23 апреля 2026 г.

Шпаргалка BI и визуализации

Карьерник — квиз-тренажёр в Telegram с 1500+ вопросами для собесов аналитика. SQL, Python, A/B, метрики. Бесплатно.

Зачем это знать

Visualization — critical skill аналитика. Bad chart → miscommunicated insights. Good chart → decision made.

Шпаргалка — chart choice, tools, design principles.

Выбор chart

Trend

  • Line chart — default для time series

Compare categories

  • Bar chart (horizontal) — long labels
  • Bar chart (vertical) — short labels
  • Stacked bar — part-to-whole + comparison

Composition

  • Donut / pie — few slices (< 5)
  • Stacked bar — more parts
  • Treemap — hierarchical

Correlation

  • Scatter plot — two numeric
  • Bubble chart — three numeric
  • Heatmap — matrix correlation

Distribution

  • Histogram — single variable
  • Box plot — summary + outliers
  • Violin plot — density + summary
  • Density plot

Flows

  • Sankey — flow between states
  • Waterfall — additive/subtractive breakdown

Geo

  • Choropleth — region colors
  • Bubble map — locations

KPI

  • Big number — key metric
  • Sparkline — inline trend
  • Gauge — progress toward target

Relationships

  • Network graph — connections

Chart types to AVOID

3D

Never useful. Distorts. Hard read.

Pie с > 7 slices

Unreadable.

Stacked area с many series

Confusing bottoms.

Dual-axis

Misleading unless сarefully.

Rainbow colormap

Not colorblind-safe. Use sequential.

Color

Principles

  • Meaningful: red/green = bad/good
  • Accessible: colorblind-safe
  • Minimal: 2-3 main colors
  • Consistent: same within report

Palettes

  • Sequential: single color, gradient. Magnitude.
  • Diverging: two-color gradient. + / - / neutral.
  • Categorical: distinct colors. Categories.

Tools

  • ColorBrewer (designer-chosen)
  • Paul Tol colors (colorblind-safe)

Design principles

Data-ink ratio

Max data, min chartjunk.

Remove:

  • Unnecessary gridlines
  • Excessive labels
  • Decorative borders
  • 3D effects
  • Shadows

Tell story

Chart title = conclusion («Revenue up 15% YoY»), не «Revenue over time».

Label directly

Avoid legends. Label lines / bars directly.

Highlight focus

Bold / color main message. Grey rest.

Consistent scales

Comparing subplots — same Y-axis scale.

Dashboard design

Hierarchy

  • Top: primary KPIs (hero numbers)
  • Middle: trend / comparison charts
  • Bottom: detail / breakdowns

Layout

Grid-based. Consistent widths. Align edges.

Filters

Date, segment — affect все charts.

Interactivity

  • Tooltips hover
  • Drill-down click
  • Date brushing

Update

Show «last updated» timestamp.

BI tools

Enterprise

  • Tableau — polished, expensive
  • Power BI — Microsoft, enterprise
  • Looker — LookML, Google
  • Qlik — associative

Open-source / cheap

  • Metabase — free, simple
  • Superset — Apache, feature-rich
  • Redash — queries + dashboards

Cloud / specialized

  • Mode — SQL + Python + dashboards
  • Hex — notebooks + dashboards
  • Observable — JavaScript-based
  • DataLens — Yandex

Free Google

  • Looker Studio — former Data Studio

Python libraries

matplotlib

Foundation. Low-level control.

seaborn

Statistical, pretty defaults. Built on matplotlib.

plotly

Interactive. HTML export.

bokeh

Alternative interactive.

altair

Declarative grammar-of-graphics.

plotnine

ggplot2 port.

Choose library

Quick EDA

seaborn / pandas

Publication

matplotlib

Interactive web

plotly / bokeh

Dashboard

Streamlit / Dash

Common mistakes

Wrong chart

Pie chart 20 slices → bar chart.

Misleading axes

Truncated Y-axis exaggerates.

Too much

30 metrics на one dashboard → overwhelmed.

No context

«5%» without comparison — meaningless.

Bad labels

«metric_val_v2» vs «Revenue (Q2 2026)».

Ignoring outliers

Mean skewed, show distribution.

Color overuse

10 colors rainbow → noise.

Specific tips

Bar chart

  • Sort by value (unless ordering matters)
  • Horizontal для > 5 categories
  • Start Y-axis at 0

Line chart

  • No markers if many points
  • Different line styles для color-blind
  • Annotate important points

Scatter

  • Don't use when too dense → alpha / hex bins
  • Add regression line if correlation matters

Histogram

  • Right bin count (rule of thumb √N or Freedman-Diaconis)
  • Clear axis labels

Dashboard examples

Executive

Hero numbers + trends + key segmentation. 5-8 items.

Operational

Real-time metrics, alerts, drill-downs.

Analytical

Exploratory, flexible filters, many views.

Customer-facing

Simplified, clear, self-serve.

Data storytelling

Structure

  1. Context (setup)
  2. Conflict (what changed)
  3. Resolution (what next)

Lead with insight

Don't build up. Start with conclusion.

Supporting data

Evidence for insight.

Recommendation

«So what?» — actionable next.

Tools comparison

Tableau Power BI Metabase Looker
Cost $$ $ Free $$$$
Polish High Medium Medium High
Ease Medium Medium Easy Hard
Enterprise Yes Yes Some Yes

На собесе

«Какой chart для X?» Map к task type.

«Dashboard для CEO?» Hero numbers + trends. 3-5 items.

«Какой tool?» Depends на task: quick exploration (Python), polished dashboard (Tableau), self-serve (Metabase).

«Best practices?» Data-ink ratio, meaningful color, clear labels.

Связанные темы

FAQ

Best book визуализация?

  • Edward Tufte classics
  • Cole Nussbaumer «Storytelling with Data»

Как practice?

Rebuild existing dashboards. Study good examples (FiveThirtyEight, NYT).

Tool должен знать?

Industry-standard (Tableau) + free (Metabase). Добавить по ситуации.

Тренируйте — откройте тренажёр с 1500+ вопросами для собесов.

Тренироваться в Telegram