matplotlib seaborn projects part 2

Matplotlib & Seaborn Projects

Phase 1 → Thinking Like a Programmer → Exploratory Data Analysis

Welcome to the Matplotlib & Seaborn Projects notebook.

This notebook is a project-based continuation of the Exploratory Data Analysis phase. Instead of learning plotting functions in isolation, you will use visualization to investigate real analytical questions.

The goal is not:

"Can I make a chart?"

The goal is:

"Can I choose, construct, interpret, and communicate a chart that helps answer a question?"

You will work through several progressively harder projects covering:

• Matplotlib fundamentals
• Seaborn fundamentals
• categorical analysis
• distributions
• relationships between variables
• correlation
• time-series visualization
• subgroups and segmentation
• outlier investigation
• multi-variable visualization
• dashboard-style EDA
• storytelling with data
• responsible interpretation

Learning objectives

By the end of this notebook, you should be able to:

1. Create publication-quality visualizations with Matplotlib.
2. Use Seaborn for statistical exploratory visualizations.
3. Select an appropriate chart for an analytical question.
4. Explain what a visualization shows without overclaiming.
5. Compare numerical distributions across categories.
6. Investigate relationships between numerical variables.
7. Visualize correlations.
8. Analyze trends over time.
9. Detect and investigate outliers visually.
10. Use color, size, position, and faceting deliberately.
11. Build a coherent EDA visualization workflow.
12. Communicate findings in Markdown rather than relying on charts alone.

Core principle

A visualization is an analytical tool, not decoration.

For every chart in this notebook, answer:

1. What question am I asking?
2. Why is this chart appropriate?
3. What pattern do I observe?
4. What does that pattern potentially mean?
5. What can I not conclude from this chart?
6. What would I investigate next?

Part 1 - Visualization Foundations

Before beginning the projects, establish a mental model for visualization.

The visual encoding hierarchy

Different visual properties communicate information with different levels of precision.

Generally, viewers are better at comparing:

1. position on a common scale
2. position on different scales
3. length
4. angle
5. area
6. color intensity
7. shape

This is one reason why a carefully designed bar chart is often easier to interpret than a pie chart.

Chart selection cheat sheet

Project 1 - Your First Analytical Chart

Scenario

You are a junior data analyst at an e-commerce company.

The product team asks:

Which products generate the most revenue?

Create a chart that answers the question clearly.

We will start with a small synthetic dataset so that the analytical logic is easy to understand.

Task 1.1 - Create a basic bar chart

Requirements:

• use Matplotlib
• put products on the x-axis
• put revenue on the y-axis
• add a descriptive title
• label both axes
• make the chart readable
• call plt.tight_layout()

Questions

Before running your code, decide:

• Should the bars be sorted?
• Should the y-axis have a currency label?
• Is the product order meaningful?

Reflection

Write a short Markdown response:

• Which product generated the most revenue?
• Which generated the least?
• Is the difference between the top and bottom products substantial?
• What business question would you investigate next?

Remember: the chart establishes an observed pattern. It does not tell you why one product generated more revenue.

Project 2 - Sorting Makes Comparisons Easier

A chart can contain correct data and still be difficult to read.

Consider what happens when categories are presented in arbitrary order.

Your goal is to create a ranked bar chart.

Requirements

1. Sort products from highest revenue to lowest revenue.
2. Create a horizontal bar chart using plt.barh().
3. Put the highest-revenue product at the top.
4. Format the chart so that the labels are easy to read.
5. Add an informative title.

Why horizontal?

Product names can be long. Horizontal bars give category labels more room.

Reflection

Explain why the horizontal version is easier or harder to interpret.

Then answer:

What visual design decision had the biggest impact on readability?

Project 3 - Understanding Distributions

A single average can hide a great deal of information.

Suppose the business reports:

"The average order value is ₹4,800."

That sounds useful, but it does not tell us whether most orders are around ₹4,800 or whether a small number of very large orders pull the average upward.

We need a distribution.

