.
2. Introduction to Altair¶
Altair is a declarative statistical visualization library. Here are some key ideas before we start:
Data: The data source (typically a pandas DataFrame).
Mark: The basic graphical shape (like
mark_bar(),mark_circle(), etc.).Encoding: A mapping between your data columns and visual properties (e.g., color, size, and position).
Data Types: Altair uses short type codes when encoding data:
Q: Quantitative (numeric values, e.g.,
height,weight).T: Temporal (time or date).
O: Ordinal (data with an order, e.g., a numeric range grouped in bins).
N: Nominal (categorical data with no intrinsic order, e.g.,
gender).
Let’s start very simply. We'll make a bar chart counting how many patients fall under each gender.
# A minimal Altair chart:
basic_1dpoints = alt.Chart(df).mark_point().encode(
# We can specify just the field names if we don't need special config yet.
x='height',
)
basic_1dpoints
Basic, right? We just plotted the height column as a 1D scatter plot by mapping it to the x-axis.
Now, let's create a bar chart counting the number of patients for each gender.
basic_bar = alt.Chart(df).mark_bar().encode(
# For a bar chart, if we only mention one dimension, Altair automatically does a 'count()' on the y-axis.
x='gender',
y='count()'
)
basic_bar
How does this work?
alt.Chart(df): We create a new Chart object using our DataFrame..mark_bar(): We want bar marks..encode(x='gender', y='count()'): We encode thegendercolumn on the x-axis.By using
count(), the bar’s height represents the number of rows for eachgender.
Note that we didn’t specify any data types (N, Q, etc.) explicitly. In many cases, Altair can infer them. However, let's be explicit to demonstrate best practices.
# Let's do the same chart but specifying types:
basic_bar_explicit = alt.Chart(df).mark_bar().encode(
x='gender:N', # Treat gender as Nominal (categorical)
y='count()' # A built-in aggregation for counting rows
)
basic_bar_explicit
3. Understanding Data Types in Altair¶
Altair recognizes several types of data and uses these type codes within the encoding channels:
Quantitative (Q): Numeric values, such as
height,weight, or any continuous measurement.Temporal (T): Time or date values, often used for time-series data.
Ordinal (O): Data with a logical order or ranking (e.g., small < medium < large).
Nominal (N): Categorical data with no intrinsic order (e.g.,
gender,cholesterollevels if treated as categories).
By default, Altair often infers these data types, but being explicit can help avoid confusion—especially when customizing how data is displayed.
Let’s explore a histogram of the weight column using Altair. Because weight is quantitative (Q), we can bin the values to see their distribution.
hist_weight = alt.Chart(df).mark_bar().encode(
alt.X('weight:Q', bin=True, title='Weight (kg)'), # Bin the weight values
alt.Y('count()', title='Count of Patients')
).properties(
width=400,
height=200,
title='Histogram of Weight'
)
hist_weight
4. Composing plots¶
Altair allows us to combine multiple plots into a single visualization. We can use the | operator to place them side by side or the & operator to stack them vertically.
Create a boxplot of height for each gender and combine
box_height_gender1 = alt.Chart(df.query("gender == 1")).mark_boxplot().encode(
y='height:Q'
).properties(
width=300,
title='Distribution of Height for Gender 1'
)
box_height_gender2 = alt.Chart(df.query("gender == 2")).mark_boxplot().encode(
y='height:Q'
).properties(
width=300,
title='Distribution of Height for Gender 2'
)
box_height_gender1|box_height_gender2
# ## 5. Customizing Encodings
#
# Altair allows us to map columns to various visual channels such as:
#
# - `alt.X()` and `alt.Y()` for positions on horizontal/vertical axes.
# - `alt.Color()` for color.
# - `alt.Size()` for size.
# - `alt.Opacity()`, `alt.Shape()`, etc.
#
# We can also customize scales, legends, and axis labels.
#
# Let’s try something a bit more interesting: a **box plot** of `height` by `gender`. Since gender is nominal, we want to ensure we specify that, and let Altair do the rest.
