Importing Libraries

While cleaning we have already imported Numpy and Pandas. Pandas and Numpy are essential for analysing the data. If you want to view the notebook where I cleaned the data, please click here.
We additionaly import :

  • Matplotlib, which will help us in creating data visualiation.
  • Seaborn, which is built on top of matplotlib with a quick & easy API for data plots & a thrid-party library
  • adjustText which will help us in labelling our plots.
  • We also import matplotlib.ticker which helps us generate the minor ticks in our plots.
    Important: If you get a Module not found error un-comment the below cell. 
 


import numpy as np
import pandas as pd

import matplotlib as mpl
import matplotlib.pyplot as plt
import seaborn as sns

# helps with arranging the texts in our plot
from adjustText import adjust_text

# helps with minor ticks in the plots
from matplotlib.ticker import MultipleLocator 

import warnings
warnings.filterwarnings('ignore')

Setting a few parameters for matplotlib 

mpl.rcParams['axes.facecolor'] = '#222222' # helps with the background color

# removes the axes of all the plots, makes it visually aesthetic
mpl.rcParams['axes.spines.right'] = False 
mpl.rcParams['axes.spines.top'] = False
mpl.rcParams['axes.spines.left'] = False
mpl.rcParams['axes.spines.bottom'] = False

# changing the tick sizes for better viewing
plt.rcParams['xtick.labelsize']=12 
plt.rcParams['ytick.labelsize']=12

# Changing to Heiti font as we have Chinese fonts
mpl.rcParams['font.family'] = ['Heiti TC']

Reading the cleaned data 

df = pd.read_csv('./audible_cleaned.csv')
df.head(3)
name author narrator time releasedate language stars price ratings
0 Geronimo Stilton #11 & #12 GeronimoStilton BillLobely 140 2008-04-08 English 5.0 468.0 34.0
1 The Burning Maze RickRiordan RobbieDaymond 788 2018-01-05 English 4.5 820.0 41.0
2 The Deep End JeffKinney DanRussell 123 2020-06-11 English 4.5 410.0 38.0

Basic exploration

While we know the shape of our dataset and the fact that we have removed all duplicate values, all null values in the Audible - Cleaner notebook but generally it's a great start working with any dataset to check for them.

print(f'Null values in the dataset: {df.isna().sum().sum()}')
print('-' * 30)
print(f'Duplicate values in the dataset: {df.duplicated().sum()}')
print('-' * 30)
print(f'Shape of our dataset: {df.shape}')

Null values in the dataset: 0
------------------------------
Duplicate values in the dataset: 0
------------------------------
Shape of our dataset: (87489, 9)

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 87489 entries, 0 to 87488
Data columns (total 9 columns):
 #   Column       Non-Null Count  Dtype  
---  ------       --------------  -----  
 0   name         87489 non-null  object 
 1   author       87489 non-null  object 
 2   narrator     87489 non-null  object 
 3   time         87489 non-null  int64  
 4   releasedate  87489 non-null  object 
 5   language     87489 non-null  object 
 6   stars        87489 non-null  float64
 7   price        87489 non-null  float64
 8   ratings      87489 non-null  float64
dtypes: float64(3), int64(1), object(5)
memory usage: 6.0+ MB

Runninng the built-in method info() in Pandas, we can come to a few conclusion about our data. We have:

  • 3 columns of datatype float64
  • 1 column of datatype int64
  • 5 columns of datatype object

    Note:We would need to convert the releasedate to a datetime object as that would give us much more detailed view on understanding our data.

df.describe().T.style.bar(subset='mean', color='crimson').background_gradient(subset=['50%'], cmap='coolwarm')
  count mean std min 25% 50% 75% max
time 87489.000000 417.497663 364.559399 1.000000 142.000000 386.000000 584.000000 8595.000000
stars 87489.000000 0.767811 1.709640 0.000000 0.000000 0.000000 0.000000 5.000000
price 87489.000000 559.009246 336.096642 0.000000 268.000000 585.000000 755.000000 7198.000000
ratings 87489.000000 3.723371 86.499601 0.000000 0.000000 0.000000 0.000000 12573.000000

Quite a few steps. Let's break them down.

