---
title: Introduction
description:
duration: 300
card_type: cue_card
---

### **Introduction to Data Visualization** (5-10 minutes)

#### Plots Presentation:

<https://docs.google.com/presentation/d/1DkLTjTe6YmGbDHtr4v9Jso553DlCuP3cfSnwvUN1mgE/edit?usp=sharing>

### Summary/Agenda

#### Where is all Data Visualization helpful? Why?

- Exploratory - EDA
- Explanatory - Storytelling

#### What is the Science in Data Visualization?

- Anatomy of a plot/chart
- How to use the right plot/chart for given data?

#### What is the Art in Data Visualization?

- Choose the right scale, labels, tick labels
- Identify and remove clutters in the plot
- Ways to highlight information in the plot

---
title: Libraries for data visualization
description:
duration: 700
card_type: cue_card
---

### **Python libraries for Data Visualization**

#### Importing Matplotlib and Seaborn

We don't need to import the entire library but just its submodule `pyplot`

We'll use the **alias name `plt`**

#### What is `pyplot`?

- `pyplot` is a **sub-module for visualization** in `matplotlib`
- Think of it as **high-level API** which **makes plotting an easy
task**
- Data Scientists **stick to using `pyplot` only unless** they want to
create **something totally new**.

For seaborn, we will be importing the whole seaborn library as alias `sns`

#### What is seaborn?

Seaborn is another visualization library which uses matplotlib in the backend for plotting

#### What is the major difference then between both matplotlib and seaborn?

- Seaborn uses **fascinating themes** and **reduces number of code lines** by doing a lot of work in the backend
- While matplotlib is used to **plot basic plots and add more functionlaity** on top of that
- Seaborn is built on the top of Pandas and Matplotlib

As we proceed through the lecture, we will see the difference between
both the libraries

import matplotlib.pyplot as plt
import seaborn as sns

Before we dive into learning these libraries, lets answer some general questions

#### Why do we even need to visualize data? When do I even need to visualise?

- It helps us to understand the data in a pictorial format
- It's **extensively used in Exploratory Data Analysis**
- And also to **present peformance results of our models**

Two reasons/scopes

- **Exploratory** - I can't see certain patterns just by crunching numbers (avg, rates, %ages)
- **Explanatory** - I can the numbers crunches and insights ready, but I'd like a visual art for storytelling

#### Data

- Rows: Samples, Data-points, Records
- Columns: Features, Variables

#### How many kinds of data do we have?

At the fundamental level, it\'s just two types:

- Numerical/Continuous
- Categorical

Categorical can be further divided into:

- **Ordinal:** Categorical Data with an order (E.g. low, medium, high)
- **Non-ordinal/nominal:** Categorical Data without any order (example
gender as Male/Female)

---
title: Video Game Data Intro
description:
duration: 500
card_type: cue_card
---
### **Video Games Analysis** (5 minutes)

You are a data scientist at \"**Tencent Games**\".
You need to analyze what kind of games they should start creating to get higher success in the market.

#### **Downloading the Dataset**

Code:

!gdown https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/021/299/original/final_vg1_-_final_vg_%281%29.csv?1670840166 -O final_vg.csv

> Output:

Downloading...
From: https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/021/299/original/final_vg1_-_final_vg_%281%29.csv?1670840166
To: /content/final_vg.csv
100% 2.04M/2.04M [00:00<00:00, 29.4MB/s]

#### **Importing Dataset using Pandas**

Code:

import pandas as pd
import numpy as np
data = pd.read_csv('final_vg.csv')
data.head()

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/588/original/Screenshot_2023-09-25_at_10.03.37_PM.png?1695659848" width="600" height="300">

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/589/original/Screenshot_2023-09-25_at_10.05.12_PM.png?1695659890" width="600" height="300">

If you notice,

- Columns like `Platform`, `Genre` are categorical
- While columns like `NA_Sales`, `Global_Sales`, `Rank` are continuous