Dataset

Generate synthetic order values with a realistic right-skewed distribution.

Task 3.1 - Histogram

Create a histogram showing the distribution of order values.

Requirements:

• use Matplotlib
• experiment with different numbers of bins
• label the axes
• include units
• give the chart a meaningful title

Try at least:

• 10 bins
• 30 bins
• 60 bins

Then decide which communicates the distribution best.

Task 3.2 - Mean vs median

Calculate:

• mean order value
• median order value
• standard deviation
• 90th percentile
• 99th percentile

Then explain why the mean and median differ.

Use:

orders["order_value"].mean()
orders["order_value"].median()
orders["order_value"].std()
orders["order_value"].quantile(0.90)
orders["order_value"].quantile(0.99)

Task 3.3 - Add a KDE

Use Seaborn's histplot() with kde=True.

Then answer:

• What does the KDE add?
• Does it make the distribution easier or harder to interpret?
• When might a KDE be misleading?

Remember:

A KDE is an estimate of a distribution, not a collection of observed values.

Project 4 - Outlier Investigation

Outliers are not automatically mistakes.

They can represent:

• genuine high-value customers
• unusual events
• fraud
• measurement errors
• data-entry errors
• rare but important observations

Your job is to investigate rather than automatically delete.

Task 4.1 - Box plot

Create a horizontal box plot of order_value.

Task 4.2 - Identify extreme orders

Use a quantile-based rule to identify unusually high-value orders.

One simple approach is to inspect the 99th percentile.

Then compare:

• number of orders above the 99th percentile
• percentage of all orders
• total revenue contributed by those orders

Important

This is an investigation threshold, not a universal definition of an error.

Reflection

Write:

"The extreme orders account for approximately _% of transactions but _% of total order value."

Then explain what this means for the business.

Would it be appropriate to report only the mean order value? Why or why not?

Project 5 - Relationships Between Variables

Now move from one variable to two.

The business asks:

Do customers with higher incomes tend to spend more?

A scatter plot is an appropriate starting point.

Dataset

Generate a synthetic customer dataset.

Task 5.1 - Scatter plot

Create a scatter plot with:

• income on x
• annual spending on y
• informative labels
• an informative title
• reasonable figure size

Use Seaborn.

Task 5.2 - Correlation

Calculate the Pearson correlation between income and annual spending.

Then answer:

1. Is the relationship positive or negative?
2. Is it weak, moderate, or strong?
3. Does correlation prove that income causes spending?
4. What other variables might matter?

Important lesson: correlation is not causation

A scatter plot can reveal association.

It cannot, by itself, establish:

• causality
• direction of causality
• absence of confounding variables
• whether the relationship generalizes beyond this dataset

Good analysts separate:

What the data shows

from:

What they think might explain it.

Project 6 - Add a Third Variable

Suppose the relationship between income and spending differs by age group.

Create an age_group column:

• 18–29 → Young Adult
• 30–44 → Adult
• 45–59 → Midlife
• 60+ → Senior

Then visualize income vs spending with age group represented by color.

Reflection

Look for:

• differences in location
• differences in spread
• clusters
• unusual observations
• whether one group appears to follow a different relationship

Then answer:

Does the overall income-spending relationship appear identical across all age groups?

Be cautious. Visual separation is evidence for further investigation, not proof of a causal mechanism.

Project 7 - Categorical Distributions

A product manager asks:

Do customers in different cities have different spending distributions?

A simple average by city may hide important differences.

We will compare distributions instead.

Task 7.1 - Box plot by city

Create a Seaborn box plot.

Requirements:

• city on x
• spending on y
• readable category labels
• descriptive title

Then compare:

• median
• spread
• potential outliers

Task 7.2 - Violin plot

Create a violin plot for the same data.

Compare the box plot and violin plot.

Question

What does the violin plot communicate that the box plot communicates less directly?

Consider:

• distribution shape
• density
• multiple peaks
• tails

Project 8 - Count Plots and Category Imbalance

Not every EDA problem is about numerical variables.