  1. Pandas comes with a basic data analysis method describe() that gives us information on mean, std(standard deviation), minimum, maximum, 25 percentile, 50 percentile & 75 percentile.
  2. The .T simply transposes the dataframe.
  3. .style.bar is a way to highlight subsets in the dataframe. Here, we select the mean and compare it with the 50 percentile to see how our data is distributed.

Insights:

  • We see that our mean for the time column is higher than the 50% meaning our data may be right skewed.
  • While the price column looks like well balanced with heavy outliers.

In-depth exploration

The goal here is to dive into our dataset looking at the the data trends, distribution , outliers and view comparison between the columns, creating informative & aesthetic data visualization.

%matplotlib inline
author_titles = df.author.value_counts().sort_values(ascending=False)[:10]

fig, ax = plt.subplots(figsize=(15, 5), facecolor='#2b2b2b')

sns.barplot(x=author_titles.values, y=author_titles.index, orient='h', color='steelblue')
ax.bar_label(ax.containers[0], color='lightcyan', size=11, padding=4)

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Authors', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Top 10 authors with highest number of titles', color='lavender', fontsize=20, pad=20)

ax.grid(alpha=0.2, ls=':')
plt.show()

Let's break down the code.

  • %matplotlib inline helps generate a matplotlib graph inside the Jupyter notebook
  • value_counts() counts the number the unique values & sort_values() arrangeds the values in an ascending order.
  • Post that we create a figure & a single axes using plt.subplots. figsize takes a tuple of (width, height) and facecolor is the outer color of the plot.
  • Then we use the Seaborn library to create a barplot.
  • ax.bar_label allows us to add label to a bar plot. ax.containers represent an container object which holds the 10 values of the graph.(Here, it's 10 objects)
  • ax.tick_params represents the ticks and the associated markers.
  • ax.set_xlabel represents the label of x-axis.
  • ax.set_title represents the title of the plot.
  • ax.grid represents the grid structure or the dotted lines in behind. alpha controls the opacity & ls the linestyle.

  • plt.show() helps display the graph.

  • labelpad,padding,pad controls the padding in between the label and respective object.

%matplotlib inline
narrator_titles = df.narrator.value_counts().sort_values(ascending=False)[:10]

fig, ax = plt.subplots(figsize=(15, 5), facecolor='#2b2b2b')

sns.barplot(x=narrator_titles.values, y=narrator_titles.index, orient='h', color='steelblue')
ax.bar_label(ax.containers[0], color='lightcyan', size=11, padding=4)

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Narrators', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Top 10 narrators with highest number of titles', color='lavender', fontsize=20, pad=20)

ax.text(x=500, y=6, s="We see that the highest number of narrations done \nare 'anonymous' followed by 'uncredited' in fourth.",
        color='azure', size=14)


ax.grid(alpha=0.2, ls=':')
plt.show()


%matplotlib inline
narrator_titles = df.language.value_counts().sort_values(ascending=False)[:10]

fig, ax = plt.subplots(figsize=(15, 5), facecolor='#2b2b2b')

sns.barplot(x=narrator_titles.values, y=narrator_titles.index, orient='h', color='steelblue')
ax.bar_label(ax.containers[0], color='lightcyan', size=11, padding=4)

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Language', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Top 10 language with highest number of titles', color='lavender', fontsize=20, pad=20)

ax.grid(alpha=0.2, ls=':')
plt.show()

Distribution of time 

%matplotlib inline
fig, ax = plt.subplots(figsize=(18, 5), facecolor='#2b2b2b')

sns.histplot(df, x=df.time, kde=True, bins=300, color='steelblue', line_kws={'linewidth': 3})
ax.lines[0].set_color('crimson')

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Time in minutes', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Distribution of time', color='lavender', fontsize=20, pad=20)

ax.text(x=3000, y=6000, s="As expected, we see that time is highly right skewed.",
        color='azure', size=14)

ax.grid(alpha=0.2, ls=':')
plt.show()


