zdjohn/lda_ex

LDA clustering with pyspark mllib

see the detailed report in notebook here

How to run this PySpark job on EMR?

Submitting the job to EMR is easy

RUN: bash emr.sh

Under the hood following things happened:

1. Python code including settings will be packaged into local ./dist folder as EMR job artifacts (see more in package.sh)
2. Local ./dist then is uploaded to s3 under artifact prefix. (later referenced by EMR step)
3. Create new EMR cluster, it read raw data from source_json, run ETL job
4. Finally, it saves outputs (such as, transformed data, model, and visualization data files) under output prefix in s3.

see s3 folder structure: (logs is EMR log folder)

What happens through the ETL process?

1. Read raw json file from source folder source_json in s3
2. Data clean and Feature engineering on product description:
   • remove html markups
   • remove punctuations and special characters
   • remove numbers
   • remove common stop words
   • then tokenize and vectoring product description corpus
3. Fitting feature data via LDA model
4. Save transformed data into s3 output folder
5. Save model artifacts
6. Save reports for data visualization

How to start developing?

check pre-requisite

• aws accounts
• credentials with emr/s3/airflow policy attached
• pyspark dev environment
• python,tox

setting up dev environment

Create virtual dev environment via tox

RUN: tox -e dev

ACTIVATE: :{your project root}$ source .tox/dev/bin/activate

OPEN NOTEBOOK: jupyter notebook

Happy coding!

How to run test?

tox -e test