On noticing further,

- `Platform` is of nominal type, no proper order between the categories
- `Year` is of ordinal type, there\'s a order to the categories

---
title: Intro to matplotlib
description:
duration: 900
card_type: cue_card
---

### **Introduction to Matplotlib** (10-15 minutes)

#### Lets learn to create a basic plot using plt

Now say, we want to draw a curve passing through 3 points:

- (0, 3)
- (1, 5)
- (2, 9)

#### How can we draw a curve using matplotlib ?

By using `plt.plot() function`

Code:

x_val = [0, 1, 2]
y_val = [3, 5, 9]
plt.plot(x_val, y_val)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/590/original/Screenshot_2023-09-25_at_10.10.30_PM.png?1695660068" width="400" height="300">

#### What can we observe from this plot ? {#what-can-we-observe-from-this-plot-}

- `plt.plot()` automatically decided the scale of the plot
- It also prints the **type of object** `matplotlib.lines.Line2D`

While this command decided a lot of things for you, you can customise each of these by understanding **components of a matplotlib plot**

#### **Anatomy of Matplotlib**

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/591/original/Screenshot_2023-09-25_at_10.12.20_PM.png?1695660180" width="700" height="400">

Woah! There is a lot of information in this image. Let's understand them one at a time.

- Figure: The **overall window** or page that everything is drawn on.
- You can create multiple independent Figures in Jupyter.
- If you run the code in terminal, separate windows will pop-up
- Axes: To the figure you can add multiple **Axes** which represents a
plot
- **Axis**: Simply the `x-axis` and `y-axis`
- **Axes**: - It is the **area** on which the **data is plotted** with
functions such as `plot()`
- **x-label**: Name of x-axis
- **y-label**: Name of y-axis
- **Major ticks**: subdivides the axis into major units. They
appear by default during plotting
- **Minor ticks**: subdivides the major tick units. They are by
default hidden and can be toggled on.
- **Title**: Title of each plot **(Axes)**, giving information about
the same
- **Legend**: describes the elements in the plot, blue and green
curves in this case
- **Suptitle**: The common title of all the plots

These are the major components of a matplotlib plot

#### Now, how to choose the right plot?

Firstly, depends on the what is your question of interest

When the question is clear:

- How many variables are involved?
- Whether the variable(s) are numerical or categorical?

#### How many variables are involved?

- 1 Variable - Univariate Analysis
- 2 Variables - Bivariate Analysis
- 2+ Variables - Multivariate Analysis

PS: Bivariate counts under multivariate, but let\'s keep it sep for ease of communication

What are the possible cases?

#### Univariate

- Numerical
- Categorical

#### Bivariate

- Numerical-Numerical
- Numerical-Categorical
- Categorical-Categorical

#### Multivariate

Let's start with 3 and then we can generalize

- Numerical-Numerical-Categorical
- Categorical-Categorical-Numerical
- Categorical-Categorical-Categorical
- Numerical-Numerical-Numerical

We will work on these one by one

---
title: Univariate Data Visualization of categorical data
description:
duration: 1800
card_type: cue_card
---

### **Univariate Data Visualization - Categorical Data**

#### What kind of questions we may want to ask for a categorical variable?

Questions like:

- What is the Distribution/Frequency of the data across different categories?
- What proportion does a particular category constitutes?

Let's take the categorical column "Genre"

#### How can we find the top-5 genres?

Recall, how could we get this data using pandas?

