Processing NYC Taxi Data using PySpark ETL pipeline


This is an project to extract, transform, and load large amount of data from NYC Taxi Rides database (Hosted on AWS S3). It extracts data from CSV files of large size (~2GB per month) and applies transformations such as datatype conversions, drop unuseful rows/columns, etc. Finally, the data is written back in parquet format. This saves time for tasks such as machine learning. It also saves a huge amount of space (~97% space reduction from csv to parquet) making it easy to store for downstream tasks.

How to use it (Using GCP as the cloud service of choice)

  • Setup a bucket on Google Cloud Storage
  • Use to download raw data from s3 in the form of CSV files to the GCS bucket
  • Setup a GCP dataproc service
  • SSH into the master node and copy the entire project folder to the Persistent Disk
  • Edit the configuration file for application
  • Submit the job: submit-spark --filename [raw_data_filename] or Execute with appropriate args

Project structure

|   requirements.txt


View Github