%matplotlib inline
fig, ax = plt.subplots(figsize=(18, 5), facecolor='#2b2b2b')

flierprops = dict(markerfacecolor='skyblue', markersize=7, marker = 'o', markeredgecolor='#06113C')
sns.boxplot(x=df.time, color='steelblue', flierprops=flierprops, linewidth=2.5,)


ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Time in minutes', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Viewing outliers in time', color='lavender', fontsize=20, pad=20)

ax.grid(alpha=0.2, ls=':')
plt.show()

Distribution of price 

%matplotlib inline
fig, ax = plt.subplots(figsize=(18, 5), facecolor='#2b2b2b')

sns.histplot(df, x=df.price, kde=True, bins=300, color='steelblue', line_kws={'linewidth': 3})
ax.lines[0].set_color('crimson')

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Price', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Distribution of price', color='lavender', fontsize=20, pad=20)

ax.text(x=3000, y=3500, s="After looking into the distribution of price \nit seems that this too is right skewed, with outliers.",
        color='azure', size=14)

ax.grid(alpha=0.2, ls=':')
plt.show()


%matplotlib inline
fig, ax = plt.subplots(figsize=(18, 5), facecolor='#2b2b2b')

flierprops = dict(markerfacecolor='skyblue', markersize=7, marker = 'o', markeredgecolor='#06113C')
sns.boxplot(x=df.price, color='steelblue', flierprops=flierprops,linewidth=2.5)

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Price', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Viewing outliers in price', color='lavender', fontsize=20, pad=20)

ax.grid(alpha=0.2, ls=':')
plt.show()

Distribution of stars 

%matplotlib inline
fig, ax = plt.subplots(figsize=(18, 5), facecolor='#2b2b2b')

sns.histplot(df, x=df.ratings, kde=True, bins=300, color='steelblue', line_kws={'linewidth': 3})
ax.lines[0].set_color('crimson')

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Ratings', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Distribution of ratings', color='lavender', fontsize=20, pad=20)

ax.text(x=4500, y=80000, s="Ratings are highly right skewed. \nIf we continue to work with this data for modelling, \nwe definelty need to transform the data.",
        color='azure', size=14)

ax.grid(alpha=0.2, ls=':')
plt.show()


%matplotlib inline
fig, ax = plt.subplots(figsize=(18, 5), facecolor='#2b2b2b')

stars_count = df.stars.value_counts()

sns.barplot(x= stars_count.index, y=stars_count.values, color='steelblue',)
ax.bar_label(ax.containers[0], color='lightcyan', size=13, padding=0)

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Stars', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Count of stars', color='lavender', fontsize=20, pad=20)

ax.grid(alpha=0.2, ls=':')
plt.show()

Trend of released audiobooks by year 

df.releasedate = pd.to_datetime(df.releasedate)
df['year'] = pd.DatetimeIndex(df['releasedate']).year

We use pd.DatetimeIndex to tap into the datetime object and extract the year using the attribute year

df.head(3)
name author narrator time releasedate language stars price ratings year
0 Geronimo Stilton #11 & #12 GeronimoStilton BillLobely 140 2008-04-08 English 5.0 468.0 34.0 2008
1 The Burning Maze RickRiordan RobbieDaymond 788 2018-01-05 English 4.5 820.0 41.0 2018
2 The Deep End JeffKinney DanRussell 123 2020-06-11 English 4.5 410.0 38.0 2020 

yearly_books = df.year.value_counts().sort_index()

fig, ax = plt.subplots(figsize=(18, 5), facecolor='#2b2b2b')
sns.lineplot(x= yearly_books.index, y=yearly_books.values, color='steelblue', linewidth=3, marker='o')

ax.tick_params(axis='x', which='both', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Stars', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Count', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Number of audiobooks released by year', color='lavender', fontsize=20, pad=20)
ax.xaxis.set_minor_locator(MultipleLocator(1))

ax.grid(alpha=0.4, ls=':', which='major')
ax.grid(alpha=0.1, ls='--', which='minor')
plt.show()

