Movie Notebook

Current Movie Industry Analysis
Author: Diane Tunnicliffe
Overview
I have been tasked with assisting Microsoft in their venture into the movie industry. My goal was to explore what
type of films are currently doing the best at the box office and to provide these findings to Microsoft's new
movie studio executives. My analysis of the movie industry, achieved by garnering data and utilizing descriptive
statistics and visualizations, has shown that a larger budget is correlated with a higher worldwide box office
gross. By allocating 75 million to 200 million dollars to produce an animated musical movie released in June or
November, or allocating 200 to 400 million dollars to produce a live action super hero movie released in April or
May, the data shows that a movie studio will be extremely likely to succeed. I have also given recommendations
as to which composers should be hired for an animated musical movie, and which directors should be hired for
a super hero movie. Microsoft can use this report to target their production budget, genre, creative type,
production method, release-time, and crew members of their upcoming movie endeavors to generate the
highest amount of revenue possible.
Business Problem
I have been informed that Microsoft wants a piece of the multi-billion dollar movie-making industry, but that they
are unsure of where to begin. The challenge for their new movie studio is that they are ready to jump into the
industry but do not have the necessary knowledge to move forward. To assist them with this goal, I have been
looking at the movies that performed highest in worldwide box office amounts for the past ten years. By
analyzing the movies that have been most successful recently, I can make recommendations about attributes
that Microsoft's movies should have in order to achieve the highest revenue. I have based my analysis on four
main factors:
Movie Type (Genre/Creative Type/Production Method): What types of movie content are currently most
successful?
Release Month: When is the most lucrative time of year to release a movie?
Production Budget: What budget amount tends to achieve the highest box office gross?
Additional Attributes: Based on these findings, what else do top-grossing movies have in common?
I chose these questions after considering the business problem and combing through the data I obtained. I have
determined that the answers to these questions are integral to the steps that should be taken when considering
how to produce the most profitable movie in today's world.
Data Understanding
I utilized three different data sources for my analysis in order to have the most comprehensive view of the
industry as it currently is.
OpusData Movie Data: a free dataset available upon request for academic research, comprised of 1,900
movies with a production year from 2006 to 2018, with a production budget greater than or equal to ten
million dollars. This dataset contains values for movie name, production budget, domestic and international
gross, genre, production method, runtime, and movie board rating.
Web-scraped data from The-Numbers.com: The Numbers is described as "the premier provider of movie
industry data and research services". This website contains domestic, international, and worldwide box
office revenue amounts per movie, and allows filtering and ordering of results based on many different
criteria. Some of the criteria provided on this site that I found especially useful were the same criteria listed
above: title, production budget, domestic and international gross, genre, and production method, in
addition to release date and worldwide gross. For the purposes of this project, I generated and scraped
reports for the top 100 movies per year, in terms of revenue, from 2010 to 2020.
The Movie Database (TMDb) API: The Movie Database is a popular database for movies and TV shows.
Their API is a system made freely available for data acquisition. There is a very large amount of data
available on TMDb; for the purposes of this project, I used it mainly to fill in missing information from my
other two datasets as I moved through my analysis.
In [2]: # importing the packages I will be using for this project
import pandas as pd
# setting pandas display to avoid scientific notation in my dataframes
pd.options.display.float_format = '{:.2f}'.format
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from bs4 import BeautifulSoup
import requests
%matplotlib inline
OpusData Movie Data
In [3]: # reading the csv file

opus_df = pd.read_csv('/Users/dtunnicliffe/Downloads/MovieData.csv')
# previewing the DataFrame
opus_df.head()
Out[3]:
movie_name production_year movie_odid production_budget domestic_box_office internation
Madea's
0 Family 2006 8220100 10000000 63257940
Reunion
1 Krrish 2006 58540100 10000000 1430721
End of the
2 2006 34620100 10000000 11748661
Spear
A Prairie
3 Home 2006 24910100 10000000 20342852
Companion
4 Saw III 2006 5840100 10000000 80238724

In [4]: # getting info for DataFrame
opus_df.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1936 entries, 0 to 1935
Data columns (total 13 columns):
movie_name 1936 non-null object
production_year 1936 non-null int64
movie_odid 1936 non-null int64
production_budget 1936 non-null int64
domestic_box_office 1936 non-null int64
international_box_office 1936 non-null int64
rating 1913 non-null object
creative_type 1923 non-null object
source 1915 non-null object
production_method 1925 non-null object
genre 1926 non-null object
sequel 1934 non-null float64
running_time 1822 non-null float64
dtypes: float64(2), int64(5), object(6)
memory usage: 196.8+ KB
In [5]: # getting counts for each value in genre column

opus_df['genre'].value_counts()
Out[5]: Drama 471

Adventure 334
Comedy 318
Action 311
Thriller/Suspense 231
Horror 104
Romantic Comedy 82
Musical 25
Black Comedy 24
Western 15
Concert/Performance 6
Documentary 5
Name: genre, dtype: int64
In [6]: # generating descriptive statistics for domestic box office values

opus_df['domestic_box_office'].describe()
Out[6]: count 1936.00

mean 64329960.75
std 87724369.60
min 0.00
25% 11003050.25
50% 36329945.00
75% 80069777.50
max 936662225.00
Name: domestic_box_office, dtype: float64
In [7]: # generating descriptive statistics for production budget values
opus_df['production_budget'].describe()
Out[7]: count 1936.00

mean 53428574.38
std 53685623.15
min 10000000.00
25% 19000000.00
50% 32750000.00
75% 65000000.00
max 425000000.00
Name: production_budget, dtype: float64
The-Numbers Web Scraping
2010
My first step was to obtain data for the top 100 grossing movies of 2010. I did this by building a report on The-
Numbers.com, scraping this data, and reading it into a pandas DataFrame.
In [8]: # url for the full customized report of top 100 movies for 2010
url = f"https://www.the-numbers.com/movies/report/All/All/All/All/All/Al
l/All/All/All/None/None/2010/2010/None/None/None/None/None/None?show-rel
ease-date=On&view-order-by=domestic-box-office&show-release-year=On&view
-order-direction=desc&show-production-budget=On&show-domestic-box-office
=On&show-international-box-office=On&show-worldwide-box-office=On&show-g
enre=On&show-production-method=On&show-creative-type=On"
response = requests.get(url)
# creating soup
soup = BeautifulSoup(response.text, 'lxml')
# finding table containing report info
table = soup.find('table')
# converting html of table into a string
table_string = f"""{table}"""
# reading html string into pandas
table_read = pd.read_html(table_string)
# converting into DataFrame
numbers_2010 = table_read[0]
# previewing DataFrame
numbers_2010.head()
Out[8]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType Pro
0
Jun 18,
0 1 2010 Toy Story 3 Adventure Digital Animation Kids Fiction
2010
Mar 5, Alice in Animation/Live

1 2 2010 Adventure Fantasy
2010 Wonderland Action
May 7,
2 3 2010 Iron Man 2 Action Live Action Super Hero
2010
The Twilight
Jun 30,
3 4 2010 Saga: Drama Live Action Fantasy
2010
Eclipse
Harry
Nov 19, Potter and Animation/Live
2010 the Deathly Action
Hallows:…
Now that I had a DataFrame to work with, I was able to start running some summary statistics and exploring the
data.
In [9]: # getting info for DataFrame
numbers_2010.info()
Unnamed: 0 100 non-null int64
Released 100 non-null object
Released.1 100 non-null int64
Title 100 non-null object
Genre 100 non-null object
ProductionMethod 100 non-null object
CreativeType 100 non-null object
ProductionBudget 100 non-null object
DomesticBox Office 100 non-null object
InternationalBox Office 100 non-null object
WorldwideBox Office 100 non-null object
dtypes: int64(2), object(9)
In [10]: # getting descriptive statistics for DataFrame

numbers_2010.describe()
Out[10]:
Unnamed: 0 Released.1
count 100.00 100.00
mean 50.50 2010.00
std 29.01 0.00
min 1.00 2010.00
25% 25.75 2010.00
50% 50.50 2010.00
75% 75.25 2010.00
max 100.00 2010.00
In [11]: # retrieving data type for domestic box office column values
numbers_2010['DomesticBox Office'].dtype
Out[11]: dtype('O')
I noted that the describe method for this DataFrame was not very helpful at this point because my dollar
amounts for domestic, international, and worldwide gross were pulled as objects (not floats or integers). I knew
that would require further adjusting in the next stage.
In [12]: # getting value counts for genre column
numbers_2010.Genre.value_counts()
Out[12]: Adventure 21
Drama 18
Comedy 16
Action 13
Horror 9
Romantic Comedy 8
Western 1
Documentary 1
Black Comedy 1
Musical 1
Name: Genre, dtype: int64
Adventure and Drama were the most common movie genres for top grossing movies in 2010.
In [13]: # getting value counts for production method column

numbers_2010['ProductionMethod'].value_counts()
Out[13]: Live Action 85

Animation/Live Action 8
Digital Animation 7
Name: ProductionMethod, dtype: int64
For 2010 production methods, Live Action was by far the most common, with 85% of the top grossing movies
being of this type.
In [14]: # getting value counts for creative type column

numbers_2010['CreativeType'].value_counts()
Out[14]: Contemporary Fiction 56

Fantasy 14
Kids Fiction 10
Science Fiction 7
Historical Fiction 5
Dramatization 5
Factual 2
Super Hero 1
Name: CreativeType, dtype: int64
Contemporary Fiction was the most common creative type by far for the top movies made in 2010.
Since I knew I'd be scraping data for each year in the exact same way as above, I created a function to speed
up the process.
In [15]: def number_scraper(year):
"""
Scrapes 100 top-grossing movies from The-Numbers.com.
Adds year input to url and scrapes resulting table.
Parameters:
year (int): user input 4-digit year for movie gross to be scraped.
Returns:
numbers_df (Pandas DataFrame): A dataframe populated with movie gros
s table values.
"""
# url for the full customized report of top 100 movies for release y
ears in range listed
url = f"https://www.the-numbers.com/movies/report/All/All/All/All/Al
l/All/All/All/All/None/None/{year}/{year}/None/None/None/None/None/None?
show-release-date=On&view-order-by=domestic-box-office&show-release-year
=On&view-order-direction=desc&show-production-budget=On&show-domestic-bo
x-office=On&show-international-box-office=On&show-worldwide-box-office=O
n&show-genre=On&show-production-method=On&show-creative-type=On"
# creating soup
soup = BeautifulSoup(response.text, 'lxml')
# finding table
table = soup.find('table')
# converting html of table into string
table_string = f"""{table}"""
# reading html string into pandas
table_read = pd.read_html(table_string)
# converting into DataFrame
numbers_df = table_read[0]
return numbers_df
2011
In [16]: # scraping 2011 data and compiling into DataFrame
numbers_2011 = number_scraper(2011)
#previewing DataFrame
numbers_2011.head()
Out[16]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType P
0
Harry Potter
Jul 15, and the Animation/Live
2011 Deathly Action
Hallows:…
Transformers:
Jun 29, Animation/Live Science
1 2 2011 Dark of the Action
2011 Action Fiction
Moon
The Twilight
Nov 18, Saga:
2 3 2011 Drama Live Action Fantasy
2011 Breaking
Dawn, Part 1
The
May 26, Contemporary
3 4 2011 Hangover Comedy Live Action
2011 Fiction
Part II
Pirates of the
May 20, Caribbean:
4 5 2011 Adventure Live Action Fantasy
2011 On
Stranger…
2012

numbers_2012.head()
Out[17]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeTyp
0
May 4, The Animation/Live

0 1 2012 Action Super He
2012 Avengers Action
The Dark
Jul 20,
1 2 2012 Knight Action Live Action Super He
2012
Rises
Mar 23, The Hunger Scienc

