Skip to content

Introduction to Data Science

Streamlit cheat sheet

Streamlit · 09 — Cheat Sheet¶

One page. Print it, paste it above your editor, keep it next to your monitor.

Running¶

pip install streamlit          # once
streamlit hello                # verify the install
streamlit run app.py           # run your app
streamlit run app.py --server.port 8502   # different port

The mental model: the script re-runs top to bottom on every interaction. Widget = variable. if around widgets controls what shows.

Text¶

Code	Result
`st.title("...")`	page title
`st.header("...")`	section heading
`st.subheader("...")`	sub-heading
`st.write(x)`	smart display — text, numbers, dataframes, anything
`st.markdown("bold it`code`")`	formatted text
`st.caption("small note")`	small grey note
`st.latex("x^2")`	math (optional)

Status messages¶

Code	Looks like
`st.success("...")`	green ✓
`st.info("...")`	blue ℹ
`st.warning("...")`	yellow ⚠
`st.error("...")`	red ✗
`st.stop()`	stops the script here

Input widgets (all return a value)¶

st.button("label")              # True only on the click's rerun
st.checkbox("label")            # True / False
st.selectbox("label", options)  # one chosen item
st.multiselect("label", options, default=[...])  # a LIST
st.slider("label", min_value=0, max_value=100, value=50, step=1)
st.number_input("label", min_value=0, value=0, step=1)
st.text_input("label")          # string, "" when empty
st.text_area("label")           # multi-line string
st.file_uploader("label", type=["csv"])   # file-like or None

Widget extras: unique labels (or key="..."), st.sidebar. prefix for the sidebar, st.columns(n) to lay out side by side.

Data & numbers¶

st.dataframe(df)                # interactive: sort, scroll, search
st.table(df)                    # static table
st.metric("label", value, delta="+3")   # labeled number with arrow
st.columns(3)                   # returns 3 containers: c1, c2, c3
c1.metric(...)                  # use them as c1.metric(...)
st.expander("Show details")     # collapsible section
with st.expander("..."):        #   ...content...

Charts¶

st.line_chart(df)               # built-in, takes a dataframe
st.bar_chart(df)
st.area_chart(df)

# Matplotlib / Seaborn — the reliable pattern:
import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(8, 4))
sns.histplot(df["col"], bins=30, kde=True, ax=ax)   # seaborn: pass ax=ax!
ax.set_title("...")
st.pyplot(fig)

Caching (run expensive work once)¶

@st.cache_data
def load_data():
    return pd.read_csv("big.csv")   # runs once, reused on reruns
df = load_data()

Session state (remember across reruns)¶

if "name" not in st.session_state:
    st.session_state.name = "default"
st.session_state.name = ...        # read / write freely

ML prediction app skeleton¶

@st.cache_data
def train_model():
    ... model.fit(X_train, y_train) ...
    return model

model = train_model()              # trained once
features = pd.DataFrame([[v1, v2]], columns=[...])
pred = model.predict(features)[0]  # cheap — runs every rerun
st.success(f"Predicted: {pred}")

st.sidebar.header("Filters")               # everything in the sidebar
page = st.sidebar.radio("Page", ["Overview", "EDA", "Model"])

if page == "Overview":
    ...                                    # one if-block per page
elif page == "EDA":
    ...

The golden rules¶

Widgets are variables; the script re-runs on every change.
Cache data loading and model training — never widget state.
st.session_state for memory across reruns; st.stop() to bail out gracefully.
Seaborn: always fig, ax = plt.subplots() and ax=ax.
Guard uploads (if uploaded is None: st.stop()), guard empty multiselect (if cols:), guard division by zero.
print() goes to the terminal — use st.write in the app.
Never deploy private data publicly.