Insights:

  • We see that the first audiobook was released as early as 1998.
  • Currently, we have data of pre-planned releases planned on 2025.
  • Looking at the trend, audiobooks started gaining popularity from 2003 hitting the highest on 2021 as of today(April 7, 2022)

Checking the relationship between ratings and price 

fig, ax = plt.subplots(figsize=(22, 7), facecolor='#2b2b2b')
sp = ax.scatter(x=df.ratings, y=df.price, c=df.stars, cmap='crest_r', edgecolor='#3A3845', s=110, alpha=0.8)

ax.tick_params(axis='x', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Ratings', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Price', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Relationship between price and ratings with stars', color='lavender', fontsize=20, pad=20)

texts= []
checker = []
for rating, amount, name in zip(df.ratings, df.price, df.name):
    if (rating) > 4000:
        if name not in checker:
            checker.append(name)
            texts.append(ax.text(rating, amount, name, fontdict=dict(color='azure', size=15)))

adjust_text(texts, force_points=0.5, force_text=3.5, expand_points=(1.2, 5.3), 
            expand_text=(0.9,0.9), autoalign=True,
            arrowprops=dict(arrowstyle = '->', alpha=0.3, lw=2, color='#B4A5A5',))


ax.grid(alpha=0.2, ls=':', which='major')

cb = plt.colorbar(sp)
cb.set_label('Stars', color='lavender', fontsize=18)
cb.ax.yaxis.set_tick_params(color='lavender', size=5)
plt.setp(plt.getp(cb.ax.axes, 'yticklabels'), color='lavender')

plt.show()

Code:

  • ax.scatter is a creating a scatter plot, this is using matplotlib and not seaborn like before
  • adjust_text is a library that helps you arrange text positions in matplotlib related plots. Here's the docs.
  • plt.colorbar generates the color bar on the right
  • setp & getp sets and gets the property of an object respectively, here it's the axes of the colorbar

Insights:

  • The higest ratings of a audiobook till date is for Atomic Habits with 1,25,690

Relationship between year and lenght of audiobooks 

fig, ax = plt.subplots(figsize=(22, 7), facecolor='#2b2b2b')
sp = ax.scatter(x=df.year, y=df.time, c=df.stars, cmap='crest_r', edgecolor='#3A3845', s=110, alpha=0.8)

ax.tick_params(axis='x', which='both', colors='gray', size=8)
ax.tick_params(axis='y', colors='gray', size=8)

ax.set_xlabel('Year', fontsize=16, color='lavender',labelpad=20)
ax.set_ylabel('Time in minutes', fontsize=16, color='lavender',labelpad=20)
ax.set_title('Yearly relationship with length of audiobooks', color='lavender', fontsize=20, pad=20)
ax.xaxis.set_minor_locator(MultipleLocator(1))

ax.text(x=2001, y=6000, s="We see a trend that the length of audiobooks \nincreased over time, with the longest on 2019.",
        color='azure', size=14)

ax.grid(alpha=0.2, ls=':', which='major')

cb = plt.colorbar(sp)
cb.set_label('Stars', color='lavender', fontsize=18)
cb.ax.yaxis.set_tick_params(color='lavender', size=5)
plt.setp(plt.getp(cb.ax.axes, 'yticklabels'), color='lavender')

plt.show()


co_relation = df.corr()
mapped = np.triu(np.ones_like(co_relation))

fig, ax = plt.subplots(figsize=(18, 7), facecolor='#2b2b2b')
sns.heatmap(co_relation, mask=mapped, annot=True, cmap='crest', linewidths=3, linecolor='#222222', cbar=False)

ax.tick_params(axis='x', colors='gray', size=15)
ax.tick_params(axis='y', colors='gray', size=15)

ax.text(x=2.5, y=2, s="We see a postive co-relation with time and price. \nIf the time of an audibook increases the price increases too.",
        color='azure', size=14)

ax.set_title('Co-relation between each columns', color='lavender', fontsize=20, pad=20)
plt.show()