2 3 2012 Thriller/Suspense Live Action
2012 Games Fictio
Nov 8, Contempora
3 4 2012 Skyfall Action Live Action
2012 Fictio
The Hobbit:
Dec 14, An Animation/Live
4 5 2012 Adventure Fanta
2012 Unexpected Action
Journey
2013

numbers_2013.head()
Out[18]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType Prod
0
The
Hunger
Nov 22, Science
0 1 2013 Games: Adventure Live Action
2013 Fiction
Catching
Fire
May 3, Animation/Live
1 2 2013 Iron Man 3 Action Super Hero
2013 Action
Nov 22,
2 3 2013 Frozen Musical Digital Animation Kids Fiction
2013
Jul 3, Despicable
3 4 2013 Adventure Digital Animation Kids Fiction
2013 Me 2
Jun 14, Man of

4 5 2013 Action Live Action Super Hero
2013 Steel
2014

numbers_2014.head()
Out[19]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType
0
Dec 25, American

0 1 2014 Drama Live Action Dramatization
2014 Sniper
The
Hunger
Nov 21, Science
1 2 2014 Games: Thriller/Suspense Live Action
2014 Fiction
Mockingjay
- Part 1
Guardians
Aug 1, Animation/Live
2 3 2014 of the Action Super Hero
2014 Action
Galaxy
Captain
Apr 4, America:
2014 The Winter
Soldier
Feb 7, The Lego

2014 Movie
2015

numbers_2015.head()
Out[20]:
Unnamed:
0
Star Wars
Dec 18, Ep. VII: Animation/Live Science
0 1 2015 Adventure
2015 The Force Action Fiction
Awakens
Jun 12, Jurassic Science

1 2 2015 Action Live Action
2015 World Fiction
Avengers:
May 1, Animation/Live
2 3 2015 Age of Action Super Hero
2015 Action
Ultron
Jun 19, Inside

2015 Out
Apr 3, Contemporary
4 5 2015 Furious 7 Action Live Action
2015 Fiction
2016
numbers_2016.head()
Out[21]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType Produc
0
Rogue
One: A
Dec 16, Animation/Live Science
0 1 2016 Star Adventure $
2016 Action Fiction
Wars
Story
Jun 17, Finding

1 2 2016 Adventure Digital Animation Kids Fiction $
2016 Dory
Captain
May 6,
2 3 2016 America: Action Live Action Super Hero $
2016
Civil War
The
Jul 8, Secret
2016 Life of
Pets
The
Apr 15, Animation/Live
4 5 2016 Jungle Adventure Fantasy $
2016 Action
Book
2017

numbers_2017.head()
Out[22]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType Produ
0
Star Wars
Dec 15, Ep. VIII: Animation/Live Science
0 1 2017 Adventure
2017 The Last Action Fiction
Jedi
Beauty
Mar 17, Animation/Live
1 2 2017 and the Musical Fantasy
2017 Action
Beast
Jun 2, Wonder
2017 Woman
Jumanji:
Dec 20, Welcome Science
3 4 2017 Adventure Live Action
2017 to the Fiction
Jungle
Guardians
May 5, of the Animation/Live
4 5 2017 Action Super Hero
2017 Galaxy Action
Vol 2
2018

numbers_2018.head()
Out[23]:
Unnamed:
0
Feb 16, Black

2018 Panther
Avengers:
Apr 27, Animation/Live
1 2 2018 Infinity Action Super Hero
2018 Action
War
Jun 15, Incredibles

2018 2
Jurassic
Jun 22, World: Science
2018 Fallen Fiction
Kingdom
Dec 21, Animation/Live

4 5 2018 Aquaman Action Super Hero
2018 Action
2019

numbers_2019.head()
Out[24]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType Produ
0
Apr 26, Avengers: Animation/Live

0 1 2019 Action Super Hero
2019 Endgame Action
Jul 19, The Lion Animation/Live

1 2 2019 Adventure Kids Fiction
2019 King Action
Star
Dec 20, Wars: The Animation/Live Science
2 3 2019 Adventure
2019 Rise of Action Fiction
Skywalker
Nov 22,
3 4 2019 Frozen II Adventure Digital Animation Kids Fiction
2019
Jun 21, Toy Story

2019 4
2020

numbers_2020.head()
Out[25]:
Unnamed:
Released Released.1 Title Genre ProductionMethod CreativeType Pro
0
Jan 17, Bad Boys Contemporary

2020 For Life Fiction
Feb 14, Sonic The Animation/Live

1 2 2020 Adventure Kids Fiction
2020 Hedgehog Action
Birds of
Prey (And
Feb 7,
2 3 2020 the Action Live Action Super Hero
2020
Fantabulous
Em…
Jan 17, Animation/Live

3 4 2020 Dolittle Adventure Fantasy
2020 Action
Feb 28, The Invisible Science

4 5 2020 Horror Live Action
2020 Man Fiction
Data Preparation
Now that my data was scraped ready to go, it was time to clean it up and prepare it for analysis.
OpusData Movie Data
Cleaning
There were a few columns in this dataset that were not relevant to my analysis: 'movie_odid', 'source', 'sequel',
'running-time', and 'rating'. I began by dropping those.
In [26]: # dropping unnecessary columns

opus_df.drop(['movie_odid', 'source', 'sequel', 'running_time', 'rating'
], axis=1, inplace=True)
I then renamed some of the column names to make them easier to work with.
In [27]: # renaming certain columns

opus_df = opus_df.rename(columns={'movie_name':'title','production_budge
t':'budget', 'domestic_box_office': 'dom_gross', 'international_box_offi
ce':'int_gross', 'production_method':'prod_method'})
In [28]: # showing all column names
opus_df.columns
Out[28]: Index(['title', 'production_year', 'budget', 'dom_gross', 'int_gross',

'creative_type', 'prod_method', 'genre'],
dtype='object')
In [29]: # saving copy of DataFrame as csv file

opus_df.to_csv('./data/opus_df.csv')
Scraped data from The-Numbers.com
Cleaning
Due to the fact that I compiled my data from tables that were completely filled in, I was pleased that I had no
null values to deal with. I did, however, have an unnecessary column called 'Unnamed: 0', which was used as
the index of each table in its original html format. I began by dropping this.
In [30]: numbers_2010.isnull().sum()
Out[30]: Unnamed: 0 0
Released 0
Released.1 0
Title 0
Genre 0
ProductionMethod 0
CreativeType 0
ProductionBudget 0
DomesticBox Office 0
InternationalBox Office 0
WorldwideBox Office 0
dtype: int64
In [31]: # dropping Unnamed column

numbers_2010 = numbers_2010.drop(columns='Unnamed: 0')
I then made all the column names lowercase for ease of use.
In [32]: # converting column names to all lowercase

numbers_2010.columns = [x.lower() for x in numbers_2010.columns]
# showing all column names
numbers_2010.columns
Out[32]: Index(['released', 'released.1', 'title', 'genre', 'productionmethod',

'creativetype', 'productionbudget', 'domesticbox office',
'internationalbox office', 'worldwidebox office'],
dtype='object')
I also wanted to rename some of the columns to make them more comprehensive.
In [33]: # renaming columns
numbers_2010 = numbers_2010.rename(columns = {'released':'release_date',
'released.1':'release_year', 'productionmethod':'prod_method', 'domestic
box office':'dom_gross', 'internationalbox office':'int_gross', 'worldwi
debox office':'world_gross', 'creativetype': 'creative_type', 'productio
nbudget': 'budget'})
# showing all column names
numbers_2010.columns
Out[33]: Index(['release_date', 'release_year', 'title', 'genre', 'prod_method',

'creative_type', 'budget', 'dom_gross', 'int_gross', 'world_gros
s'],
dtype='object')
Finally, I wanted to convert the dollar amounts to numbers I could actually work with.
In [34]: # removing dollar signs and commas from dollar amounts

# converting dollar amounts from strings into integers
numbers_2010['dom_gross'] = numbers_2010['dom_gross'].str.replace(',',
'').str.replace('$', '').astype(int)
numbers_2010['int_gross'] = numbers_2010['int_gross'].str.replace(',',
'').str.replace('$', '').astype(int)
numbers_2010['world_gross'] = numbers_2010['world_gross'].str.replace(
',', '').str.replace('$', '').astype(int)
numbers_2010['budget'] = numbers_2010['budget'].str.replace(',', '').str
.replace('$', '').astype(int)
numbers_2010.head()
Out[34]:
release_date release_year title genre prod_method creative_type budget do
Digital
0 Jun 18, 2010 2010 Toy Story 3 Adventure Kids Fiction 200000000 41
Animation
Alice in Animation/Live
1 Mar 5, 2010 2010 Adventure Fantasy 200000000 33
Wonderland Action
2 May 7, 2010 2010 Iron Man 2 Action Live Action Super Hero 170000000 31
The Twilight
3 Jun 30, 2010 2010 Saga: Drama Live Action Fantasy 68000000 30
Eclipse
Harry
Potter and Animation/Live
4 Nov 19, 2010 2010 Adventure Fantasy 125000000 29
the Deathly Action
Hallows:…
Since this is how I intended to clean all my DataFrames, I wrote a function to execute all the steps I had taken.
In [35]: def clean(df):
"""
Cleans and modifies a given dataframe according to criteria set for
this particular project.
Drops column called 'Unnamed:0'.

Converts column names to lowercase.
Renames certain columns to make them shorter/more comprehensive.
Removes dollar signs and commas from dollar amounts.
Converts dollar amounts from strings into integers.
Parameters:
df (Pandas DataFrame): user input dataframe based on previously scra
ped table values from The-Numbers.com.
Returns:
df (Pandas DataFrame): A dataframe cleaned and adjusted as per crite
ria listed above.
"""
# drop 'Unnamed' column
df = df.drop(columns='Unnamed: 0')
# make column names lowercase
df.columns = [x.lower() for x in df.columns]
# rename certain columns
df = df.rename(columns = {'released':'release_date', 'released.1':'r
elease_year', 'productionmethod':'prod_method',
'domesticbox office':'dom_gross', 'interna
tionalbox office':'int_gross', 'worldwidebox office':'world_gross',
'creativetype':'creative_type', 'productio
nbudget': 'budget'})
# removing dollar signs and commas from dollar amounts
# converting dollar amounts from strings into integers
df['dom_gross'] = df['dom_gross'].str.replace(',', '').str.replace(
'$', '').astype(int)
df['int_gross'] = df['int_gross'].str.replace(',', '').str.replace(
'$', '').astype(int)
df['world_gross'] = df['world_gross'].str.replace(',', '').str.repla
ce('$', '').astype(int)
df['budget'] = df['budget'].str.replace(',', '').str.replace('$', ''
).astype(int)
return df
In [36]: # cleaning data
numbers_2011 = clean(numbers_2011)
# previewing cleaned data
numbers_2011.head()
Out[36]:
release_date release_year title genre prod_method creative_type budget d
Harry Potter
and the Animation/Live
0 Jul 15, 2011 2011 Adventure Fantasy 125000000
Deathly Action
Hallows:…
Transformers:
Animation/Live Science
1 Jun 29, 2011 2011 Dark of the Action 195000000
Action Fiction
Moon
The Twilight
Saga:
2 Nov 18, 2011 2011 Drama Live Action Fantasy 127500000
Breaking
Dawn, Part 1
The
Contemporary
3 May 26, 2011 2011 Hangover Comedy Live Action 80000000
Fiction
Part II
Pirates of the
Caribbean:
4 May 20, 2011 2011 Adventure Live Action Fantasy 379000000
On
Stranger…








