Visualization: Are more movies set in the future or the past?

Visualization: Are more movies set in the future or the past?#

Motivation#

I was curious about the balance between media set in the future vs movies set in the past. Luckily, people on Wikipedia have tagged movies both for when they are produced/released, as well as whether they are set in certain years, decades or centuries.

For example, Category:Films_set_in_1945 contains list of movies set in 1945, while Category:2022_films contains list of movies from 2022.

Obtain data#

The data are scraped from Wikipedia using category pages.

The produced years are from Category:Films_by_year.

The set-in years are from 3 categories:

  • Category:Films_by_century_of_setting

  • Category:Films_by_decade_of_setting

  • Category:Films_by_year_of_setting

See the scrape.ipynb notebook for more information.

Import modules#

Hide code cell source 
import os, re, pickle, time
import numpy as np
import pandas as pd

import matplotlib.pyplot as plt
from matplotlib import rcParams
import seaborn as sns
Hide code cell source 
# plot configs
rcParams['font.family'] = 'Overpass Nerd Font'
rcParams['font.size'] = 18
rcParams['axes.titlesize'] = 20
rcParams['axes.labelsize'] = 18
rcParams['axes.linewidth'] = 1.5
rcParams['lines.linewidth'] = 1.5
rcParams['lines.markersize'] = 20
rcParams['patch.linewidth'] = 1.5
rcParams['xtick.labelsize'] = 18
rcParams['ytick.labelsize'] = 18
rcParams['xtick.major.width'] = 2
rcParams['xtick.minor.width'] = 2
rcParams['ytick.major.width'] = 2
rcParams['ytick.minor.width'] = 2
rcParams['savefig.dpi'] = 200
rcParams['savefig.transparent'] = False
rcParams['savefig.facecolor'] = 'white'
rcParams['savefig.format'] = 'svg'
rcParams['savefig.pad_inches'] = 0.5
rcParams['savefig.bbox'] = 'tight'

Load data#

data_path = 'data/films_set_in_and_produced.csv'
df = pd.read_csv(data_path)
df
pageid title source_year_setin year_produced year_setin year_setin_type
0 3217 Army of Darkness 13th 1992 1200 century
1 3333 The Birth of a Nation 1860s 1915 1860 decade
2 3333 The Birth of a Nation 19th 1915 1800 century
3 3333 The Birth of a Nation 1870s 1915 1870 decade
4 3746 Blade Runner 2019 1982 2019 year
... ... ... ... ... ... ...
20150 74597836 The Sacrifice Game 1971 2023 1971 year
20151 74607682 Joachim and the Apocalypse 12th 2023 1100 century
20152 74612478 The Kitchen (2023 film) 2040s 2023 2040 decade
20153 74619750 Jailer (2023 Malayalam film) 1950s 2023 1950 decade
20154 74621258 Robot Dreams (film) 1980s 2023 1980 decade

20155 rows × 6 columns

Brief notes on data#

It is intuitive that a movie can be set in multiple years.

But, a movie entry (by pageid) can be appear in multiple Wikipedia categories ABCD films. Technically they’re not always produced in such years.

As I understand there may be different scenarios as to how this happens, for example release dates or an entry is a TV series.

df.query('pageid == 67231914')
pageid title source_year_setin year_produced year_setin year_setin_type
19286 67231914 Edge of the World (2021 film) 19th 2021 1800 century
19287 67231914 Edge of the World (2021 film) 19th 2023 1800 century

For example, Edge of the World (2021 film) belongs to both Category: 2021 films and Category: 2023 films. The Wikipedia page says

Edge of the World was released in the US […] on 4 June 2021, […] then in the UK […] on 21 June 2021.

In Malaysia and Brunei, the film has been slated as Rajah in the countries and scheduled on 9 March 2023 in cinemas.

Similarly, the Wikipedia box says

Release dates

4 June 2021 (US)

9 March 2023 (Malaysia and Brunei)

So this is due to different release dates.

df.query('pageid == 6819660')['year_produced'].unique()

array([1967, 1968, 1969, 1970, 1971, 1973])

Another example is Adventures of Mowgli, with 6 different years for production/release. This is not entirely like the above example with similar content and different release dates, but actually 5 different contents (1967-1971) and then later combined into one content released in 1973.