Suppose a company wants to understand the distribution of customers across cities.

Create a count plot.

Reflection

Answer:

1. Which city has the most observations?
2. Which city has the fewest?
3. Could unequal sample sizes affect comparisons between cities?
4. Why might comparing averages alone be dangerous when group sizes differ substantially?

This is an important analytical lesson:

A group with 100 observations and a group with 100,000 observations do not provide the same amount of evidence.

Project 9 - Time-Series Visualization

Time is one of the most common dimensions in real-world data.

The business asks:

How has monthly revenue changed over time?

Create a synthetic daily transaction dataset for two years.

Task 9.1 - Daily revenue

Plot daily revenue over time.

Then answer:

• Is there a long-term trend?
• Is there short-term noise?
• Is there visible seasonality?
• Is the daily chart too noisy to communicate the broader trend?

Task 9.2 - Aggregate to monthly revenue

Use Pandas to aggregate the daily data to month level.

Then plot monthly revenue.

This is an important EDA concept:

The right aggregation level can make a pattern easier to see.

Reflection

Compare daily and monthly charts.

Which is better for answering:

"Is the business growing?"

Which is better for answering:

"What happened on a particular day?"

There is no universally best chart. The correct visualization depends on the question.

Project 10 - Rolling Averages

Daily data can be noisy.

A rolling average can help reveal an underlying trend.

Calculate a 30-day rolling mean.

Interpretation

Explain:

• what the rolling average smooths away
• what information it preserves
• what information it hides
• why smoothing can be useful
• why smoothing can also be dangerous

A smoothed chart should complement the raw data, not automatically replace it.

Project 11 - Correlation Heatmap

Now investigate relationships among multiple numerical variables.

Create a richer customer dataset with:

• age
• income
• spending
• purchase frequency
• average order value
• discount usage
• customer tenure

Then calculate a correlation matrix.

Task 11.1 - Correlation matrix

Calculate correlations among all numerical variables.

Then visualize them using a Seaborn heatmap.

Requirements:

• annotate correlation values
• use a clear title
• make the figure large enough to read

Interpretation exercise

Find:

1. the strongest positive relationship
2. the strongest negative relationship
3. a relationship close to zero

For each, write one sentence describing what the correlation suggests.

Then write:

"Correlation does not establish causation because..."

Complete the sentence in your own words.

Project 12 - Pair Plot Investigation

A pair plot can provide a fast overview of relationships among a small set of variables.

Select:

• income
• spending
• purchase frequency
• average order value

Create a pair plot.

Why pair plots should be used carefully

Pair plots are useful for exploration, but they have limitations:

• they become unreadable with many variables
• large datasets can produce dense plots
• correlation may be visually difficult to assess
• categorical structure may be hidden
• they encourage broad exploration rather than focused questions

Use them as an investigative tool, not as a substitute for thoughtful analysis.

Project 13 - Faceting

Suppose the relationship between income and spending may differ by customer segment.

Create a customer_segment variable based on income.

Then use Seaborn's FacetGrid or relplot() to create separate scatter plots for each segment.

Reflection

Compare the faceted plots.

Ask:

• Does the relationship look consistent across groups?
• Does the spread change?
• Are there different outliers?
• Does segmentation reveal a pattern that the single overall scatter plot hides?

This is an example of how stratification can reveal structure hidden in aggregate data.

Project 14 - Build a Mini EDA Dashboard

Now combine multiple visualizations into a coherent analytical story.

Scenario

You are presenting to a product manager who has five minutes.

They want to know:

1. What does customer spending look like?
2. Which city has the highest typical spending?
3. Is income associated with spending?
4. How large is the customer base by city?
5. Are there unusual high-value customers?

Create a set of five separate figures, one for each question.

Do not put everything into one overcrowded chart.

Required charts

Figure 1

Spending distribution.

Figure 2

Spending by city.

Figure 3

Income vs spending.

Figure 4

Customer count by city.

Figure 5

Outlier-focused spending visualization.