Combining
Now that all the data had been cleaned, I was ready to combine all my DataFrames into a single DataFrame for
further analysis.
In [46]: # concatenating my DataFrames for box office data from years 2015-2020
numbers_df = pd.concat([numbers_2010, numbers_2011, numbers_2012, number
s_2013,
numbers_2014, numbers_2015, numbers_2016, number
s_2017,
numbers_2018, numbers_2019, numbers_2020], axis=
0, ignore_index=True)
# getting info on my new DataFrame
numbers_df.info()
release_date 1100 non-null object
release_year 1100 non-null int64
title 1100 non-null object
genre 1099 non-null object
prod_method 1100 non-null object
creative_type 1100 non-null object
budget 1100 non-null int64
dom_gross 1100 non-null int64
int_gross 1100 non-null int64
world_gross 1100 non-null int64
dtypes: int64(5), object(5)
At this point I realized that, somehow, one singular null value had snuck in there.
In [47]: # checking for null values

numbers_df.loc[numbers_df.genre.isnull()]
Out[47]:
release_date release_year title genre prod_method creative_type budget dom_gross
Science
1091 Jul 10, 2020 2020 Archive NaN Live Action 0 139593
Fiction
Since it was just one null value for genre for the movie 'Archive', it was easy enough to perform a quick search
on IMDb. The primary genre listed for this movie is Drama, so I filled in the value manually.
In [48]: # for movie with title value Archive, set genre to Drama
numbers_df.loc[numbers_df['title']=='Archive', 'genre'] = 'Drama'
In [49]: # checking for successful value change

numbers_df.loc[numbers_df['title'] == 'Archive']
Out[49]:
release_date release_year title genre prod_method creative_type budget dom_gross
Science
1091 Jul 10, 2020 2020 Archive Drama Live Action 0 139593
Fiction
Feature Engineering
I was curious about how the timing of a release impacted the revenue. So my next step was to create a column
for the month that each movie was released in.
In [50]: # convert release date column to datetime values

numbers_df['release_date'] = pd.to_datetime(numbers_df['release_date'])
# create release month column
numbers_df['release_month'] = numbers_df['release_date'].dt.strftime('%
B')
In [51]: # checking for successful column creation

numbers_df['release_month'].value_counts()
Out[51]: February 104

December 104
August 104
November 103
July 97
March 92
January 89
June 89
September 86
October 85
May 79
April 68
Name: release_month, dtype: int64
In [52]: # saving copy of DataFrame as csv file

numbers_df.to_csv('./data/thenumbers_df.csv')
Putting it all together!

In [53]: # show all column names
numbers_df.columns

s',
'release_month'],
dtype='object')
In [54]: # show number of rows and columns

numbers_df.shape
Out[54]: (1100, 11)

opus_df.columns
Out[55]: Index(['title', 'production_year', 'budget', 'dom_gross', 'int_gross',

'creative_type', 'prod_method', 'genre'],
dtype='object')
In [56]: # show number of rows and columns of DataFrame

opus_df.shape
Out[56]: (1936, 8)
In [57]: # merging the DataFrames

merged_df = pd.merge(numbers_df, opus_df, how='outer')
# previewing the new DataFrame
merged_df.shape
Out[57]: (2441, 12)
In [58]: # filter DataFrame to only include movies with a production year greater
than or equal to 2010 (or null)
merged_df = merged_df[(merged_df['production_year']>=2010.00) | (merged_
df['production_year'].isnull())]
In [59]: # show number of rows and columns

merged_df.shape
Out[59]: (1745, 12)

In [60]: # previewing DataFrame
merged_df.head()
Out[60]:
Digital
0 2010-06-18 2010.00 Toy Story 3 Adventure Kids Fiction 200000000 41
Animation
Alice in Animation/Live
1 2010-03-05 2010.00 Adventure Fantasy 200000000 33
Wonderland Action
2 2010-05-07 2010.00 Iron Man 2 Action Live Action Super Hero 170000000 31
The Twilight
3 2010-06-30 2010.00 Saga: Drama Live Action Fantasy 68000000 30
Eclipse
Harry
Potter and Animation/Live
4 2010-11-19 2010.00 Adventure Fantasy 125000000 29
the Deathly Action
Hallows:…

merged_df.columns

s',
'release_month', 'production_year'],
dtype='object')
Cleaning Duplicates
In [62]: # generating rows where movies are duplicates in terms of title
merged_df.loc[merged_df.duplicated(subset=['title'])]
Out[62]:
release_date release_year title genre prod_method creative_type budget
Harry
Potter and
Animation/Live
100 2011-07-15 2011.00 the Adventure Fantasy 125000000
Action
Deathly
Hallows:…
Robin Historical
887 2018-11-21 2018.00 Action Live Action 99000000
Hood Fiction
Contemporary
1076 2020-10-02 2020.00 The Call Horror Live Action 0
Fiction
Midnight in Romantic
1858 NaT nan Live Action Fantasy 30000000
Paris Comedy
A
Nightmare
1866 NaT nan Horror Live Action Fantasy 35000000
on Elm
Street
... ... ... ... ... ... ... ...
Robin Historical
2432 NaT nan Action Live Action 99000000
Hood Fiction
Mary
2435 NaT nan Poppins Musical Live Action Kids Fiction 130000000
Returns
The
Nutcracker
2436 NaT nan and the Adventure Live Action Fantasy 132900000
Four
Realms
2437 NaT nan Aquaman Action Live Action Super Hero 160000000
Ralph
Breaks Digital
2438 NaT nan Adventure Kids Fiction 175000000
The Animation
Internet
185 rows × 12 columns
In [63]: merged_df.loc[merged_df['title']=='Aquaman']
Out[63]:
release_date release_year title genre prod_method creative_type budget dom_
Animation/Live
804 2018-12-21 2018.00 Aquaman Action Super Hero 160000000 3350
Action
2437 NaT nan Aquaman Action Live Action Super Hero 160000000 3338
In [64]: merged_df.loc[merged_df['title']=='Ralph Breaks The Internet']
Out[64]:
release_date release_year title genre prod_method creative_type budget dom
Ralph
Breaks Digital
813 2018-11-21 2018.00 Adventure Kids Fiction 175000000 201
The Animation
Internet
Ralph
Breaks Digital
2438 NaT nan Adventure Kids Fiction 175000000 200
The Animation
Internet
In [65]: merged_df.loc[merged_df['title']=='Midnight in Paris']
Out[65]:
Midnight Romantic
157 2011-05-20 2011.00 Live Action Fantasy 30000000 568
in Paris Comedy
Midnight Romantic
1858 NaT nan Live Action Fantasy 30000000 568
in Paris Comedy
After reviewing many of the duplicates, I was able to determine that if a movie was showing up as a duplicate in
terms of title, it was truly a duplicate entry of the same movie. I wanted to give priority to the data that had come
from numbers_df since it was more current (and had more non-null values) than the data that had come from
opus_df. So I opted to keep the first instance of each duplicate and drop the rest.
In [66]: # dropping duplicates

merged_df.drop_duplicates(subset=['title'], inplace=True)

merged_df.isnull().sum()
Out[67]: release_date 474

release_year 474
title 0
genre 1
prod_method 1
creative_type 3
budget 0
dom_gross 0
int_gross 0
world_gross 474
release_month 474
production_year 502
dtype: int64
Based on the amount of null values in the world_gross and release_month columns, I could see that there were
474 movies that were in my scraped Numbers dataset but not in my Opus dataset. And based on the amount of
null vlaues in the production_year column, I could see that there were 501 movies that were in my Opus dataset
but not my Numbers dataset.
In [68]: # creating a list of titles that did not have an associated release_date
value
dateless_titles = merged_df['title'].loc[merged_df['release_date'].isnul
l()]
I created a list of the movie titles that originated from the Opus dataset which did not contain release date
information. I then ran a function to query The Movies Database (TMDb) via API for each specific title and pull
the release date for the title.
In [69]: import json

from datetime import datetime
In [70]: def get_keys(path):

"""Retrieves API key from files as api_key."""
with open(path) as f:
return json.load(f)
keys = get_keys("/Users/dtunnicliffe/.secret/TMDb_api.json")
api_key = keys['api_key']
In [71]: def get_date(title):
"""
Updates release date information for movie in dataframe.
Queries TMDB for a given movie title.

Retrieves release date information for title.
Adds new release date value to movie's row in dataframe.
Parameters:
title (str): user input movie title.
Returns:
df (Pandas DataFrame): A dataframe cleaned and adjusted as per crite
ria listed above.
"""
title_r = title.replace(' ', '+')
url = f"https://api.themoviedb.org/3/search/movie?api_key={api_key}&
query={title_r}"
if len(response.json()['results']) > 0:
rdate = (response.json()['results'][0]['release_date'])
if rdate:
x = datetime.strptime(rdate, '%Y-%m-%d').strftime('%b %d, %
Y')
merged_df.loc[merged_df['title']==title, 'release_date'] = x
else:
pass
In [72]: # getting release dates for list of titles lacking release dates
for title in dateless_titles:
get_date(title)

release_year 474
title 0
genre 1
prod_method 1
creative_type 3
budget 0
dom_gross 0
int_gross 0
world_gross 474
release_month 474
production_year 502
dtype: int64
My get_date function successfully took care of almost all the null values for release_date! I had a look at the few
null values that remained.
In [74]: # show rows that contain null values for release date
merged_df[merged_df['release_date'].isnull()]
Out[74]:
release_date release_year title genre prod_method creative_type bu
Jin líng shí Historical

1962 NaT nan Drama Live Action 10000
san chai Fiction
San cheng
2160 NaT nan Drama Live Action Dramatization 1200
ji
Savva.
Digital
2202 NaT nan Serdtse Adventure Kids Fiction 3000
Animation
voyna
Baahubali
Historical
2282 NaT nan 2: The Action Live Action 3000
Fiction
Conclusion
Contemporary
2345 NaT nan Jìyì dàshī Thriller/Suspense Live Action 2000
Fiction
Chāi dàn Contemporary

2350 NaT nan Action Live Action 2300
zhuānjiā Fiction
Based on the fact that all these movies had many columns of null values, I opted to drop them from the data.
In [75]: # dropping rows that have null release dates

merged_df = merged_df.dropna(axis=0, subset=['release_date'])

release_year 468
title 0
genre 1
prod_method 1
creative_type 3
budget 0
dom_gross 0
int_gross 0
world_gross 468
release_month 468
production_year 502
dtype: int64
Next, I dealt with movies that did not have a world_gross value.
In [77]: # creating a list of titles that did not have a world_gross value
worldless_titles = merged_df['title'].loc[merged_df['world_gross'].isnul
l()]
worldless_titles
Out[77]: 1785 You Got Served: Beat The World

1786 A Better Life
1787 Cedar Rapids
1788 Blood Done Sign My Name
1789 MacGruber
...
2428 Dr. Seuss’ The Grinch
2433 Mortal Engines
2434 How to Train Your Dragon: The Hidden World
2439 Mission: Impossible—Fallout
2440 Fantastic Beasts: The Crimes of Grindelwald
Name: title, Length: 468, dtype: object
To fill in this data, I ran a function to add the domestic gross and international gross column values together for
each row to equal the world gross.
In [78]: def worldsum(title):