Code:

cat_counts = data['Genre'].value_counts()
cat_counts

> Output:

    Action          3316
    Sports          2400
    Misc            1739
    Role-Playing    1488
    Shooter         1310
    Adventure       1286
    Racing          1249
    Platform         886
    Simulation       867
    Fighting         848
    Strategy         681
    Puzzle           582
    Name: Genre, dtype: int64

#### Now what kind of plot can we use to visualize this information?

- We can perhaps plot categories on X-axis and their corresponding frequencies on Y-axis
- Such chart is called a Bar Chart or a Count Plot
- Can also plot horizontally when the #categories are many

### **Bar Chart** (25-30 minutes)

The data is binned here into categories

#### How can we draw a Bar plot ?

Using `plt.bar()`

Code:

x_bar=cat_counts.index
y_bar=cat_counts
plt.bar(x_bar,y_bar)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/592/original/Screenshot_2023-09-25_at_10.17.52_PM.png?1695660555" width="400" height="300">

The names seem to be overlapping a lot

#### How can we handle overlapping labels?

- Maybe decrease the font size (not preferred though)
- Or maybe increase the figure size
- Or rotate the labels

#### How can we change the plot size?

Code:

plt.figure(figsize=(12,8))
plt.bar(x_bar,y_bar)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/593/original/Screenshot_2023-09-25_at_10.21.30_PM.png?1695660708" width="500" height="400">

#### And how can we rotate the tick labels, also maybe increase the fontsize of the same?

plt.figure(figsize=(12,8))
plt.bar(x_bar,y_bar)
plt.xticks(rotation=90, fontsize=12)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/594/original/Screenshot_2023-09-25_at_10.22.44_PM.png?1695660782" width="500" height="400">

If you notice, the width of each bar is **1**

#### Can we change the width of these bars?

Code:

# same code
plt.figure(figsize=(10,8))
plt.bar(x_bar,y_bar,width=0.2)
plt.xticks(rotation = 90, fontsize=12)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/595/original/Screenshot_2023-09-25_at_10.24.04_PM.png?1695660859" width="500" height="400">

#### What about any additional styling to add to the bars ?

- We can **change colour of bars**
- We can add a **title to the axes**
- We can also add x and y labels

plt.figure(figsize=(10,8))
plt.bar(x_bar,y_bar,width=0.2,color='orange')
plt.title('Games per Genre',fontsize=15)
plt.xlabel('Genre',fontsize=12)
plt.ylabel('Count',fontsize=12)
plt.xticks(rotation = 90, fontsize=12)
plt.yticks(fontsize=12)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/596/original/Screenshot_2023-09-25_at_10.25.34_PM.png?1695660948" width="500" height="400">

If you notice, there\'s some text printed always before the plots.

This contains the data information of the plot

#### How can we remove the text printed before the plot and just display the plot?

Code:

plt.figure(figsize=(10,8))
plt.bar(x_bar,y_bar,width=0.2,color='orange')
plt.title('Games per Genre',fontsize=15)
plt.xlabel('Genre',fontsize=12)
plt.ylabel('Count',fontsize=12)
plt.xticks(rotation = 90, fontsize=12)
plt.yticks(fontsize=12)
plt.show()

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/597/original/Screenshot_2023-09-25_at_10.27.47_PM.png?1695661083" width="500" height="400">

#### How can we draw a bar-chart in Seaborn?

- In Seaborn, the same plot is called as **countplot**.
- Countplot automtically does even the counting of frequencies for you

#### Why not called a barplot?

There is **another function** in Seaborn called **barplot which has some other purpose** - discuss later

Code:

sns.countplot(x = 'Genre', data = data, order=data['Genre'].value_counts().index, color='cornflowerblue')
plt.xticks(rotation=90)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/598/original/Screenshot_2023-09-25_at_10.29.16_PM.png?1695661171" width="500" height="400">

The top 5 genres are action, sports, misc, role player, and shooter

### **Pie charts** (5-10 minutes)

#### What if instead of actual frequencues, I want see the proportion of the categories with each other?

Say, we want to compare the distrubution/proportion of sales across the different regions?

Which plot can we use for this?

