Skip to content

Streamlit Lab 3 — Interactive EDA Dashboard

Week: Module C (Streamlit module) · CLO-1, CLO-3 · Time: ~75 min

Learning objectives

  • Build matplotlib/seaborn figures correctly for Streamlit (fig, ax = plt.subplots(), ax=ax, st.pyplot(fig)).
  • Make charts respond to widgets (selectbox, slider, checkbox).
  • Apply a global filter that updates every chart.

Problem statement

Build one app (app.py) on the built-in penguins dataset (sns.load_dataset("penguins")):

  1. Overview: 3 metrics (rows, missing values, mean body mass).
  2. Interactive histogram of any numeric column (selectbox).
  3. Boxplot of body mass by species (fixed) and scatter of bill length vs bill depth (fixed), both responding to a global filter.
  4. Global filter: checkbox "Drop rows with missing values" (penguins has 2 rows with missing values!) plus a slider "Minimum body mass".

Tasks

  1. Create streamlit-lab3/app.py.
  2. Load penguins once (cache it — @st.cache_data).
  3. Metrics: rows, missing values (df.isna().sum().sum()), mean body mass (f"{df['body_mass_g'].mean():,.0f} g").
  4. Histogram: selectbox of numeric columns → sns.histplot(..., kde=True, ax=ax) → st.pyplot(fig).
  5. Fixed charts: boxplot of body_mass_g by species; scatter of bill_length_mm vs bill_depth_mm (sns.scatterplot(data=df, x="bill_length_mm", y="bill_depth_mm", hue="species", ax=ax)).
  6. Global filter (applied before any chart): 
    drop_missing = st.checkbox("Drop rows with missing values")
min_mass = st.slider("Minimum body mass (g)", 2700, 6300, 2700)
if drop_missing:
    df = df.dropna()
df = df[df["body_mass_g"] >= min_mass]
  7. Verify the missing-value checkbox changes the histogram and the boxplot and scatter together.

Expected output (verify all)

  • Metrics: 344 rows, 2 missing values, mean body mass ≈ 4201 g (before filtering).
  • Histogram redraws when switching columns.
  • Checking "Drop rows with missing values" → missing count becomes 0 and every chart rebuilds.
  • Moving the slider upward removes the smallest penguins from all charts.

Questions to answer in your submission:

  1. Why does every chart update when you change a single global filter?
  2. What would happen if you forgot ax=ax in the seaborn calls?

Challenge

Add a correlation heatmap page using st.tabs or a radio: one tab for the charts above, one tab for a heatmap of numeric correlations (sns.heatmap(df.select_dtypes("number").corr(), annot=True, ax=ax)) that also respects the global filter.