"""Gets sum of dom_gross and int_gross values for movie title, sets
as world_gross value"""
dg = merged_df.loc[merged_df['title'] == title, 'dom_gross']
ig = merged_df.loc[merged_df['title'] == title, 'int_gross']
merged_df.loc[merged_df['title'] == title, 'world_gross'] = dg + ig
In [79]: # generating world_gross values for all titles lacking world_gross

for title in worldless_titles:
worldsum(title)

release_year 468
title 0
genre 1
prod_method 1
creative_type 3
budget 0
dom_gross 0
int_gross 0
world_gross 0
release_month 468
production_year 502
dtype: int64
I no longer needed the production_year column at this point, so I went ahead and dropped that from the
DataFrame.
In [81]: # dropping production_year column
merged_df.drop('production_year', axis=1, inplace=True)
I once again made use of the datetime methods I had used previously for numbers_df prior to the merge, in
order to fill in the null values for 'month' in merged_df.
In [82]: # converting release_date column to datetime object

merged_df['release_date'] = pd.to_datetime(merged_df['release_date'])
# generating release_month value based on datetime release_date
merged_df['release_month'] = merged_df['release_date'].dt.strftime('%B')

release_year 468
title 0
genre 1
prod_method 1
creative_type 3
budget 0
dom_gross 0
int_gross 0
world_gross 0
release_month 0
dtype: int64
Finally, I had to set the release_year for the titles whose release_dates had been scraped after the merge.
In [84]: # setting release_year based on datetime release_date values

merged_df['release_year'] = merged_df['release_date'].dt.year

release_year 0
title 0
genre 1
prod_method 1
creative_type 3
budget 0
dom_gross 0
int_gross 0
world_gross 0
release_month 0
dtype: int64
In [86]: # getting counts of each value in release_year column
merged_df.release_year.value_counts()
Out[86]: 2011 162

2016 158
2013 155
2015 152
2012 148
2014 146
2010 142
2017 133
2018 119
2019 109
2020 102
2009 8
2005 3
2007 2
1968 1
1996 1
1969 1
1982 1
1985 1
1988 1
1994 1
2006 1
1997 1
1999 1
2000 1
2001 1
2004 1
2008 1
1967 1
Name: release_year, dtype: int64
When I checked the value counts for the release_year column, I found 31 values where the release year was not
within my parameters.
In [87]: merged_df.loc[merged_df['release_year']<2010]
Out[87]:
Io sono Historical
1795 2009-09-05 2009 Drama Live Action 1000
l’amore Fiction
Contemporary
1805 2009-03-25 2009 Chloe Thriller/Suspense Live Action 1300
Fiction
I Love You,
Contemporary
1808 2009-01-18 2009 Phillip Comedy Live Action 1300
Fiction
Morris
Enter the
1820 2009-06-17 2009 Drama Live Action Fantasy 1600
Void
Youth in Contemporary
1824 2009-09-11 2009 Comedy Live Action 1800
Revolt Fiction
The Last
1826 2009-09-04 2009 Drama Live Action Dramatization 1800
Station
Historical
1833 2009-05-17 2009 Middle Men Comedy Live Action 2000
Fiction
Contemporary
1840 2001-11-16 2001 Stone Drama Live Action 2200
Fiction
1853 2009-08-13 2009 Case 39 Horror Live Action Fantasy 2700
Going the Romantic Contemporary

1862 2004-08-19 2004 Live Action 3200
Distance Comedy Fiction
Historical
1880 2000-05-26 2000 Shanghai Drama Live Action 5000
Fiction
Les Contemporary
1902 1969-04-01 1969 Comedy Live Action 1080
Intouchables Fiction
1946 1982-06-25 1982 The Thing Horror Live Action Fantasy 3800
The
1948 2006-07-28 2006 Drama Live Action Dramatization 4000
Impossible
Dangerous Historical
1996 1988-06-09 1988 Drama Live Action 2420
Liaisons Fiction
Contemporary
2053 1994-09-23 1994 Redemption Thriller/Suspense Live Action 2300
Fiction
Historical
2095 2007-01-05 2007 Freedom Drama Live Action 1450
Fiction
2120 1967-11-27 1967 Viy Adventure Live Action Fantasy 2600
Contemporary
2129 2005-01-01 2005 Aloha Drama Live Action 3700
Fiction
Contemporary
2161 2005-08-26 2005 Brothers Action Live Action 1300
Fiction
Robinson Digital
2166 1997-05-12 1997 Adventure Kids Fiction 1300
Crusoe Animation
2195 1985-07-19 1985 Legend Thriller/Suspense Live Action Dramatization 2500
Historical
2210 2008-12-12 2008 Yip Man 3 Action Live Action 3600
Fiction
Contemporary
2253 1968-01-01 1968 Sultan Action Live Action 1100
Fiction
Historical
2296 2007-03-16 2007 Silence Drama Live Action 4650
Fiction
2353 1996-08-02 1996 Matilda Drama Live Action Dramatization 2500
Contemporary
2414 2005-09-03 2005 Serenity Thriller/Suspense Live Action 2500
Fiction
Dr. Seuss’ Digital

2428 1999-10-05 1999 Adventure Kids Fiction 7500
The Grinch Animation
All of these were movies with release dates prior to 2010 that had been added from my get_date function after
my earlier filtering. I went ahead and dropped those as well.
In [88]: # show number of rows and columns before filtering

merged_df.shape
Out[88]: (1554, 11)
In [89]: # filter DataFrame to only include titles with release_year >= 2010
merged_df = merged_df[merged_df['release_year']>=2010.00]
In [90]: # show number of rows and columns after filtering

merged_df.shape
Out[90]: (1526, 11)
In [91]: merged_df.isnull().sum()
release_year 0
title 0
genre 1
prod_method 1
creative_type 3
budget 0
dom_gross 0
int_gross 0
world_gross 0
release_month 0
dtype: int64
Looking good! I had exactly 5 null values left in my whole DataFrame.
In [92]: # show columns where value for genre is null

merged_df.loc[merged_df['genre'].isnull()]
Out[92]:
release_date release_year title genre prod_method creative_type budget dom_gro
Le Contemporary
1976 2012-04-25 2012 NaN Live Action 11000000 616
prénom Fiction
When I searched this entry on IMDb, there was no useful information on this title, so I dropped this as well.
In [93]: # dropping row with null value for genre

merged_df = merged_df.dropna(axis=0, subset=['genre'])
In [94]: # show rows where creative type value is null

merged_df.loc[merged_df['creative_type'].isnull()]
Out[94]:
1789 2010-05-21 2010 MacGruber Comedy Live Action NaN 10000000
The Killer
1807 2010-02-19 2010 Drama Live Action NaN 13000000
Inside Me
2054 2013-12-18 2013 Dhoom 3 Action NaN NaN 24000000
In [95]: # getting counts for each value in creative_type column

merged_df['creative_type'].value_counts()

Dramatization 164
Science Fiction 160
Kids Fiction 154
Fantasy 136
Super Hero 53
Factual 18
Multiple Creative Types 5
Name: creative_type, dtype: int64
In [96]: # filling null values for creative_type with the most common value for t
his column
merged_df['creative_type'].fillna('Contemporary Fiction', inplace=True)

release_year 0
title 0
genre 0
prod_method 1
creative_type 0
budget 0
dom_gross 0
int_gross 0
world_gross 0
release_month 0
dtype: int64
In [98]: # show rows where value for prod_method is null
merged_df.loc[merged_df['prod_method'].isnull()]
Out[98]:
release_date release_year title genre prod_method creative_type budget dom_gro
Dhoom Contemporary
2054 2013-12-18 2013 Action NaN 24000000 80319
3 Fiction
In [99]: # getting value counts for prod_method column

merged_df['prod_method'].value_counts()
Out[99]: Live Action 1283

Digital Animation 124
Animation/Live Action 103
Stop-Motion Animation 8
Hand Animation 4
Multiple Production Methods 2
Name: prod_method, dtype: int64
In [100]: # fill null values for prod_method with most common value for column
merged_df['prod_method'].fillna('Live Action', inplace=True)
In [101]: # saving copy of DataFrame as csv

merged_df.to_csv('./data/merged_df.csv')
Data Analysis
In order to answer the questions I had posed at the onset of this project, I performed exploratory analyses on
the basis of genre, creative type, production method, time of year, and budget. For each of these themes, I
utilized statistical methods followed by plotting of visualizations to determine the relationship between each of
these key elements and movie revenue (world gross). While I had the data for domestic and international gross
for most movies as well, I was most focused on world gross because I wanted the total amount of money that
Microsoft could expect to make on a given production.
One thing I realized as soon as I began my analysis was that my dollar amounts were not very plot-friendly. I had
to reassess my gross and budget columns at this point and add new columns for budget in millions and world
gross in millions.
In [102]: merged_df['budget_in_mil'] = merged_df['budget'] / 1000000
In [103]: merged_df['world_gross_mil'] = merged_df['world_gross']/1000000

In [104]: merged_df['world_gross_mil'].describe()
Out[104]: count 1525.00

mean 180.81
std 265.90
min 0.00
25% 32.40
50% 82.18
75% 211.74
max 2797.80
Name: world_gross_mil, dtype: float64
Genre/Creative Type/Production Method: What types of movies are

currently most successful?
Genre
In [105]: # getting value counts for genre column

merged_df['genre'].value_counts()
Out[105]: Drama 340

Adventure 270
Action 243
Comedy 223
Horror 118
Romantic Comedy 59
Black Comedy 23
Musical 19
Western 12
Documentary 11
Concert/Performance 8
Multiple Genres 2
Name: genre, dtype: int64
In [106]: # plotting the number of movies per genre in dataset
plt.figure(figsize=(8,4))
sns.countplot(x='genre', data=merged_df, palette='viridis')
plt.title('Number of Movies Per Genre', fontsize=18)
plt.ylabel('Count', fontsize=16)
plt.xlabel('Genre', fontsize=16)
plt.xticks(fontsize=14)
plt.yticks(fontsize=12)
plt.xticks(rotation=-80);
#saved in images as fig1
#plt.tight_layout()
#plt.savefig('./images/fig1.png')
The Drama genre was most common in this dataset, followed by Adventure and Action.
In [107]: # getting mean and median world gross amounts by genre
genre_stats = merged_df.groupby('genre')['world_gross_mil'].agg(['media
n', 'mean'])
genre_stats.sort_values(by='mean', ascending=False)
Out[107]:
median mean
genre
Musical 392.88 403.18
Adventure 227.53 345.30
Action 181.98 332.11
Western 75.52 124.18
Thriller/Suspense 60.28 114.65
Horror 82.12 105.72
Black Comedy 61.79 97.35
Comedy 70.55 93.16
Romantic Comedy 61.62 91.49
Drama 43.33 84.27
Concert/Performance 27.96 33.47
Documentary 10.53 23.66
Multiple Genres 1.72 1.72
In [108]: # getting Q1 (25th percentile) for world gross of each genre

a = merged_df.groupby('genre')['world_gross_mil'].quantile(0.25)

b = merged_df.groupby('genre')['world_gross_mil'].quantile(0.75)
In [110]: # getting interquartile range (IQR) for world gross of each genre
iqr = b - a
iqr
Out[110]: genre
Action 372.28
Adventure 390.04
Black Comedy 54.42
Comedy 97.13
Concert/Performance 34.49
Documentary 34.69
Drama 84.75
Horror 96.79
Multiple Genres 1.58
Musical 445.58
Romantic Comedy 93.10
Thriller/Suspense 104.24
Western 175.06
For my visualizations, I found it helpful to show my plots both with and without outliers. While outliers paint the
whole picture of the data being analyzed, it can be helpful to see the data more closesly and "zoom in" (both
literally and figuratively) on the trends without outliers.
In [111]: # generating box plot of world gross statistics per genre
sns.set_style('darkgrid')
sns.boxplot(x='genre', y='world_gross_mil', data=merged_df, palette='vir
idis_r')
plt.xticks(rotation=-80)
plt.ylabel('World Gross in Millions', fontsize=16)
plt.xlabel('Genre', fontsize = 16)
plt.title('World Gross by Genre (With Outliers)', fontsize = 18)
plt.yticks(fontsize=14);
#plt.subplots_adjust(bottom=0.2)
In [112]: # generating box plot of world gross statistics per genre
sns.boxplot(x='genre', y='world_gross_mil', data=merged_df, showfliers=F
alse, palette='viridis_r')
plt.xlabel('Genre', fontsize = 16)
plt.title('World Gross by Genre (Without Outliers)', fontsize = 18)
Based on mean and median, Musicals appeared to be the most lucrative genre for the time period I had
explored. However, as can be seen from the box plot above, this was also the genre with the largest interquartile
range (IQR), meaning that the middle values of world gross for these movies were the most spread out. After
Musicals, Action and Adventure were the highest-grossing categories, with both high means and medians.
These categories did have high IQR's as well, with a large variance in world gross values. Both Action and
Adventure also had many high-grossing outliers, as can be seen in the first box plot above.
In [113]: # creating a copy of merged_df sorted in order of release date

