imdb-sqlite 1.0.0

Imports IMDB TSV files into a SQLite database

imdb-sqlite

Imports IMDB TSV files into a SQLite database.

It will fetch the files from IMDB unless you've already fetched them earlier.

The program relies on the following IMDB tab separated files:

• title.basics.tsv.gz: Video titles such as movies, documentaries, tv series, episodes etc.
• name.basics.tsv.gz: People in the entertainment business.
• title.akas.tsv.gz: Alternative names for titles, for different languages.
• title.principals.tsv.gz: Mapping of who participated in which title (movie / show).
• title.episode.tsv.gz: Season and episode numbers, for episodes of shows.
• title.ratings.tsv.gz: Current rating and vote count for the titles.

Installation

pip install imdb-sqlite

Usage

usage: imdb-sqlite [OPTIONS]

Imports imdb tsv interface files into a new sqlitedatabase. Fetches them from
imdb if not present onthe machine.

optional arguments:
  -h, --help       show this help message and exit
  --db FILE        Connection URI for the database to import into. (default:
                   imdb.db)
  --cache-dir DIR  Download cache dir where the tsv files from imdb will be
                   stored before the import. (default: downloads)
  --verbose        Show database interaction (default: False)

Just run the program with no arguments, and you'll get a file named imdb.db in the current working directory.

Hints

• Make sure the disk the database is written to has sufficient space. About 5 GiB is needed.
• Use a SSD to speed up the import.
• To check the best case import performance, use an in-memory database: --db :memory:.

Example

$ imdb-sqlite

2018-07-08 16:00:00,000 Populating database: imdb.db
2018-07-08 16:00:00,001 Applying schema

2018-07-08 16:00:00,005 Importing file: downloads\name.basics.tsv.gz
2018-07-08 16:00:00,005 Reading number of rows ...
2018-07-08 16:00:11,521 Inserting rows into table: people
100%|█████████████████████████| 8699964/8699964 [01:23<00:00, 104387.75 rows/s]

2018-07-08 16:01:34,868 Importing file: downloads\title.basics.tsv.gz
2018-07-08 16:01:34,868 Reading number of rows ...
2018-07-08 16:01:41,873 Inserting rows into table: titles
100%|██████████████████████████| 5110779/5110779 [00:58<00:00, 87686.98 rows/s]

2018-07-08 16:02:40,161 Importing file: downloads\title.akas.tsv.gz
2018-07-08 16:02:40,161 Reading number of rows ...
2018-07-08 16:02:44,743 Inserting rows into table: akas
100%|██████████████████████████| 3625334/3625334 [00:37<00:00, 97412.94 rows/s]

2018-07-08 16:03:21,964 Importing file: downloads\title.principals.tsv.gz
2018-07-08 16:03:21,964 Reading number of rows ...
2018-07-08 16:03:55,922 Inserting rows into table: crew
100%|███████████████████████| 28914893/28914893 [03:45<00:00, 128037.21 rows/s]

2018-07-08 16:07:41,757 Importing file: downloads\title.episode.tsv.gz
2018-07-08 16:07:41,757 Reading number of rows ...
2018-07-08 16:07:45,370 Inserting rows into table: episodes
100%|█████████████████████████| 3449903/3449903 [00:21<00:00, 158265.16 rows/s]

2018-07-08 16:08:07,172 Importing file: downloads\title.ratings.tsv.gz
2018-07-08 16:08:07,172 Reading number of rows ...
2018-07-08 16:08:08,029 Inserting rows into table: ratings
100%|███████████████████████████| 846901/846901 [00:05<00:00, 152421.27 rows/s]

2018-07-08 16:08:13,589 Creating table indices ...
2018-07-08 16:09:16,451 Import successful

Note

The import may take a long time, since there are millions of records to process.

The above example used python 3.6.4 on windows 7, with the working directory being on a SSD.

Data model

The IMDB dataset gravitates around the notion of a title. It is the primary entity. The title ID is what you see in the URL when you visit imdb.com. It is the defacto ID that other movie and TV sites use to uniquely reference a movie or show. So a bit of clarification on that ID and how the tables in the dataset reference it is in order.

A movie has a title, a TV show has one. An episode has one as well. Well two actually; the title of the show, and the title of the episode itself. That is why there are two links to the same title_id attribute in the titles table.

