Streamlit для аналитика

Зачем это знать

Streamlit позволяет аналитику build interactive web apps без фронтенд-знаний. 30 минут — от Python-скрипта к working dashboard. В startups и data-teams — популярный tool для quick prototypes, internal tools, demo моделей.

Что такое Streamlit

Python library для web apps:

• Declarative API
• No HTML/CSS/JS нужно
• Real-time reactive
• Deploy easy

pip install streamlit
streamlit run app.py

Hello world

# app.py
import streamlit as st
import pandas as pd

st.title("Мой первый dashboard")
st.write("Здесь будет аналитика")

df = pd.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6]})
st.dataframe(df)
st.line_chart(df)

streamlit run app.py → opens browser с app.

Widgets

Interactive inputs:

# Slider
age = st.slider("Your age", 0, 100, 25)

# Selectbox
color = st.selectbox("Pick color", ["red", "green", "blue"])

# Date input
date = st.date_input("Select date")

# File upload
file = st.file_uploader("Upload CSV")

# Multi-select
countries = st.multiselect("Countries", ["RU", "US", "KZ"])

# Button
if st.button("Run analysis"):
    st.write("Analyzing...")

Layout

Columns

col1, col2 = st.columns(2)
with col1:
    st.metric("Revenue", "$1M", "+10%")
with col2:
    st.metric("Users", "5k", "+5%")

Sidebar

with st.sidebar:
    filter_country = st.selectbox("Country", options)

Filters в sidebar — classic.

Tabs

tab1, tab2 = st.tabs(["Metrics", "Details"])
with tab1:
    st.write("Metrics here")
with tab2:
    st.write("Details here")

Caching

Для slow queries — @st.cache_data:

@st.cache_data
def load_data():
    return pd.read_csv("huge_file.csv")

df = load_data()  # первый запуск — загружает
# Subsequent runs — cached

Critical для performance.

Typical dashboard

import streamlit as st
import pandas as pd
import plotly.express as px

st.set_page_config(page_title="Sales Dashboard", layout="wide")

st.title("📊 Sales Dashboard")

# Sidebar filters
country = st.sidebar.selectbox("Country", ["All", "RU", "US", "KZ"])

# Load data (cached)
@st.cache_data
def load_data():
    return pd.read_csv("sales.csv")

df = load_data()

if country != "All":
    df = df[df['country'] == country]

# Metrics row
col1, col2, col3 = st.columns(3)
col1.metric("Revenue", f"${df['revenue'].sum():,.0f}")
col2.metric("Orders", f"{df['orders'].sum():,}")
col3.metric("AOV", f"${df['revenue'].sum() / df['orders'].sum():.0f}")

# Chart
fig = px.line(df.groupby('date')['revenue'].sum().reset_index(),
              x='date', y='revenue')
st.plotly_chart(fig, use_container_width=True)

# Table
st.dataframe(df.head(100))

30 строк = работающий dashboard.

ML model interface

Для demo models:

import streamlit as st
import pickle

model = pickle.load(open("model.pkl", "rb"))

st.title("Churn prediction")

age = st.slider("Age", 18, 100, 30)
tenure = st.number_input("Tenure (months)", 0, 100, 12)
payments = st.number_input("Payments count", 0, 50, 5)

if st.button("Predict"):
    features = [[age, tenure, payments]]
    prob = model.predict_proba(features)[0][1]
    st.write(f"Churn probability: {prob:.2%}")

Stakeholders играются, understand model.

State management

# Session state
if 'count' not in st.session_state:
    st.session_state.count = 0

if st.button("Increment"):
    st.session_state.count += 1

st.write(f"Count: {st.session_state.count}")

Persist data across reruns.

Deploy

Streamlit Cloud (free)

Connect к GitHub repo → auto-deploy. 1 public app free.

Internal

Docker + cloud:

FROM python:3.10
RUN pip install streamlit pandas
COPY app.py .
CMD ["streamlit", "run", "app.py"]

Use cases

1. Internal tools

Ad-hoc queries с filters. Non-technical users.

2. ML demo

Show model to stakeholders.

3. Prototype

Before full BI investment.

4. Portfolio

Deploy your project — shareable.

5. Experiments

A/B analysis with custom logic.

Streamlit vs альтернативы

Streamlit

• Simplest Python
• Quick prototyping
• Single file

Dash (Plotly)

• More flexible
• Callback-based
• Heavier

Gradio

• ML model-focused
• Simpler than Streamlit для ML
• Less general-purpose

Panel

• Holoviews ecosystem
• Flexible
• Steeper curve

Streamlit — best starting point.

Частые ошибки

Heavy load без caching

Query runs на каждый widget change → slow.

Always @st.cache_data для heavy queries.

Too many widgets

10+ filters → confusion. Prioritize.

Production claims

Streamlit — prototype. Для production SaaS — proper frontend.

No error handling

User uploads wrong file format → app crashes.

try:
    df = pd.read_csv(file)
except:
    st.error("Invalid file format")

Limitations

• Single user: не multi-tenant
• No auth built-in: use wrapper (Streamlit Cloud has)
• Performance: limited на huge datasets
• Fancy UI: not as polished as dedicated frontend

Но для internal analytics — хватает.

На собесе

«Как поделились бы anal?»

Options:

1. Jupyter notebook
2. Streamlit app
3. BI tool (Tableau, Metabase)
4. Email report

Streamlit — когда нужен interactivity without BI invest.

Связанные темы

• Data visualization в Python
• Dashboard design принципы
• Как автоматизировать отчёт

FAQ

Free?

Open-source + Streamlit Cloud free tier.

Production?

Internal tools — ok. Customer-facing — consider proper frontend.

Mobile?

Works, но не mobile-optimized.

---

Тренируйте навыки — откройте тренажёр с 1500+ вопросами для собесов.