time_df = merged_df.sort_values(by='release_date')
In [179]: # plotting the relationship between musical movies, world gross, and rel
ease month
sns.lineplot(data=time_df.loc[time_df['genre']=='Musical'], x="release_d
ate", y="world_gross_mil", markers='o', style=True, dashes=[(2,2)], line
width=3, color='royalblue', legend=False)
plt.title('World Gross of Musical Movies by Year', fontsize=18)
plt.xlabel('Release Date', fontsize=16)
plt.ylim(-50, 1500)
for w, x, y, z in zip(time_df['genre'], time_df['release_date'], time_df
['world_gross_mil'], time_df['title']):
if (w == 'Musical') & (y>500):
plt.text(x = x, y = y+20, s = z, fontsize=12, color='black').set
_backgroundcolor('white');
#plt.tight_layout()
For musical movies in the past ten years, there were five titles that were much more successful than any others.
There was also an obvious dip in the year 2020, thanks to the global COVID-19 pandemic.
In [115]: # creating subset of DataFrame where genre is Musical
musicals = merged_df.loc[merged_df['genre']=='Musical']
musicals.sort_values(by='world_gross_mil', ascending=False).head()
Out[115]:
release_date release_year title genre prod_method creative_type budget dom_g
Digital
302 2013-11-22 2013 Frozen Musical Kids Fiction 150000000 40073
Animation
Beauty
Animation/Live
701 2017-03-17 2017 and the Musical Fantasy 160000000 50401
Action
Beast
Digital
712 2017-11-22 2017 Coco Musical Kids Fiction 175000000 21032
Animation
Digital
610 2016-11-23 2016 Moana Musical Kids Fiction 150000000 24875
Animation
Digital
9 2010-11-24 2010 Tangled Musical Kids Fiction 260000000 20082
Animation
I wanted to see which other attributes the most profitable movies in the Musical genre had shared. What I found
was that, for the top five highest-grossing Musical movies in this time period, 4 out of 5 also happened to be:
Digitally Animated, Kids Fiction, and released in November!
Based on my analysis of genre, my recommendation to Microsoft was to consider making a musical movie.
Creative Type
In [116]: # getting value counts for creative type column

merged_df['creative_type'].value_counts()

Dramatization 164
Science Fiction 160
Kids Fiction 154
Fantasy 136
Super Hero 53
Factual 18
Multiple Creative Types 5
Name: creative_type, dtype: int64
In [117]: # plotting the number of movies per creative type in dataset
sns.countplot(x='creative_type', data=merged_df, palette='viridis')
plt.title('Number of Movies Per Creative Type', fontsize=18)
plt.xlabel('Creative Type', fontsize=16)
#plt.tight_layout()
Contemporary Fiction was the most common creative type in this dataset by far.
In [118]: # getting mean and median world gross amounts by creative type
genre_stats = merged_df.groupby('creative_type')['world_gross_mil'].agg
(['median', 'mean'])
genre_stats.sort_values(by='mean', ascending=False)
Out[118]:
median mean
creative_type
Super Hero 668.00 714.09
Kids Fiction 197.54 310.47
Fantasy 156.96 288.07
Science Fiction 171.24 279.50
Contemporary Fiction 62.60 109.09
Historical Fiction 44.55 99.84
Dramatization 53.55 93.46
Factual 32.51 47.64
Multiple Creative Types 31.16 45.12
Super Hero movies did substantially better at the box office than all other creative types, with Kids Fiction
coming in second place.

a = merged_df.groupby('creative_type')['world_gross_mil'].quantile(0.25)

b = merged_df.groupby('creative_type')['world_gross_mil'].quantile(0.75)
In [121]: # getting interquartile range (IQR) for world gross of each genre
iqr = b - a
iqr
Out[121]: creative_type
Contemporary Fiction 105.01
Dramatization 99.43
Factual 59.02
Fantasy 279.87
Historical Fiction 124.20
Kids Fiction 344.91
Multiple Creative Types 56.62
Science Fiction 329.81
Super Hero 498.54
In [122]: # generating box plot of world gross statistics per creative type
sns.boxplot(x='creative_type', y='world_gross_mil', data=merged_df, pale
tte='viridis_r')
plt.xlabel('Creative Type', fontsize = 16)
plt.title('World Gross by Creative Type', fontsize = 18)
plt.ylim(None, 2000);
In [123]: # generating box plot of world gross statistics per creative type
sns.boxplot(x='creative_type', y='world_gross_mil', data=merged_df, show
fliers=False, palette='viridis_r')
plt.xlabel('Creative Type', fontsize = 16)
plt.title('World Gross by Creative Type (without outliers)', fontsize =
18)
Based on mean and median, Super Hero movies were far above all the other creative types. Kids Fiction was in
second place, with many high-grossing outliers (mentioned previously). Science Fiction and Fantasy had
relatively high means and medians as well, and both creative types also contained many high-grossing outliers.
In [124]: # plotting relationship between world gross and release month for super
hero movies
sns.lineplot(data=time_df.loc[time_df['creative_type']=='Super Hero'], x
="release_date", y="world_gross_mil", markers='o', style=True, dashes=[(
2,2)], linewidth=3, color='darkorange', legend=False)
plt.title('World Gross of Super Hero Movies by Year', fontsize=18)
plt.xlabel('Release Date', fontsize=16)
for w, x, y, z in zip(time_df['creative_type'], time_df['release_date'],
time_df['world_gross_mil'], time_df['title']):
if (w == 'Super Hero') & (y>1150):
plt.text(x = x, y = y+20, s = z, fontsize=12, color='black').set
_backgroundcolor('white');
#plt.tight_layout()
Super Hero movies seemed to do consistently well over the past ten years, although the line plot showed some
ups and downs. Still, even the lows for Super Hero movies would be considered highs for other types of movies,
so perspective is important. The plot showed seven titles that did extremely well for their movie type.
In [125]: # creating subset of DataFrame where creative type is Super Hero
superhero = merged_df.loc[merged_df['creative_type']=='Super Hero']
superhero.sort_values(by='world_gross_mil', ascending=False).head(7)
Out[125]:
Avengers: Animation/Live
900 2019-04-26 2019 Action Super Hero 400000000 8583
Endgame Action
Avengers:
Animation/Live
801 2018-04-27 2018 Infinity Action Super Hero 300000000 6788
Action
War
The Animation/Live
Avengers Action
Avengers:
Animation/Live
502 2015-05-01 2015 Age of Action Super Hero 365000000 4590
Action
Ultron
Black
800 2018-02-16 2018 Action Live Action Super Hero 200000000 7000
Panther
Iron Man Animation/Live

3 Action
Captain
602 2016-05-06 2016 America: Action Live Action Super Hero 250000000 4080
Civil War
As I had anticipated, The Avengers movies dominated the Super Hero category. All of these movies were
produced by Live Action in the Action genre, and 6 out of 7 were released in either April or May. Based on my
analysis of creative type, my recommendation to Microsoft was to explore the idea of making a Super Hero
movie.
Production Method
In [126]: # plotting the number of movies per production method in dataset
sns.countplot(x='prod_method', data=merged_df, palette='viridis')
plt.title('Number of Movies Per Production Method', fontsize=18)
plt.xlabel('Production Method', fontsize=16)
Live Action was the most common production method by far.

In [127]: # getting mean and median world gross amounts by production method
prod_stats = merged_df.groupby('prod_method')['world_gross_mil'].agg(['m
edian', 'mean'])
prod_stats.sort_values(by='mean', ascending=False)
Out[127]:
median mean
prod_method
Animation/Live Action 393.15 577.76
Digital Animation 247.91 345.38
Live Action 68.27 134.29
Stop-Motion Animation 91.54 90.09
Hand Animation 29.76 54.91
Multiple Production Methods 17.23 17.23
However, the Animation/Live Action category had the highest mean and median, with Digital Animation coming
in second.
In [128]: # generating box plot of world gross statistics per production method
sns.boxplot(x='prod_method', y='world_gross_mil', data=merged_df, palett
e='viridis_r')
plt.xlabel('Production Method', fontsize = 16)
plt.title('World Gross by Production Method', fontsize = 18)
Based on mean and median, Animation/Live Action and Digital Animation appeared to be the most successful
production methods for the time period I had explored.
In [129]: # creating subset of DataFrame where prod_method is Animation/Live Actio
n or Digital Animation
anim_df = merged_df.loc[(merged_df['prod_method']=='Animation/Live Actio
n') | (merged_df['prod_method']=='Digital Animation')]
anim_df.sort_values(by='world_gross_mil', ascending=False).head(10)
Out[129]:
Endgame Action
Star Wars
Ep. VII: Animation/Live Science
500 2015-12-18 2015 Adventure 306000000 9
The Force Action Fiction
Awakens
Avengers:
Animation/Live
Action
War
The Lion Animation/Live

901 2019-07-19 2019 Adventure Kids Fiction 260000000 5
King Action
The Animation/Live
Avengers Action
Digital
903 2019-11-22 2019 Frozen II Adventure Kids Fiction 150000000 4
Animation
Avengers:
Animation/Live
Action
Ultron
Harry
Potter
and the Animation/Live
1964 2011-07-07 2011 Adventure Fantasy 125000000 3
Deathly Action
Hallows:
Part II
Star Wars
Ep. VIII: Animation/Live Science
700 2017-12-15 2017 Adventure 200000000 6
The Last Action Fiction
Jedi
Digital
Animation
I immediately noticed some overlap between this subset and the Musical and Super Hero subsets. Many of the
top titles for this animation subset were either musicals or super hero movies as well. I also noted that while
Frozen II is generally classified as a musical like the original Frozen, the data had it listed as an adventure movie.
I wondered if there may be other children's animated movies in the data that were musical as well, but labeled
with a different genre.
In [130]: # creating subset of dataframe where production method is digital animat
ion
diganim_df = merged_df.loc[merged_df['prod_method']=='Digital Animation'
]
diganim_df.sort_values(by='world_gross_mil', ascending=False).head(10)
Out[130]:
Digital
903 2019-11-22 2019 Frozen II Adventure Kids Fiction 150000000 4
Animation
Digital
Animation
Incredibles Digital
2 Animation
Digital
505 2015-07-10 2015 Minions Adventure Kids Fiction 74000000 3
Animation
Toy Story Digital

4 Animation
Toy Story Digital

3 Animation
Despicable Digital
Me 3 Animation
Finding Digital
Dory Animation
Digital
606 2016-03-04 2016 Zootopia Adventure Kids Fiction 150000000 3
Animation
Despicable Digital
Me 2 Animation
As I had suspected, many of the top films with a production method of digital animation also had musical
components, but were labeled with a genre other than Musical. All of these were also Kids Fiction creative type,
which was the second most profitable creative type as seen above.
In [131]: # creating column to signify whether a title is animated, musical, or bo

th
merged_df.loc[merged_df['prod_method']=='Digital Animation', 'animated_o
r_musical']='animated'
merged_df.loc[merged_df['genre']=='Musical', 'animated_or_musical']='mus
ical'
merged_df.loc[(merged_df['genre']=='Musical') & (merged_df['prod_method'
]=='Digital Animation'), 'animated_or_musical']='both'
merged_df['animated_or_musical'].value_counts()
Out[131]: animated 120

musical 15
both 4
Name: animated_or_musical, dtype: int64
In [132]: # plotting distribution and world gross of animated/musical movies
g = sns.catplot(data=merged_df.loc[(merged_df['genre']=='Musical') | (me
rged_df['prod_method']=='Digital Animation')], kind="swarm", x="animated
_or_musical", y="world_gross_mil", hue='animated_or_musical', order=['an
imated', 'musical', 'both'], s=6)
g.fig.set_size_inches(9,5)
plt.title('World Gross for Animated/Musical Movies', fontsize=20)
plt.xlabel('Movie Type', fontsize=16)
plt.axhline(y=1000, ls='--', c='green');
#plt.legend(loc='upper left', fontsize=14);
#plt.tight_layout()
From reviewing the data above and analyzing this plot, I was able to determine that broadening the recomended
movie type from simply musicals to animated movies with a musical component would ensure that no high-
grossing animated movies were excluded. And again, since all of these animated and animated/musical movies
were Kids Fiction as well, they all had a combination of factors that were correlated with high world gross.
Based on my analysis of production methods, my recommendation to Microsoft was that they should prioritize
animation, whether solely digital animation or in conjunction with live action, in order to achieve the highest
possible movie gross.
Movie Type Recommendations

After careful consideration of the findings above, I had two recommendations regarding movie type:
1. Release a super hero movie.