#
# ### 5.1 Box Plot of Height Grouped by Gender
box_height_gender = alt.Chart(df).mark_boxplot().encode(
x='gender:N', # Nominal
y='height:Q', # Quantitative
# color="gender:N",
).properties(
width=300,
title='Distribution of Height by Gender'
)
box_height_gender
This gives us a quick statistical summary of the height distribution for each gender.
6. Plotting Multiple Variables: Scatter Plots¶
One of the most common ways to look for relationships between two quantitative variables is through a scatter plot. In our dataset, weight and height are both good examples of continuous (Q) variables.
6.1 Basic Scatter Plot of Weight vs. Height¶
scatter_wh = alt.Chart(df).mark_point().encode(
x='height:Q',
y='weight:Q'
).properties(
width=400,
height=300,
title='Scatter Plot: Weight vs. Height'
)
scatter_wh
This basic scatter plot shows how height (x-axis) relates to weight (y-axis), but it's just a simple cloud of points.
6.2 Adding Color to Represent a Third Variable¶
Often, we want to see how a third variable might influence the relationship between two numeric variables. For example, let's color our points by gender.
genderis categorical, so we use:Nor:O.We can directly specify
alt.Color('gender:N').
scatter_wh_color = alt.Chart(df).mark_point(
filled=True, # Filled circles
size=50 # Larger size
).encode(
x='height:Q',
y='weight:Q',
color='gender:N' # Color points by gender
).properties(
width=400,
height=300,
title='Weight vs. Height Colored by Gender'
)
scatter_wh_color
Now we can see how male and female patients (in this dataset, labeled 2 and 1, respectively) may occupy different regions in the height-weight space.
Tip: If you want to make the legend or color scale more descriptive, you can replace the numeric codes (1 and 2) with actual labels. One way is to create a mapping in your DataFrame before plotting:
gender_map = {1: 'Female', 2: 'Male'}
df['gender_label'] = df['gender'].map(gender_map)
Then encode:
color='gender_label:N'
7. Putting It All Together¶
Below is a final example combining some of these concepts:
Histogram to see the distribution of
height.Scatter plot of
heightvs.weightwith color bygender.We also add some descriptive properties like titles and axis labels.
# Histogram of height (binned)
hist_height = alt.Chart(df).mark_bar().encode(
alt.X('height:Q', bin=True, title='Height (cm)'),
alt.Y('count()', title='Count')
).properties(
width=300,
height=200,
title='Histogram of Height'
)
# Scatter plot of height vs. weight colored by gender
scatter_hw_gender = alt.Chart(df).mark_point().encode(
alt.X('height:Q', title='Height (cm)'),
alt.Y('weight:Q', title='Weight (kg)'),
alt.Color('gender:N', title='Gender')
).properties(
width=300,
height=200,
title='Scatter: Height vs Weight'
)
# Combine the two charts side by side
combined_charts = hist_height | scatter_hw_gender
combined_charts
8. Cholesterol levels vs. Cardiovascular Disease¶
High cholesterol levels have long been associated with an increased risk of heart disease. In this chart, we compare the proportion of cardiovascular disease (cardio) across three categories of cholesterol: Normal, Above Normal, and Well Above Normal.
Create descriptive labels for cholesterol categories
Stacked bar chart showing the proportion of patients with/without CVD in each cholesterol category
cholesterol_cvd_chart = (
alt.Chart(df)
.mark_bar()
.encode(
x=alt.X('cholesterol:O', title='Cholesterol Level'),
# Use stack='normalize' on the y-axis to see proportions instead of raw counts
y=alt.Y('count()', stack='normalize', title='Proportion of Patients'),
color=alt.Color('cardio:N', title='Has CVD?'),
tooltip=['count()'] # Optional: see raw counts on hover
)
.properties(title='Proportion of Cardiovascular Disease by Cholesterol Level')
)
cholesterol_cvd_chart