A pie-chart!
<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/021/280/original/download.png?1670819443" width="300" height="250">

Code:

sales_data = data[['NA_Sales', 'EU_Sales', 'JP_Sales', 'Other_Sales']]
region_sales = sales_data.T.sum(axis='columns')

plt.pie(region_sales,
       labels=region_sales.index,
       startangle=90,
       explode=(0.2,0,0,0))
plt.show()

---
title: Quiz-1
description:
duration: 60
card_type: quiz_card
---

# Question
In the state “Haryana”, we want to find the proportion of people who smoke. Which will be the preferred plot?

# Choices
- [x] Pie Chart
- [ ] Bar Chart
- [ ] Count Plot
- [ ] BoxPlot

---
title: univariate data visualisation of Numerical Data
description:
duration: 1800
card_type: cue_card
---

### **Univariate Data Visualisation - Numerical Data**

#### What kind of questions we may have regarding a numerical variable?

1. How is the data distributed? Say distribution of number of games published in a year.
2. Is the data skewed? Are there any outliers? - Extremely high selling games maybe?
3. How much percentage of data is below/above a certain number?
4. Some special numbers - Min, Max, Mean, Median, nth percentile?

Now say, you want to find the distribution of games released every year.

Unlike barplot, **to see the distribution we will need to `bin` the data**.

**How can we understand popularity of video games year by year?**

#### **Histogram**

Code:

plt.hist(data['Year'])
plt.show()

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/599/original/Screenshot_2023-09-25_at_10.36.43_PM.png?1695661619" width="500" height="400">

- The curve is left skewed, with a lot more games being published in 2005-2015

- This shows that games started being highly popular in the last 1-2 decades, maybe could point to increased usage of internet worldwide!

If you notice, histograms are basically frequency charts

We can also vary the number of bins, the **default number of bins is 10**

So if we would need to see this data per decade, we would need 40 years in 4 bins.

Code:

plt.hist(data['Year'], bins=4)
plt.show()

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/600/original/Screenshot_2023-09-25_at_10.38.09_PM.png?1695661717" width="500" height="400">

We can also get the data of each bin, such as range of the boundaries, values, etc.

Code:

count, bins, _ = plt.hist(data['Year'])

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/601/original/Screenshot_2023-09-25_at_10.38.19_PM.png?1695661762" width="500" height="400">

count

> Output:

    array([ 112.,   70.,   92.,  449., 1274., 2440., 3921., 5262., 2406.,
            355.])

bins

> Output:

    array([1980., 1984., 1988., 1992., 1996., 2000., 2004., 2008., 2012.,
           2016., 2020.])

#### Now what do these `count` and `bins` mean?

- **bins** provides bin edges
- **counts** provides it corresponding counts

#### What is the length of `count`?

10

#### What shoud be the length of `bins`?

10 + 1 = 11

##### How can we plot histogram in Seaborn?

sns.histplot(data['Year'], bins=10)

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/602/original/Screenshot_2023-09-25_at_10.40.49_PM.png?1695661866" width="600" height="400">

Notice,

- The boundaries are more defined than matplotlib\'s plotting
- The x and y axis are labelled automatically

---
title: Quiz-2
description:
duration: 60
card_type: quiz_card
---

# Question
For analyzing marks by an edtech company, we want to find the range in which the most number of students have scored.
Which one will be the best plot ?

# Choices
- [x] Histogram
- [ ] Line Chart
- [ ] Bar Chart
- [ ] BoxPlot

---
title: Quiz-2 Explanation, Univariate data viz of Num Data (contd.)
description:
duration: 900
card_type: cue_card
---

#### Quiz-2 Explanation
Since we want to find number of studnets in a bin, we have to study the distribution of scores of students, which is a numerical variable. Hence we will use Histogram.

