DBIS Pipeline

This pipline can be used to run analyses in a structured way, and stores configurations and results in a database.

Usage

the user writes a minimal plan file which contains only the following information:

• "how do I get the data?", by providing a dataloader
• "what to do with the data?", by providing a scikit pipeline
• "how to process the result?", by providing result handlers.
• "where to additionally store results?" by providing storage handlers.

Please have a look at the examples for more information.

CLI

We provide a dbispipeline-link tool that can be used to link datasets to the data directory. This ensures that datasets are linked in a consistent way even on different machines. The general process is as follows:

1. Either in the dbispipeline.ini or as an argument in the cli call, one can define where in general datasets are stored on the local machine. For example, many datasets are available on /storage/nas3/datasets/text. In this case, this would be the value in the configuration:

   #dbispipeline.ini:
   [project]
   dataset_dir = /storage/nas3/datasets
   

2. In a file data/links.yaml, one can define specific datasets that are used by the software. Thereby, the first path segment will be cut off (not sure why). For example, the following yaml file:

   ---
   datasets:
     - music/acousticbrainz
     - music/billboard
     - music/millionsongdataset
   

   would assume that a physical directory exists at /storage/nas3/datasets/music/billboard and after calling the script dbispipeline-link without parameters using the above configuration, the following symlinks will be created:

   data/acousticbrainz -> /storage/nas3/datasets/music/acousticbrainz
   data/billboard -> /storage/nas3/datasets/music/billboard
   data/millionsongdataset -> /storage/nas3/datasets/music/millionsongdataset
   

   The value of dataset_dir from the config can be overwritten in the cli script by using the -p option.

Requirements

• python >= 3.6
• a PostgreSQL database
• an email server if you want to use notification emails

Installation

1. Install dbispipeline in your python. We recommend using pipenv to keep your dependencies clean: pipenv install dbispipeline This call will install a virtual environment as well as all dependencies.
2. Write your plan(s). See the example plan files for guidance.
3. call pipenv run dp <yourplanfile.py>

Enjoy!

Configuration

The framework look in multiple directories for its configuration files.

• /usr/local/etc/dbispipeline.ini used for system wide default.
• $HOME/.config/dbispipeline.ini used for user specific configurations.
• ./dbispipeline.ini for project specific configurations.

And example configuration file looks like this:

[database]

# url to your postgres database
host = your.personal.database

# your database user name
user = user

# port of your postgres database, default = 5432
# port = 5432

# password of your database user
password = <secure-password>

# database to use
database = pipelineresults

# table to be used
result_table = my_super_awesome_results

[project]
# this will be stored in the database
name = dbispipeline-test

# this is used to store backups of the execution
# it is possible to override this by setting the DBISPIPELINE_BACKUP_DIR
# environment variable
# the default is the temp dir of the os if this option is not set.
backup_dir = tmp

# this is used to linke the used datasets spcified in data/links.yaml
# it is possible to override this by setting the DBISPIPELINE_DATASET_DIR
# environment variable
dataset_dir = /storage/nas/datasets

[mail]
# email address to use as sender
sender = botname@yourserver.com

# recipient. This should probably be set on a home-directory-basis.
recipient = you@yourserver.com

# smtp server address to use
smtp_server = smtp.yourserver.com

# use smtp authentication, default = no
# authenticate = no

# username for smtp authentication, required if authenticate = yes
# username = foo

# password for smtp authentication, required if authenticate = yes
# password = bar

# port to use for smtp server connection, default = 465
# port = 465