Data engineering course

1. System requirements
2. Prerequisities
3. Projects
   • Data Cubes
     • Installation instructions
     • Files
   • Apache Airflow
     • Installation instructions
     • Files
   • Provenance
   • SKOS & DCAT-AP

System requirements

• Linux / Windows 10 or higher with WSL installed
• Nodejs v19.8.1 or higher
• NPM v9.6.2 or higher
• Python v3.8
• PyPI v20.0.2 or higher

Prerequisities

• Nodejs
• NPM
• TypeScript execution engine (ts-node)
• Python
• PyPI

Projects

Data Cubes

We're generating the following data cubes:

• Care providers (Poskytovatelé zdravotních služeb)
• Population 2021 (Obyvatelé okresy 2021)

Installation instructions

Configuration

To install required project dependencies, run the following commands:

cd data-cubes
npm ci

Run

Scripts for data cubes generation can be launched from data-cubes directory using a command:

npm start

To generate both data cubes and test their integrity constraints, run the following:

npm test

Files

Input

Input files are stored in the data-cubes/input directory. There're 3 source CSV files:

• care-providers-registry.csv (data sources for care providers data cube)
• population-cs-2021.csv (data sources for mean population data cube)
• county-codes.csv (mapping for translation of county codes from population-cs-2021.csv into standardized NUTS codes)

The remaining files contain RDF schemas for the generated data cubes.

Output

Output files are stored in the data-cubes/output directory. The following files will be generated into the output directory:

• care-providers.ttl (Care providers data cube with metadata)
• population.ttl (Mean population data cube with metadata)
• datasets.ttl (merged data cubes from care-providers.ttl and population.ttl)

Scripts

The following scripts need to be run to generate both data cubes and to validate their integrity constraints:

• care-providers.ts
• population.ts
• constraints-validation.ts

Apache Airflow

We're generating the same data cubes as in the previous section:

• Care providers (Poskytovatelé zdravotních služeb)
• Population 2021 (Obyvatelé okresy 2021)

Installation instructions

This project needs to be run on a Linux machine or in the WSL environment.

Configuration

First, open project's directory:

cd airflow

Then proceed with installing dependencies for Nodejs scripts from package.json:

npm ci

Next step will be configuration of a virtual environment for Python. You can install Apache Airflow from the exported dependencies in requirements.txt file or follow the official installation guide.

To set up Python virtual environment called venv with all required dependencies, run the following commands:

# Create Python venv
python3 -m venv venv

# Activate venv
. venv/bin/activate

# Install Python dependencies from requirements.txt
pip install -r requirements.txt

You'll also need to define a home directory for Apache Airflow:

export AIRFLOW_HOME=~/airflow

To finish Airflow setup, run the following:

# Initialize a database
airflow db init

# Create an administrator
airflow users create --username "admin" --firstname "Harry" --lastname "Potter" --role "Admin" --email "[email protected]"

# Check the existing users
airflow users list

You should also update the generated airflow.cfg file:

1. dags_folder needs to point to the dags directory, e.g. /home/kristyna/data-engineering/airflow/dags
2. load_examples should better be set to False if you want to avoid seeing tons of example DAGs

Run

You'll need 2 commands to launch Airflow. Each command needs to be triggered in its own terminal.

1. airflow scheduler

2. airflow webserver --port 8080

Now you can visit http://localhost:8080/home in your web browser. If you set load_examples = False in airflow.cfg, you should only see one DAG - data-cubes.

DAGs can be easily triggered using the web interface. You need to provide a JSON configuration with output_path field. An example of the right configuration can be seen at dag-configuration.json. The generated data cubes will be stored into the directory specified by output_path parameter. If you forget to pass this parameter, {dags_folder}/output will be used as a default location for output files.

Files

Input

Input files are gathered in airflow/dags/input directory. Three CSV files will be downloaded (same as in Data Cubes project):

• care-providers-registry.csv (data sources for care providers data cube)
• population-cs-2021.csv (data sources for mean population data cube)
• county-codes.csv (mapping for translation of county codes from population-cs-2021.csv into standardized NUTS codes)

The remaining files contain RDF schemas for the generated data cubes.

Output

Output files will be stored into the output_path directory or {dags_folder}/output by default. The following files will be generated:

• health_care.ttl (Care providers data cube with metadata)
• population.ttl (Mean population data cube with metadata)
• datasets.ttl (merged data cubes from health_care.ttl and population.ttl)

Airflow pipeline

There's exactly one DAG defined in data-cubes.py. It specifies 6 tasks for the Airflow pipeline:

1. health_care_providers_download
2. population_2021_download
3. county_codes_download
4. care_providers_data_cube
5. population_data_cube
6. integrity_constraints_validation

Tasks 1-3 take care of downloading the source CSV files into the input directory. Tasks 4 and 5 are responsible of generating the target data cubes health_care.ttl and population.ttl. The last task merges the output datasets and validates them against a set of integrity constraints.

For a better idea of how the tasks are connected, see the DAG's graph visualization:

Scripts

Scripts operating with CSV and RDF files were adapted from the Data Cubes project. They're stored in airflow/dags/scripts and they're triggered by data-cubes.py. The most important (entry) scripts are again the following:

• care-providers.ts
• population.ts
• constraints-validation.ts

Provenance

A provenance document describing a process of generating datasets from Data Cubes project can be found in provenance directory. It is stored as provenance.trig file in RDF TriG format. It follows PROV-O specification and its attached examples.

SKOS & DCAT-AP

The datasets from Data Cubes project are extended with SKOS hierarchy and DCAT-AP metadata in the skos-and-dcat project.

For project installation & usage instructions, refer to the original Data Cubes section. Project structure is the same as in data-cubes project, except for 2 changes:

1. Metadata

Compared to the original Data Cubes project, metadata were removed from population dataset completely and they were moved from care-providers dataset into a separate dataset care-providers-metadata. Metadata are described in a file skos-and-dcat/input/care-providers-metadata.ttl. File's content is loaded into an RDF store, normalized and dumped into skos-and-dcat/output/care-providers-metadata.ttl.

2. SKOS hierarchy

A SKOS hierarchy was employed for regions and counties in both data cubes - care-providers and population. Regions and counties are defined as separate SKOS concepts (skos:Concept) and they're connected to the hierarchy through Broader and Narrower relationships (skos:broader, skos:narrower).