#### **Kernel Density Estimate (KDE) Plot**

- A KDE plot, similar to histrogram, is a method for visualizing the distributions
- But instead of bars, KDE represents data using a **continuous probability density curve**

#### Now, Why do we even need KDE plots?

- Compared to histogram, KDE produces a plot which is **less cluttered** and **more interpretable**
- Think of it as a **smoothened version** of histogram

Let's plot KDE using `seaborn`'s `kdeplot`

Code:

sns.kdeplot(data['Year'])

> Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/603/original/Screenshot_2023-09-25_at_10.44.26_PM.png?1695662080" width="600" height="400">

#### Can you notice the difference between KDE and histogram?

Y-Axis has **probability density estimation** instead of count

You can read more about this on:

<https://en.wikipedia.org/wiki/Kernel_density_estimation>
<https://www.youtube.com/watch?v=DCgPRaIDYXA>

**`Instructor Note:`**

Just for a brief idea, to find the probability, say probability of `Year` from 1990-2000, we will need to find the `area under the curve` for that interval

#### **Boxplot**

Now say I want to find the typical earnings of a game when it is published.

Or maybe find the aggregates like median, min, max and percentiles of the data.

#### What kind of plot can we use to understand the typical earnings from a game?

Box Plot

#### What exactly is a Box Plot?

- A box plot or **box-and-whisker plot** shows the **distribution of quantitative data**
- It facilitates comparisons between
- attributes
- across levels of a categorical attribute.

The **box**: Shows the **quartiles** of the dataset

The **whiskers**: Show the **rest of the distribution**

#### Box plots show the five-number summary of data:

1. Minimum score,
2. first (lower) quartile
3. Median
4. Third (upper) quartile
5. maximum score

#### **Diagram**

<img src=https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/063/627/original/iqr3.jpg?1706525143 width="650" height="200">

#### **Minimum Score**

- It is the **lowest value**, excluding outliers
- It is shown at the **end of bottom whisker**

#### **Lower Quartile**

- **25% of values** fall below the lower quartile value
- It is also known as the **first quartile**.

#### **Median**

- Median marks the **mid-point of the data**
- **Half the scores are greater than or equal to this value and half are less**.
- It is sometimes known as the **second quartile**.

#### **Upper Quartile**

- **75% of the values fall below the upper quartile value**
- It is also known as the **third quartile**.

#### **Maximum Score**

- It is the **highest value**, excluding outliers
- It is shown at the **end of upper whisker**.

#### **Whiskers**

- The upper and lower whiskers represent **values outside the middle 50%**
- That is, the **lower 25% of values** and the **upper 25% of values**.

#### **Interquartile Range (or IQR)**

- This is the box plot showing the **middle 50% of scores**
- It is the **range between the 25th and 75th percentile**.

#### Now, Let's plot a box plot to find the average typical earnings for a game

Code:

plt.figure(figsize=(15,10))
sns.boxplot(y = data["Global_Sales"])
plt.yticks(fontsize=20)
plt.ylabel('Global Sales (in million dollars)', fontsize=20)
plt.title('Global Sales of video games', fontsize=20)

>Output:

<img src="https://d2beiqkhq929f0.cloudfront.net/public_assets/assets/000/050/605/original/Screenshot_2023-09-25_at_10.48.46_PM.png?1695662360" width="500" height="400">

What can we infer from this?

The 5 point estimates (approx.) here are:

- Minimum, excluding outliers: 0
- Maximum, excluding outliers: 20 million dollars
- 25th Quantile: 6 million
- Median: around 7 million
- 75th Quantile: 12 million

There are quite a few outliers above 20 million dollars, represented by black diamonds

Key Takeaways:

Categorical - Barplot, Pie Chart

Numerical - Histogram, KDE, Boxplot

Can explore more types: Violin plot, bee-swarm plot, etc.

---
title: Quiz-3
description:
duration: 60
card_type: quiz_card
---

# Question
The telecom company "Airtel", wants to find the count of each payment mode opted by the customer. Which will be the preferred plot?

# Choices
- [ ] Boxplot
- [ ] Pie Chart
- [x] Count Plot
- [ ] Line Plot

---
title: Quiz-3 Explanation
description:
duration: 60
card_type: cue_card
---

We are using a single variable, which is categorical in this case. To find the count, we will be using countplot