2. Release a an animated kids fiction movie with a musical component.
Release Month: When is the most lucrative time of year to release a movie?
In [133]: # getting value counts for release month column

merged_df['release_month'].value_counts()
Out[133]: September 161

August 142
December 142
February 141
October 139
November 135
July 125
March 123
January 117
June 107
May 100
April 93
Name: release_month, dtype: int64
In [178]: # plotting the number of movies per release month in dataset
months = ['January', 'February', 'March', 'April','May','June', 'July',
'August','September', 'October', 'November', 'December']
sns.countplot(x='release_month', data=merged_df, order=months, palette=
'viridis')
plt.title('Release Month Counts for Movie Data', fontsize=20)
plt.xlabel('Month', fontsize=18)
In this dataset, movie release months were fairly evenly distributed throughout the year, with the most releases
in September and the least in April.
In [135]: # getting mean and median world gross amounts by release month
months_df = merged_df['world_gross_mil'].groupby(merged_df['release_mont
h']).agg(['median', 'mean'])
months_df.sort_values(by='mean', ascending=False)
Out[135]:
median mean
release_month
May 169.07 305.07
June 189.62 291.25
July 146.60 275.07
November 128.26 250.93
December 125.04 234.83
April 70.69 197.66
March 87.15 172.53
October 65.25 115.02
February 53.09 114.67
August 73.14 108.64
January 61.64 93.31
September 45.17 89.21

In [136]: # generating boxplots of world gross by release month
plt.figure(figsize=(12, 5))
sns.boxplot(x='release_month', y='world_gross_mil', data=merged_df, orde
r=months, palette='viridis')
plt.xlabel('Release Month', fontsize = 16)
plt.title('World Gross by Release Month', fontsize = 18)
In [137]: # generating box plots of world gross by release month with fewer outlie
rs
sns.boxplot(x='release_month', y='world_gross_mil', data=merged_df, orde
r=months, palette='viridis')
plt.xlabel('Release Month', fontsize =16)
plt.title('World Gross by Release Month (Fewer Outliers)', fontsize = 18
)
Based on release month alone, May, June, and July seemed to be the most profitable months with the highest
mean value for world box office gross. But I wondered, how much does the type of movie impact the time of
year that a movie does best?
In [138]: # creating column to signify whether a title is animated or superhero

merged_df.loc[merged_df['prod_method']=='Digital Animation', 'animated_o
r_super']='animated'
merged_df.loc[merged_df['creative_type']=='Super Hero', 'animated_or_sup
er']='super'
In [139]: # plotting world gross by month for animated and super hero movies
g = sns.catplot(data=merged_df, kind="swarm", x="release_month", y="worl
d_gross_mil", hue="animated_or_super", order=months, s=6, legend=False)
plt.title('World Gross by Month for Animated/Musical and Super Hero Movi
es', fontsize=18)
plt.xlabel('Release Month', fontsize=16)
plt.legend(loc='upper left', fontsize=14)
#plt.tight_layout()
By plotting world gross by release month for only the types of movies I was recommending, super hero and
animated, I was able to determine that there were certain trends that could be observed. While the super hero
movies seemed to do best in April and May, the animated movies seemed to do best in June, July, and
November.
In [140]: # getting mean and median world gross amounts by release month for anima
ted movies
animated_months_df = merged_df.loc[merged_df['animated_or_super']=='anim
ated']
animated_months_gb = animated_months_df['world_gross_mil'].groupby(anima
ted_months_df['release_month']).agg(['median', 'mean'])
animated_months_gb.sort_values(by='mean', ascending=False)
Out[140]:
median mean
release_month
June 679.09 656.05
November 355.09 482.04
July 251.73 400.31
May 307.56 342.84
March 269.81 326.02
December 227.54 254.13
February 191.93 222.17
April 110.47 205.19
September 185.62 201.90
January 122.75 195.74
October 97.65 173.41
August 84.24 95.30

In [141]: # generating boxplots of world gross by release month for animated movie
s
sns.boxplot(x='release_month', y='world_gross_mil', data=merged_df.loc[m
erged_df['animated_or_super']=='animated'], order=months, palette='virid
is')
plt.title('World Gross by Release Month for Animated Movies', fontsize =
18)
After taking a closer look at release month statistics for animated movies, I was able to confirm that June and
November were the highest-grossing months for an animated movie release.
In [142]: # plotting world gross and release month for animated movies
g = sns.catplot(data=merged_df.loc[merged_df['prod_method']=='Digital An
imation'], kind="swarm", x="release_month", y="world_gross_mil", color=
'royalblue', order=months, s=7, legend=False)
plt.title('World Gross by Month for Animated Movies', fontsize=18)
#plt.tight_layout()
Almost all of the animated movies over the one-billion mark for world gross were released in June or November,
with the exception of one in July and one right on the mark in March. The highest-grossing ones were released
in November.
In [143]: # getting mean and median world gross amounts by release month for super
hero movies
super_months_df = merged_df.loc[merged_df['animated_or_super']=='super']
super_months_gb = super_months_df['world_gross_mil'].groupby(super_month
s_df['release_month']).agg(['median', 'mean'])
super_months_gb.sort_values(by='mean', ascending=False)
Out[143]:
median mean
release_month
April 714.40 1192.24
May 786.68 909.47
October 853.63 853.63
December 759.50 759.50
July 690.52 722.41
November 655.95 695.60
March 743.30 663.41
February 493.44 589.13
August 485.00 442.74
June 246.36 399.82
January 229.16 229.16

In [144]: # generating box plots for world gross by relese month of super hero mov
ies
sns.boxplot(x='release_month', y='world_gross_mil', data=merged_df.loc[m
erged_df['animated_or_super']=='super'], order=months, palette='viridis_
r')
plt.title('World Gross by Release Month for Super Hero Movies', fontsize
= 18)
After taking a closer look at release month statistics for super hero movies, I was able to confirm that April and
May were the highest-grossing months for a super hero movie release.
In [145]: # plotting world gross by month for super hero movies
g = sns.catplot(data=merged_df.loc[merged_df['creative_type']=='Super He
ro'], kind="swarm", x="release_month", y="world_gross_mil", color='darko
range', order=months, s=6, legend=False)
plt.title('World Gross by Month for Super Hero Movies', fontsize=18)
#plt.tight_layout()
Most of the super hero movies over the one-billion mark were released in May. The two highest-grossing ones
were released in April. There were a couple released in other months as well.
Release Month Recommendations

After careful consideration of the findings above, I had two recommendations regarding movie release months:
1. Release a kids fiction animated movie with a musical component in June or November.
2. Release a super hero movie in April or May.
Production Budget: What budget amount tends to achieve the highest box office
gross?
The first thing I did was run the .corr method on merged_df to see if there were any strong correlations in the
numerical data. I then generated a heatmap based on these correlations.
In [146]: # generating correlations
corr = merged_df.corr()
corr
Out[146]:
release_year budget dom_gross int_gross world_gross budget_in_mil world_g
release_year 1.00 -0.06 0.01 0.02 0.02 -0.06
budget -0.06 1.00 0.69 0.78 0.78 1.00
dom_gross 0.01 0.69 1.00 0.87 0.94 0.69
int_gross 0.02 0.78 0.87 1.00 0.98 0.78
world_gross 0.02 0.78 0.94 0.98 1.00 0.78
budget_in_mil -0.06 1.00 0.69 0.78 0.78 1.00
world_gross_mil 0.02 0.78 0.94 0.98 1.00 0.78
In [147]: # plotting heatmap of correlations

plt.title("Correlation Heatmap for Dollar Amounts", fontsize=20)
sns.heatmap(corr, cmap='rocket_r');
plt.tight_layout()
The heatmap shows darker colors where two values are highly correlated and lighter colors where there is less
of a correlation. It was clear from both the correlation method and the heatmap that there was a strong
correlation between budget and world gross, as well as budget and international gross. While the correlation
between budget and domestic gross was not quite as strong, it was still right at the threshold of a high
correlation. From this visualization, I could conclude that, generally speaking, the more money you put into a
movie, the more money you are likely to make from it.
In [148]: # plotting world gross by budget in terms of genre with line of best fit
sns.lmplot(x='budget_in_mil', y='world_gross_mil', data=merged_df, aspec
t=4, line_kws={'color': 'crimson'})
plt.title('World Gross by Budget', fontsize=20)
plt.xlabel('Budget in Millions', fontsize=18)
plt.ylim(0,None)
plt.xlim(0, None);
#plt.tight_layout()
The regression line in this plot shows the general increase in movie gross as the budget increases. While this
varies greatly from movie to movie, it was a good starting point as I worked towards combining my findings and
forming my recommendations. This plot confirmed my hypothesis that a higher production budget typically
leads to a higher box office gross.
In [149]: # plotting world gross by budget for animated and super hero movies
g = sns.lmplot(x='budget_in_mil', y='world_gross_mil', data=merged_df.lo
c[(merged_df['prod_method']=='Digital Animation') | (merged_df['creative
_type']=='Super Hero')], hue='animated_or_super', aspect=3, legend=False
)
plt.title('World Gross by Budget for Animated and Super Hero Movies', fo
ntsize=20)
plt.legend(loc='upper left', fontsize=14)
#for w, x, y, z in zip(merged_df['animated_or_super'], merged_df['budget
_in_mil'], merged_df['world_gross_mil'], merged_df['title']):
#if (w == 'animated') & (y>1250):
#plt.text(x = x+5, y = y-15, s = z.upper(), fontsize=14, color
='black')
#for w, x, y, z in zip(merged_df['animated_or_super'], merged_df['budget
_in_mil'], merged_df['world_gross_mil'], merged_df['title']):
#if (w == 'super') & (y>1250):
#plt.text(x = x+5, y = y-15, s = z.upper(), fontsize=14, color
='black')
#plt.tight_layout()
For the specific movie types we are looking at, we can see that super hero movies benefit from a very large
budget, while animated movies benefit from a moderately large budget. Both regression lines show that as the
budget increases, the world gross tends to increase as well.
In [150]: # getting descriptive stats for animated movie budgets
merged_df.loc[merged_df['prod_method']=='Digital Animation', 'budget_in_
mil'].describe()
Out[150]: count 124.00
mean 87.68
std 59.33
min 0.00
25% 37.50
50% 76.75
75% 135.00
max 260.00
Name: budget_in_mil, dtype: float64
In [151]: # plotting world gross by budget for animated movies