Adventures of Mowgli (Russian: Маугли; also spelled Maugli) is an animated feature-length story originally released as five animated shorts of about 20 minutes each between 1967 and 1971 in the Soviet Union. […]

In 1973, the five films were combined into a single 96-minute feature film.

Hence, take note that year_produced should be taken with a jitter of a few years in mind in some cases. Plus, it is not correct that is is produced, it may be a combination of production and release, maybe even (re)distribution.

For simplicity, I will not consider movie titles and ignore this for the time being, and instead analyze all pairs of (year_produced, year_setin).

Get time difference#

This is to calculate the time difference between year_setin and year_produced, with consideration of units.

If the time difference is positive, the movie is set in the future of the time that it is produced/released, e.g. post-apocalyptic movies.

If it is negative, the movie is set in the past, e.g. war movies.

df['year_diff'] = df['year_setin'] - df['year_produced']
df['diff_unit'] = df['year_setin_type']
df['num_years'] = df['diff_unit'].map({'century': 100, 'decade': 10, 'year': 1})
df['time_diff'] = df['year_diff'] / df['num_years']

df
pageid title source_year_setin year_produced year_setin year_setin_type year_diff diff_unit num_years time_diff
0 3217 Army of Darkness 13th 1992 1200 century -792 century 100 -7.92
1 3333 The Birth of a Nation 1860s 1915 1860 decade -55 decade 10 -5.50
2 3333 The Birth of a Nation 19th 1915 1800 century -115 century 100 -1.15
3 3333 The Birth of a Nation 1870s 1915 1870 decade -45 decade 10 -4.50
4 3746 Blade Runner 2019 1982 2019 year 37 year 1 37.00
... ... ... ... ... ... ... ... ... ... ...
20150 74597836 The Sacrifice Game 1971 2023 1971 year -52 year 1 -52.00
20151 74607682 Joachim and the Apocalypse 12th 2023 1100 century -923 century 100 -9.23
20152 74612478 The Kitchen (2023 film) 2040s 2023 2040 decade 17 decade 10 1.70
20153 74619750 Jailer (2023 Malayalam film) 1950s 2023 1950 decade -73 decade 10 -7.30
20154 74621258 Robot Dreams (film) 1980s 2023 1980 decade -43 decade 10 -4.30

20155 rows × 10 columns

Visualization#

Plot distribution of years set-in#

Hide code cell source 
plt.figure(figsize=(18,5))

count_offset = 15

select_periods = {
    'US Civil War': [1861, 1865],
    'World War I': [1914, 1918],
    'World War II': [1939, 1945],
}

select_years = {
    1969: 'Apollo 11\nMoon Landing',
    1999: 'before Y2K',
    2012: 'defunct apocalypse',
    2023: 'present year'
}

select_year_list = list(select_years.keys())

df_year_target = df.query('year_setin_type == "year"')

select_counts = (
    df_year_target
    .query('year_setin in @select_year_list')
    .value_counts('year_setin')
    .to_frame('count')
    .join(
        pd.Series(select_years).to_frame('desc'), 
        how='outer'
    )
    .reset_index()
    .rename(columns={'index': 'year_setin'})
)

select_counts['count_offset'] = select_counts['count'] + count_offset

    
sns.histplot(
    df_year_target.query('year_setin > 1800 and year_setin < 2050'),
    x='year_setin',
    discrete=True,
    element='step',
    color='k',
    edgecolor='none',
    alpha=0.1
)


sns.scatterplot(
    select_counts,
    x='year_setin',
    y='count_offset',
    c='#12a4b6',
    s=25,
    marker='o',
    edgecolor='none'
)

for _, r in select_counts.iterrows():
    t_ha = 'left'
    t_va = 'center'
    t_yr = r['year_setin']
    c_os = r['count_offset']
    desc = r['desc']
    if pd.isna(desc):
        continue
    desc = f"{t_yr} ({desc})"
    
    x_text = t_yr + 2
    y_text = c_os
    
    if t_yr == 1969:
        # to avoid blocking
        x_text = t_yr
        y_text = c_os + 10        
        t_ha='center'
        t_va='bottom'
        
    plt.text(
        x_text, y_text, desc,
        fontdict=dict(
            size=15,
            ha=t_ha,
            va=t_va
        )
    )

for desc, (y0, ye) in select_periods.items():
    c_os = (
        df_year_target
        .query('year_setin >= @y0 and year_setin <= @ye')
        .value_counts('year_setin')
        .max()
    ) + count_offset
    