For every figure, include:

• title
• axis labels
• units
• appropriate scale
• readable categories
• a short Markdown interpretation

Project 15 - Visualization Critique

A strong analyst should be able to critique visualizations, not merely create them.

For each chart below, imagine that it has been presented to a stakeholder.

Evaluate:

• Is the chart appropriate?
• Is the scale appropriate?
• Is the title informative?
• Are the units clear?
• Is the comparison easy?
• Could the chart mislead?
• What alternative chart might be better?

Critique A

A pie chart with 25 categories showing product revenue.

Questions:

• Is a pie chart appropriate?
• Can viewers accurately compare many slices?
• Would a sorted bar chart be better?

Critique B

A line chart with 10,000 daily observations and no aggregation.

Questions:

• Is the chart too noisy?
• Would weekly or monthly aggregation help?
• Should the raw series still be available?

Critique C

A bar chart with a y-axis starting at 95 instead of 0, showing values of 96, 97, and 98.

Questions:

• Does the truncated axis exaggerate differences?
• When can a non-zero baseline be appropriate?
• What should be communicated explicitly?

Critique D

A scatter plot containing 2 million points.

Questions:

• Will overplotting hide patterns?
• Could sampling help?
• Could transparency help?
• Could hexbin or density visualization be more appropriate?

Project 16 - Tell a Data Story

This is the most important project in the notebook.

You will produce an analytical narrative rather than a collection of charts.

Your task

Choose one dataset you care about.

Possible themes:

• movies
• sports
• music
• gaming
• finance
• e-commerce
• education
• climate
• transportation
• housing
• public policy
• technology
• social media

Define one central question.

Example:

What factors are associated with higher movie ratings?

Then build a visualization story.

Required structure

1. Question

State the question in one or two sentences.

2. Dataset

Explain:

• source
• observations
• variables
• time period
• known limitations

3. Data quality

Investigate:

• missing values
• duplicates
• invalid values
• inconsistent categories
• suspicious outliers

4. Univariate exploration

At least two variables.

5. Bivariate exploration

At least two relationships.

6. Multivariate exploration

At least one analysis involving three or more variables.

7. Visualization

Create at least five meaningful charts.

8. Findings

Write at least five findings.

9. Limitations

Write at least three limitations.

10. Next questions

Write at least five questions that would be worth investigating next.

Project 17 - Advanced Challenge: Reproduce a Visualization

Find an interesting visualization online or in a public report.

Do not copy the underlying data blindly.

Instead:

1. identify the analytical question
2. identify the chart type
3. identify the variables
4. identify the visual encodings
5. recreate the visualization using Python
6. improve one aspect of it
7. explain your design choices

Your goal is to learn visualization by reverse-engineering good examples.

Questions to answer

• Why did the original author choose this chart?
• What is the most important visual element?
• How is attention directed?
• What would make the visualization easier to understand?
• What information could be lost?

Project 18 - Advanced Challenge: Make a Bad Chart Better

Create an intentionally poor chart.

Examples:

• too many categories
• unclear title
• missing units
• unreadable labels
• excessive clutter
• poor chart type
• misleading scale

Then redesign it.

Your notebook should contain:

1. the bad chart
2. an explanation of what is wrong
3. the improved chart
4. an explanation of each improvement

This teaches an important skill:

Visualization is communication design.

Project 19 - Analytical Writing Challenge

For each major chart in your final project, write three layers of interpretation.

Layer 1 - Observation

What can we directly see?

Example:

"Revenue is highest in the fourth quarter."

Layer 2 - Interpretation

What might explain the pattern?

Example:

"This may indicate seasonal demand."

Layer 3 - Limitation

What can we not conclude?

Example:

"The chart alone cannot determine whether seasonality caused the increase."

This three-layer structure helps prevent overclaiming.

Project 20 - Final Capstone

Build an EDA Visualization Portfolio

Create a final notebook using a real dataset.

Your submission should demonstrate all major skills from this notebook.