sns.lmplot(x='budget_in_mil', y='world_gross_mil', data=merged_df.loc[me
rged_df['prod_method']=='Digital Animation'], aspect=3)
plt.title('World Gross by Budget for Animated Movies', fontsize=20)
plt.xlim(0, None)
plt.axhline(y=1000, ls='--', c='green')
for v, w, x, y, z in zip(merged_df['genre'], merged_df['prod_method'], m
erged_df['budget_in_mil'], merged_df['world_gross_mil'], merged_df['titl
e']):
if (z=='Frozen'):
plt.text(x=x-15, y=y-15, s=z.upper(), fontsize=12, color='black'
)
elif z=='Toy Story 3':
)
elif z=='Zootopia':
plt.text(x=x+2, y=y, s=z.upper(), fontsize=12, color='black')
elif (z=='Toy Story 4'):
plt.text(x = x+2, y = y, s = z.upper(), fontsize=12, color='blac
k')
elif (w=='Digital Animation') and y>1000:
plt.text(x = x+2, y = y-15, s = z.upper(), fontsize=12, color='b
lack');
#plt.tight_layout()
The mean budget for animated movies is 87.68 million dollars. For the highest-grossing animated movies, the
optimal budget is between 75 million dollars and 200 million dollars, with the two highest-grossing movies
having a budget of 150 million dollars.
In [152]: # getting descriptive statistics for super hero movie budgets

merged_df.loc[merged_df['creative_type']=='Super Hero', 'budget_in_mil']
.describe()
Out[152]: count 53.00

mean 165.21
std 76.16
min 0.00
25% 125.00
50% 165.00
75% 200.00
max 400.00
Name: budget_in_mil, dtype: float64
In [153]: # plotting world gross by budget for super hero movies
sns.lmplot(x='budget_in_mil', y='world_gross_mil', data=merged_df.loc[me
rged_df['creative_type']=='Super Hero'], line_kws={'color': 'darkorange'
}, scatter_kws={'color': 'darkorange'}, aspect=3)
plt.title('World Gross by Budget for Super Hero Movies', fontsize=20)
plt.xlim(0, None)
for w, x, y, z in zip(merged_df['animated_or_super'], merged_df['budget_
in_mil'], merged_df['world_gross_mil'], merged_df['title']):
if (z=='Avengers: Endgame'):
)
elif (z=='Avengers: Infinity War'):
plt.text(x=x+2, y=y-15, s=z.upper(), fontsize=12, color='black')
elif (z=='Avengers: Age of Ultron'):
)
elif (z=='The Avengers'):
elif (z=='Spider-Man: Far From Home'):
)
elif (z=='Aquaman'):
elif (z=='Captain Marvel'):
elif (z=='The Dark Knight Rises'):
elif (w == 'super') & (y>1000):
lack')
#plt.tight_layout()
The mean budget for super hero movies is 165.21 million dollars--almost double the mean budget for an
animated movie. For the highest-grossing super hero movies, the optimal budget is between 200 million dollars
and 400 million dollars, with the two highest-grossing movies having budgets between 300 and 400 million
dollars.
Budget Recommendations
After careful consideration of the findings above, my main observation was that a higher production budget
leads to a higher gross. I had two specific recommendations regarding production budget for the preferred
movie types:
1. For the highest box office gross, an animated move should have a budget of 75 to 200 million dollars.
2. For the highest box office gross, a super hero movie should have a budget of 200 to 400 million dollars.
Putting It All Together!
Now that I had answers to my three main questions regarding the business problem, I wanted to put them all
together into one visualization.
This plot shows the specifics of the top-grossing movies in the animation category. As discussed previously, the
highest-grossing movies were mostly released in June or November and had budgets between 75 and 200
million dollars.
In [173]: # plotting world gross by budget and release month for animated
g = sns.relplot(x='budget_in_mil', y='world_gross_mil', data=merged_df.l
oc[(merged_df['prod_method']=='Digital Animation')], hue='release_month'
, hue_order=['January', 'February', 'March', 'April', 'May', 'June', 'Ju
ly', 'August', 'September', 'October', 'November', 'December'], s=120, a
spect=3, palette='Paired')
plt.title('World Gross by Budget and Release Month for Animated Movies',
fontsize=20)
plt.xlim(0, None)
for v, w, x, y, z in zip(merged_df['genre'], merged_df['prod_method'], m
erged_df['budget_in_mil'], merged_df['world_gross_mil'], merged_df['titl
e']):
if (z=='Frozen'):
)
elif z=='Toy Story 3':
)
elif z=='Zootopia':
elif (w=='Digital Animation') and y>1000:
lack')
#plt.tight_layout()
In [175]: # plotting world gross by budget and release month for super hero movies
g = sns.relplot(x='budget_in_mil', y='world_gross_mil', data=merged_df.l
oc[merged_df['creative_type']=='Super Hero'], hue='release_month', hue_o
rder=['January', 'February', 'March', 'April', 'May', 'June', 'July', 'A
ugust', 'September', 'October', 'November', 'December'], s=130, aspect=3
, palette='Paired')
plt.title('World Gross by Budget and Release Month for Super Hero Movie
s', fontsize=20)
plt.xlim(0, None)
for w, x, y, z in zip(merged_df['animated_or_super'], merged_df['budget_
in_mil'], merged_df['world_gross_mil'], merged_df['title']):
if (z=='Avengers: Endgame'):
)
elif (z=='Avengers: Infinity War'):
elif (z=='Avengers: Age of Ultron'):
)
elif (z=='The Avengers'):
elif (z=='Spider-Man: Far From Home'):
)
elif (z=='Aquaman'):
plt.text(x=x, y=y+10, s=z.upper(), fontsize=12, color='black')
elif (z=='Captain Marvel'):
elif (w == 'super') & (y>1000):
lack')
#plt.tight_layout()
This plot shows the specifics of the top-grossing movies in the super hero category. As discussed previously,
the highest-grossing movies were mostly released in April, May, or July and most had budgets between 200 and
400 million dollars.
These visualizations show the relationship between all the key factors that I analyzed: movie type, release
month, and production budget. Based on the top movies that meet these criteria, super hero movies do best
when given a very high budget and are released in April or May. Animated musicals find the most success when
released in June or November, and don't necessarily need quite as high of a budget to achieve a high gross,
although their gross is also comparatively lower than that of super hero movies.
Additional Attributes: Based on these findings, what else do top-grossing movies

have in common?
Directors
Now that we know that a high-budget super hero movie released in April or May is the most rewarding
combination of factors, how can we make this movie even better and ensure that it reaches its full potential?
One way that we can increase our chances of success is by hiring the right director for this type of movie.
To further this point, I utilized the TMDb API to obtain the director names for all the super hero movies in my
dataset.
In [156]: def get_director(title):
"""
Updates director information for movie in dataframe.

Retrieves TMDB movie_id for title.
Retrieves director information based on movie_id.
Adds director information to a list.
Converts director information from list to string.
Adds new director value as string to movie's row in dataframe.
Parameters:
Returns:
Updated cells in Pandas DataFrame.
"""
query={title_r}"
movie_id = response.json()['results'][0]['id']
url2 = f"https://api.themoviedb.org/3/movie/{movie_id}/credits?a
pi_key={api_key}"
response2 = requests.get(url2)
crew = response2.json()['crew']
directors = []
for member in crew:
if member['job'] == 'Director':
directors.append(member['name'])
d = str(directors)
d = d.replace('[', '').replace(']', '').replace("'","")
merged_df.loc[merged_df['title']==title, 'director'] = d
else:
pass
In [157]: # creating a list of all the super hero movie titles

superhero_titles = [title for title in superhero['title']]
In [158]: # getting director info for movies in list and updating data accordingly
for title in superhero_titles:
get_director(title)
In [159]: # getting director value counts
merged_df.director.value_counts()
Out[159]: Anthony Russo, Joe Russo 4

Zack Snyder 3
James Gunn 2
Marc Webb 2
Bryan Singer 2
James Mangold 2
Jon Watts 2
Joss Whedon 2
Dave Green 2
Taika Waititi 2
Peyton Reed 2
Tim Story 1
Ruben Fleischer 1
Martin Campbell 1
Alan Taylor 1
Patty Jenkins 1
Ryan Coogler 1
Jon Favreau 1
Shane Black 1
Laurent Bouzereau 1
Josh Boone 1
Ryan Fleck, Anna Boden 1
Jonathan Liebesman 1
Scott Derrickson 1
Simon Kinberg 1
James Wan 1
David Ayer 1
David Leitch 1
Dave Wilson 1
Neil Marshall 1
Tim Miller 1
David F. Sandberg 1
Christopher Nolan 1
Michel Gondry 1
Joe Johnston 1
Chris Williams, Don Hall 1
Name: director, dtype: int64
In [160]: # getting mean and median stats for directors
director_stats = merged_df.groupby('director')['world_gross_mil'].agg([
'median', 'mean']).reset_index()
director_stats.sort_values(by='mean', ascending=False)
Out[160]:
director median mean
1 Anthony Russo, Joe Russo 1598.23 1677.17
18 Joss Whedon 1455.60 1455.60
27 Ryan Coogler 1339.73 1339.73
30 Shane Black 1215.39 1215.39
12 James Wan 1143.97 1143.97
28 Ryan Fleck, Anna Boden 1129.73 1129.73
4 Christopher Nolan 1082.23 1082.23
15 Jon Watts 1005.06 1005.06
26 Ruben Fleischer 853.63 853.63
10 James Gunn 819.98 819.98
24 Patty Jenkins 818.79 818.79
9 David Leitch 786.68 786.68
33 Tim Miller 785.03 785.03
7 David Ayer 746.85 746.85
20 Marc Webb 733.44 733.44
35 Zack Snyder 668.00 732.11
29 Scott Derrickson 676.35 676.35
32 Taika Waititi 650.36 650.36
3 Chris Williams, Don Hall 649.69 649.69
2 Bryan Singer 645.20 645.20
0 Alan Taylor 644.60 644.60
14 Jon Favreau 621.16 621.16
25 Peyton Reed 571.00 571.00
11 James Mangold 515.33 515.33
16 Jonathan Liebesman 485.00 485.00
13 Joe Johnston 370.57 370.57
8 David F. Sandberg 363.66 363.66
19 Laurent Bouzereau 355.41 355.41
31 Simon Kinberg 246.36 246.36
5 Dave Green 245.33 245.33
22 Michel Gondry 229.16 229.16
21 Martin Campbell 219.54 219.54
34 Tim Story 167.85 167.85
17 Josh Boone 43.15 43.15

director median mean
23 Neil Marshall 40.79 40.79
6 Dave Wilson 37.32 37.32
In [161]: # plotting directors in order of highest-grossing, descending

ordered_df = director_stats.sort_values(by='mean')
my_range=range(1,len(director_stats.index)+1)
plt.hlines(y=my_range, xmin=0, xmax=ordered_df['mean'], color='mediumspr
inggreen')
plt.plot(ordered_df['mean'], my_range, "o")
plt.yticks(my_range, ordered_df['director'], fontsize=18)
plt.title("Mean Movie Gross by Director", fontsize=22)
plt.xlabel('Mean Movie Gross', fontsize=18)
plt.ylabel('Director', fontsize=18)
plt.xlim(0, 1750);
#plt.tight_layout()
In [162]: # seeing which movies were directed by Anthony Russo, Joe Russo
merged_df.loc[merged_df['director']=='Anthony Russo, Joe Russo']
Out[162]:
Captain
America:
403 2014-04-04 2014 The Action Live Action Super Hero 170000000 25974
Winter
Soldier
Captain
602 2016-05-06 2016 America: Action Live Action Super Hero 250000000 4080
Civil War
Avengers:
Animation/Live
Action
War
Endgame Action
In [163]: # seeing which movies were directed by Joss Whedon

merged_df.loc[merged_df['director']=='Joss Whedon']
Out[163]:
The Animation/Live
Avengers Action
Avengers:
Animation/Live
Action
Ultron
As both the director_stats DataFrame and the Mean Movie Gross by Director plot show, for the past ten years,
the top five directors (or combinations of directors) for super hero movies by mean world gross are:
1. Anthony Russo, Joe Russo

2. Joss Whedon
3. Ryan Coogler
4. Shane Black
5. James Wan
Hiring one of these top directors to work on a super hero movie can further increase the chances of a
successful movie venture.
Composers
For the animated movie, I had found that the musical ones tended to have the highest box office gross. To that
end, I again used the TMDb API, this time to obtain composer information for all the animated movies in my
dataset. Because not every animated movie was a musical one, they did not all have composer information
available. Only the titles that had an associated composer value available were used in the following analysis.
In [164]: def get_composer(title):

"""
Updates composer information for movie in dataframe.

