@article{dynomant2017medoc,
title={MEDOC: a Python wrapper to load MEDLINE into a local MySQL database},
author={Dynomant, Emeric and Gorieu, Mathilde and Perrin, Helene and Denorme, Marion and Pichon, Fabien and Desfeux, Arnaud},
journal={arXiv preprint arXiv:1710.06590},
year={2017}
}
Thanks to rafspiny for his multiple corrections and feedback !
MEDOC had multiple changes. The most important is about the XML parsing, which should return less errors than before. The way the data is parsed has been improved.
Then, the stacking of the SQL INSERT() has been removed. Files are now process in parallel by many threads and inserted during this streaming of the file.
I now assume that it is easy and cheap to get a decent multi-thread machine for 24h of processing (AWS, GoogleCloud, MAzure ...) with a decent amount of RAM.
MEDOC has been tested on a 144 cores XEON E7 with 500Go of RAM. If your want the old version, please clone the sequential branch.
MEDLINE is a database of scientitifc articles released by the NIH. Pubmed is the most common way to query this database, used daily by many scientists around the world.
The NIH provides free APIs to build automatic queries, however a relational database could be more efficient.
The aim of this project is to download XML files provided by MEDLINE on a FTP and to build a relational mySQL database with their content.
The first step is to clone this Github repository on your local machine.
Open a terminal:
git clone "https://github.com/MrMimic/MEDOC"
cd ./MEDOC
Here prerequisites and installation procedures will be discussed.
XML parsing libraries may be needed. You can install them on any Debian-derived system with:
sudo apt-get install libxml2-dev libxslt1-dev zlib1g-dev
You may also need python-dev
. You can also install it with the same command:
sudo apt-get install python-dev
The second step is to install external dependencies. First, create a virtual environment and load it:
python3 -m venv .venv
source .venv/bin/activate
Then, simply run the following commands:
python3 -m pip install --upgrade pip
python3 -m pip install -r requirements.txt
Avery libraries will now be installed into this environment.
Before you can run the code, you should first create a configuration.cfg file (in the MEDOC folder) and customize it according to your environment. Below is the dist.config:
# ================================ GLOBAL =============================================
[informations]
version: 1.3.0
author: [email protected]
# =========================== MYSQL ============================================
[database]
path_to_sql: ./utils/database_creation.sql
user: <YOUR_SQL_USER>
password: <YOUR_SQL_PASSWORD>
host: <YOUR_SQL_HOST>
port: <YOUR_SQL_PORT>
database: pubmed
# =========================== PATH ============================================
[paths]
program_path: ./
pubmed_data_download: ./pudmed_data/
sql_error_log: ./log/errors.log
already_downloaded_files: ./log/inserted.log
# =========================== PARALLELISM ============================================
[threads]
parallel_files: 10
Then, simply execute :
python3 medoc.py
It took 26H to insert 29,058,362 articles with a XEON E7. Among them, 420 insert command generated an error, aminly due to under-sized mySQL columns.
Recreate the SQL database after dropping it, by running the following command:
DROP DATABASE pubmed ;
Then, comment every line about indexes (CREATE INDEX) or foreigns keys (ALTER TABLE) into the SQL creation file. Indexes are slowing up insertions.
When the database is full, launch the indexes and alter commands once at a time.
Make sure you have all the right dependencies installed
On Debian based machines try running:
sudo apt-get install python-dev libxml2-dev libxslt1-dev zlib1g-dev