To make the relationships a bit clearer, following are a few query examples

Find movies named Casablanca and their ratings

SELECT t.title_id, t.type, t.primary_title, t.premiered, t.genres, r.rating, r.votes
FROM titles t  INNER JOIN ratings r ON ( r.title_id = t.title_id  )
WHERE t.primary_title = 'Casablanca' AND t.type = 'movie'

If the title type is omitted, we'd get results for tv series and other production types named Casablanca. The current set of title types in IMDB are the following:

{ movie, short, tvEpisode, tvMiniSeries, tvMovie, tvPilot,
  tvSeries, tvShort, tvSpecial, video, videoGame }

Find all episodes of the TV show Better off Ted and rank them by rating

-- // table aliases: st = show-title, et = episode-title
SELECT st.primary_title, st.premiered, st.genres, e.season_number,
       e.eposide_number, et.primary_title, r.rating, r.votes
FROM  titles AS st
INNER JOIN       episodes  e ON ( e.show_title_id = st.title_id )
INNER JOIN       titles   et ON ( e.episode_title_id = et.title_id )
LEFT  OUTER JOIN ratings   r ON ( et.title_id = r.title_id )
WHERE st.primary_title = 'Better Off Ted'
AND   st.type = 'tvSeries'
ORDER BY r.rating DESC

Find which productions both Robert Deniro and Al Pacino acted together on

SELECT t.title_id, t.type, t.primary_title, t.premiered, t.genres,
       c1.characters AS 'Pacino played', c2.characters AS 'Deniro played'
FROM people p1
INNER JOIN crew   c1 ON ( c1.person_id = p1.person_id )
INNER JOIN titles  t ON ( t.title_id = c1.title_id )
INNER JOIN crew   c2 ON ( c2.title_id = t.title_id )
INNER JOIN people p2 ON ( p2.person_id = c2.person_id )
WHERE p1.name = 'Al Pacino'
AND   p2.name = 'Robert De Niro'
AND   c1.category = 'actor' AND c1.category = c2.category

As indicated in the query, each person can participate in a production in different roles. The crew.category designates the participation role in the production. The current set of crew categories are:

{ actor, actress, archive_footage, archive_sound, cinematographer, composer,
  director, editor, producer, production_designer, self, writer }

Performance tips

Many relations over a ten gigabyte database is where a relational DB really starts showing its limits. The above query takes quite a while to run. Optimally a graph DB would be used instead, since the query is more of that nature. Some things can still be done to get an OK response time. If you use IDs instead of names, some joins can be eliminated. For example by using the IMDB IDs for Pacino and De Niro instead of searching for them by name, the above query is sped up by an order of magnitude or so, by eliminating the index search through two people tables.

SELECT t.title_id, t.type, t.primary_title, t.premiered, t.genres,
       c1.characters AS 'Pacino played', c2.characters AS 'Deniro played'
FROM crew c1
INNER JOIN crew c2 ON ( c1.title_id = c2.title_id  )
INNER JOIN titles t ON ( c2.title_id = t.title_id  )
WHERE c1.person_id = 'nm0000199' -- Pacino
AND   c2.person_id = 'nm0000134' -- De Niro
AND   c1.category = 'actor' AND c2.category = 'actor'

Another thing to try is to further normalize the data. There is quite some duplication in the data, such at title types, crew categories etc. That duplication prevents the DB from efficiently joining tables. Even though a lot of indices are used, any collection of duplicates in a column forces the engine to essentially do brute force scans through all those duplicates. It's not much else it can do, and this takes time. Normalizing away such duplicates will mean more joins are required, but it should hopefully allow the engine to more quickly whittle down the dataset. I've not tried this, but it seems something plausible to invetigate if your queries are too slow for your taste.

Finally, when in doubt. Prefix your query with the EXPLAIN QUERY PLAN. If you see SCAN TABLE in there, particularly in the beginning, it means the DB is doing a brute-force search through all the data in the column. This is very slow. You want the query plan to say SEARCH everywhere. Create an index for the column indicated as being scanned and rerun the query plan. Hopefully that resulted in orders of magnitude query speed improvement.

For example sqlite3 imdb.db "CREATE INDEX myindex ON <table-name> (<slow-column>)"

PyPI

Current status of the project is:

This project uses an automated build and release process. The module in the pypi repository is automatically built and released from the github source, upon any version tagged commit to the master branch.

Click the status link and check out the logs if you're interested in the package lineage; meaning how the released pypi module was constructed from source.