Retrieves TMDB movie_id for title.
Retrieves composer information based on movie_id.
Adds composer information to a list.
Converts composer information from list to string.
Adds new composer value as string to movie's row in dataframe.
Parameters:
Returns:
Updated cells in Pandas DataFrame.
"""
query={title_r}"
movie_id = response.json()['results'][0]['id']
url2 = f"https://api.themoviedb.org/3/movie/{movie_id}/credits?a
pi_key={api_key}"
response2 = requests.get(url2)
crew = response2.json()['crew']
composers = []
for member in crew:
if member['job'] == 'Original Music Composer':
composers.append(member['name'])
c = str(composers)
c = c.replace('[', '').replace(']', '').replace("'","")
merged_df.loc[merged_df['title']==title, 'composer'] = c
else:
pass
In [165]: # creating a list of animated titles

animation_titles = [title for title in merged_df['title'].loc[merged_df[
'prod_method'] == 'Digital Animation']]
In [166]: # obtaining composer information for animated titles

for title in animation_titles:
get_composer(title)
In [167]: composer_stats = merged_df.groupby('composer')['world_gross_mil'].agg([
'median', 'mean']).reset_index()
composer_stats.sort_values(by='mean', ascending=False)
Out[167]:
composer median mean
8 Christophe Beck, Robert Lopez, Kristen Anderso... 1357.67 1357.67
20 Heitor Pereira, Pharrell Williams 1032.60 1032.60
42 Thomas Newman 1025.01 1025.01
32 Michael Giacchino 826.70 804.89
38 Randy Newman 743.59 733.51
14 Hans Zimmer, John Powell 664.84 664.84
2 Alexandre Desplat 659.88 659.88
29 Mark Mancina 636.34 636.34
25 Joby Talbot 632.48 632.48
1 Alan Menken, Kevin Kliesch 585.73 585.73
19 Heitor Pereira 352.33 570.57
22 Henry Jackman 542.14 557.62
36 Patrick Doyle 554.61 554.61
21 Heitor Pereira, Pharrell Williams, Chris Westlake 543.46 543.46
17 Hans Zimmer, Steve Mazzaro 527.97 527.97
18 Harry Gregson-Williams 452.98 452.98
26 John Powell 490.18 432.12
30 Mark Mothersbaugh 423.30 390.96
13 Hans Zimmer 383.45 379.62
27 John Williams 373.99 373.99
15 Hans Zimmer, Lorne Balfe 321.89 321.89
28 Lorne Balfe 310.57 310.57
3 Alexandre Desplat, Michael Allen 306.90 306.90
9 Danny Elfman 262.79 262.79
41 Theodore Shapiro 254.25 254.25
7 Christophe Beck 250.09 250.09
34 Mychael Danna, Jeff Danna 191.64 212.32
23 James Newton Howard, Chris Bacon 193.74 193.74
39 Rupert Gregson-Williams 187.89 187.89
0 Alan Menken, Christopher Lennertz 141.34 141.34
10 David Hirschfelder 139.72 139.72
35 Norihito Sumitomo 122.75 122.75
40 Steven Price 115.12 115.12
16 Hans Zimmer, Richard Harvey 102.03 102.03

composer median mean
12 Gustavo Santaolalla 97.65 97.65
37 Ramin Djawadi 71.59 71.59
11 David Newman 53.05 53.05
24 Jared Faber 51.62 51.62
6 Chris Bacon 48.96 48.96
31 Matthieu Chedid 27.00 27.00
5 Bruno Coulais 14.53 14.53
33 Michiru Ōshima 4.00 4.00
4 Atli Örvarsson 0.65 0.65
43 Walter Murphy 0.07 0.07
In [168]: plt.figure(figsize=(18, 16))

ordered_df2 = composer_stats.sort_values(by='mean')
my_range=range(1,len(composer_stats.index)+1)
plt.hlines(y=my_range, xmin=0, xmax=ordered_df2['mean'], color='darkviol
et')
plt.plot(ordered_df2['mean'], my_range, "o")
plt.yticks(my_range, ordered_df2['composer'], fontsize=18)
plt.title("Mean Movie Gross by Composer", fontsize=22)
plt.xlabel('Mean Movie Gross', fontsize=18)
plt.ylabel('Composer', fontsize=18)
plt.xlim(0, 1400);
#plt.tight_layout()
In [169]: # seeing which movies had music composed by Christophe Beck, Robert Lope
z, Kristen Anderson-Lopez
merged_df.loc[merged_df['composer']=='Christophe Beck, Robert Lopez, Kri
sten Anderson-Lopez']
Out[169]:
Digital
Animation
Frozen Digital
II Animation
In [170]: # seeing which movies had music composed by Christophe Beck, Robert Lope
z, Kristen Anderson-Lopez
merged_df.loc[merged_df['composer']=='Heitor Pereira, Pharrell Williams'
]
Out[170]:
Despicable Digital
Me 3 Animation
As both the composer_stats DataFrame and the Mean Movie Gross by Composer plot show, for the past ten
years, the top five composers (or combinations of composers) for animated movies by mean world gross are:
1. Christophe Beck, Robert Lopez, Kristen Anderson-Lopez

2. Heitor Pereira, Pharrell Williams
3. Thomas Newman
4. Michael Giacchino
5. Randy Newman
These composers have a proven track record of success and should be hired to work on an animated musical
to increase box office success.
Now that I had my general answers to each of the primary business questions that were laid out at the onset of
this project, it was time to combine them into an actionable plan for Microsoft's up-and-coming movie studio.
Evaluation
Microsoft's ticket to worldwide box office success can be attained by releasing a Super Hero movie in April or
May, or by releasing an animated children's musical movie in June or November. My findings have shown that
the more funds a studio invests in their movie production, the more money (worldwide box office gross) they are
likely to receive as a result. Though there are some outliers, the majority of high-grossing movies for these
particular types of movies are ones with higher budgets. For an animated movie, this amount is between 75 and
200 million dollars. For a super hero movie, that amount is between 200 and 400 million dollars. Throughout my
analysis, I have found that the optimal time to release a movie depends on the type of movie that it is. While
animated musical movies tend to fare very well in November and June, super hero movies have seen immense
box office success in April and May. I chose to investigate a bit further and narrow down some additional
attributes that may increase a movie's value, such as the highest-grossing composers and directors for the past
ten years, based on mean world gross.
I am confident that the results I extrapolated from this analysis would generalize beyond the data that I have,
with the exception of this year and next year due to the COVID-19 pandemic. By looking at the data up until this
year, the trends and correlations I found were true for the past ten years, so I am confident that they will again
be true once the world returns to some semblance of normalcy.
If the recommendations that I made are put to use, I am confident that Microsoft will have a successful break
into the movie-making industry. From the data, it is clear that all the attributes I have discussed are correlated
with high worldwide box office gross, which is exactly what Microsoft will want for their first movies and beyond.
Conclusion
In conclusion, I would recommend that Microsoft release one of the following two movies, each with four
specific recommendations that have proven to be successful combinations:
Movie Option #1
an animated kids fiction movie
with a production budget of 75 to 200 million dollars
released in June or November
containing songs by a high-grossing composer with a track record of successful work in digital animation
movies, such as Christophe Beck, Robert Lopez, and Kristen Anderson-Lopez, or Heitor Pereira and
Pharrell Williams
Movie Option #2
a live action/animation super hero movie
with a production budget of 200 to 400 million dollars
released in April or May
directed by a top-grossing director with a history of proven successful superhero movies, such as Anthony
Russo, Joe Russo, or Joss Whedon
While the past ten years of data show that this should be a good recipe for success, one limitation is that we are
currently in a global pandemic, which has negatively affected many facets of the global economy. The
visualizations above displaying movie gross over time clearly show a significant drop in movie gross for this year
(2020). However, since movies take quite a bit of time to produce, the expectation is that the market will be
trending in the right direction by the time a future movie would be released.
In the future, this analysis could be improved by adding additional data as it becomes available. It could also be
expanded upon by determining how much money there is to be made on a streaming platform movie release
while theaters remain at low audience capacity.

Movie Notebook

Uploaded by

Copyright:

Available Formats

Movie Notebook

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Movie Notebook

Uploaded by

Copyright:

Available Formats

Current Movie Industry Analysis

Author: Diane Tunnicliﬀe

OpusData Movie Data

In [3]: # reading the csv file

1 Krrish 2006 58540100 10000000 1430721

4 Saw III 2006 5840100 10000000 80238724

In [5]: # getting counts for each value in genre column

Out[5]: Drama 471

In [6]: # generating descriptive statistics for domestic box office values

Out[6]: count 1936.00

Out[7]: count 1936.00

The-Numbers Web Scraping

Mar 5, Alice in Animation/Live

In [10]: # getting descriptive statistics for DataFrame

count 100.00 100.00

mean 50.50 2010.00

std 29.01 0.00

min 1.00 2010.00

25% 25.75 2010.00

50% 50.50 2010.00

75% 75.25 2010.00

max 100.00 2010.00

In [13]: # getting value counts for production method column

Out[13]: Live Action 85

In [14]: # getting value counts for creative type column

Out[14]: Contemporary Fiction 56

Adds year input to url and scrapes resulting table.

In [17]: # scraping 2012 data and compiling into DataFrame

May 4, The Animation/Live

Mar 23, The Hunger Scienc

In [18]: # scraping 2013 data and compiling into DataFrame

Jun 14, Man of

In [19]: # scraping 2014 data and compiling into DataFrame

Dec 25, American

Feb 7, The Lego

In [20]: # scraping 2015 data and compiling into DataFrame

Jun 12, Jurassic Science

Jun 19, Inside

Jun 17, Finding

In [22]: # scraping 2017 data and compiling into DataFrame

In [23]: # scraping 2018 data and compiling into DataFrame

Feb 16, Black

Jun 15, Incredibles

Dec 21, Animation/Live

In [24]: # scraping 2019 data and compiling into DataFrame

Apr 26, Avengers: Animation/Live

Jul 19, The Lion Animation/Live

Jun 21, Toy Story

In [25]: # scraping 2020 data and compiling into DataFrame

Jan 17, Bad Boys Contemporary

Feb 14, Sonic The Animation/Live

Jan 17, Animation/Live

Feb 28, The Invisible Science

OpusData Movie Data

In [26]: # dropping unnecessary columns

In [27]: # renaming certain columns

Out[28]: Index(['title', 'production_year', 'budget', 'dom_gross', 'int_gross',

In [29]: # saving copy of DataFrame as csv file

Scraped data from The-Numbers.com