Required analytical components

A. Dataset overview

Show:

• shape
• columns
• data types
• sample rows
• descriptive statistics

B. Data-quality analysis

Show:

• missing values
• duplicates
• categorical consistency
• suspicious numerical values

C. Univariate analysis

Include:

• at least one histogram
• at least one count plot
• at least one box plot

D. Bivariate analysis

Include:

• at least one scatter plot
• at least one categorical comparison

E. Multivariate analysis

Include at least one:

• heatmap
• faceted chart
• hue-based visualization
• pair plot

F. Time analysis

If your dataset contains dates, include:

• a time-series chart
• an appropriate aggregation
• interpretation of trends

If your dataset has no dates, explain why time-series analysis is not applicable.

G. Outlier analysis

Identify and investigate at least one unusual observation or group.

H. Findings

Write at least five meaningful findings.

I. Limitations

Write at least three limitations.

J. Next steps

Write at least five follow-up questions.

Final Assessment Rubric

Use this rubric before considering the project complete.

Final Checklist

Before submitting your notebook, verify:

Code

• [ ] All imports are included.
• [ ] Code runs from top to bottom.
• [ ] Random operations use a fixed seed where appropriate.
• [ ] Variables have meaningful names.
• [ ] Repeated logic is avoided where possible.
• [ ] Charts render correctly.

Data

• [ ] You understand what each row represents.
• [ ] You understand the important variables.
• [ ] Missing data has been investigated.
• [ ] Duplicates have been investigated.
• [ ] Outliers have been investigated.
• [ ] Categories have been checked.
• [ ] Dates have been parsed correctly when applicable.

Visualization

• [ ] Every chart answers a question.
• [ ] Every chart has a descriptive title.
• [ ] Axes are labeled.
• [ ] Units are explicit.
• [ ] Categories are readable.
• [ ] Scales are appropriate.
• [ ] Visual clutter is minimized.
• [ ] Colors have a purpose.
• [ ] You avoid unnecessary decoration.

Interpretation

• [ ] You distinguish observation from interpretation.
• [ ] You do not claim causation from correlation alone.
• [ ] You discuss important limitations.
• [ ] You identify potential confounders.
• [ ] You identify useful follow-up questions.

Communication

• [ ] Someone unfamiliar with your code can understand the story.
• [ ] Markdown explains why each analysis exists.
• [ ] Findings are written in plain language.
• [ ] The conclusion answers the original question.
• [ ] The notebook ends with clear next steps.

Final Challenge: Think Like an Analyst

Before moving to the next phase, complete this exercise without writing code.

You receive a dataset containing:

customer_id
age
city
income
product
quantity
price
purchase_date

A stakeholder asks:

"Tell me something interesting about our customers."

Do not immediately make a chart.

Write down:

Step 1 - Questions

What would you want to know?

Step 2 - Assumptions

What do you need to understand about the dataset?

Step 3 - Quality checks

What could be wrong?

Step 4 - Variables

Which variables might be useful?

Step 5 - Visualizations

Which charts would help answer your questions?

Step 6 - Interpretation

What would count as a meaningful finding?

Step 7 - Next steps

What would you investigate after discovering an interesting pattern?

This exercise is the real objective of Phase 1.

What You Should Take Away

Matplotlib and Seaborn are not the skill.

The skill is visual reasoning.

You should now be moving from:

"I know how to make a scatter plot."

to:

"I know when a scatter plot is useful."

Then from:

"I found a correlation."

to:

"I found an association, and I understand what it does and does not establish."

And finally from:

"Here are five charts."

to:

"Here is a question, here is the evidence, here is what the evidence suggests, here are the limitations, and here is what I would investigate next."

That is the foundation of strong Exploratory Data Analysis.

Next Phase

Once you can independently investigate a dataset with NumPy, Pandas, Matplotlib, and Seaborn, you are ready to move from:

Exploring data

toward:

Using data to answer questions systematically.

The next phase should build on this foundation rather than skipping it.

Do not rush past EDA.

The ability to understand data before modeling it is one of the most important skills in the entire data-science and AI workflow.