    plt.plot(
        [y0-0.5,ye+0.5], [c_os, c_os],
        color='#12a4b6', lw=2, 
        marker='.', markersize=10
    )
    
    desc = f'{desc}\n({y0} - {ye})'
    
    plt.text(
        (y0 + ye)/2, c_os + count_offset/2, desc,
        fontdict=dict(
            size=15,
            ha='center',
            va='bottom',
        )
    )

plt.xlabel('Years that movies are set in (restricted to 1800 - 2050, set-in type = "year")')
plt.ylabel('Number of movies')
plt.suptitle('Distribution of years that movies are set in', fontsize=20)

sns.despine(trim=True, offset=10)
plt.tight_layout()

plt.savefig('figures/distrib-years-setin.pdf')
plt.savefig('figures/distrib-years-setin.png')
../_images/0ed9796ff0da240ce31e4af4a27e28a9cd3728d5c0a1bfb96b2f92bd979d0b9c.png 
(
    df_year_target
    .query('year_setin in [1944,1945]')
    .value_counts('year_setin')
)

year_setin
1944    195
1945    167
dtype: int64

Many movies tend to be set in war periods, especially World War II (1939 - 1945). Interesting, there are more movies set in 1944 (195 movies) than in 1945 (167 movies). Not sure why.

Additionally, besides war periods, some notable events are annotated in the distribution and have high peaks as well, such as:

  • 1969: Apollo 11 Moon Landing

  • 1999: the year before Y2K. I annotated as such because I suspect this is why there’s a high peak in 1999 instead of 2000.

  • 2012: the supposed end-of-the-world

Plot time difference between set-in and produced#

Hide code cell source 
plt.figure(figsize=(18, 5))

setin_types = ['year', 'decade', 'century']
t_range = [1880, 2030]

limits = dict(
    year = dict(x1=t_range, x2=[-0.001,0.05], y=[-450, 150]),
    decade = dict(x1=t_range, x2=[-0.01,0.6], y=[-60, 20]),
    century = dict(x1=t_range, x2=[-0.01,0.6], y=[-30, 10]),
)

num_cols = len(setin_types)

for i, st in enumerate(setin_types):
    ax = plt.subplot(1, num_cols, i+1)
    df_st = df.query('year_setin_type == @st')
    
    sns.scatterplot(
        df_st,
        x = 'year_produced',
        y = 'time_diff',
        s = 30,
        c = 'k',
        alpha = 0.02,
        edgecolor = 'none',
        ax = ax
    )
    
    ax.hlines(0, *t_range, colors='k', linestyles=':', alpha=0.4)
    
    ax.set_xlim(limits[st]['x1'])
    ax.set_ylim(limits[st]['y'])

    ax_twin = ax.twiny()
    sns.kdeplot(
        df_st,
        y = 'time_diff',
        color = 'k',
        ax = ax_twin
    )
    
    scale_bar_y = limits[st]['y'][-1]/2
    scale_bar_x = [abs(limits[st]['x2'][0])*2, limits[st]['x2'][1]/10]
    scale_bar_x[1] += scale_bar_x[0]
    ax_twin.hlines(scale_bar_y, *scale_bar_x, colors='k')
    ax_twin.text(
        sum(scale_bar_x)/2,
        scale_bar_y * 1.1,
        'density\n%.3f' %(scale_bar_x[1] - scale_bar_x[0]),
        fontdict=dict(size=12, ha='center', va='bottom', style='italic')
    )
        
    ax_twin.set_xlim(limits[st]['x2'])
    ax_twin.set_xlabel(None)
    ax_twin.set_xticks([])

    ax.set_xlabel('Production (or release) year')
    ax.set_ylabel(f'# {st}')
    ax.set_title(f'Set in {st}')
    
plt.suptitle(
    'Time difference between set-in years and production years',
    fontsize=22
)

sns.despine(trim=True)
plt.tight_layout()

plt.savefig('figures/produced-vs-timediff.pdf')
plt.savefig('figures/produced-vs-timediff.png')
../_images/b31b0b279441065e18a1dec01b616dcaebb7e2c1d129f434e3a1902e855fd893.png

Considerable number of movies are set in the past, even more so than setting in the present. The tail towards the past is much heavier than it is for the future.

Contents