Load Dow Jones DNA Snapshot to Elasticsearch

Sample code to read data from snapshot files, enrich articles metadata, and import the output into Elasticsearch. It uses some methods created in a common library (djdna_common) included as a symbolic link.

djdna_common

Set of common methods that eases operations like reading DNA Snapshots AVRO files, calculating new features or interacting with Elasticsearch. These methods are for illustration purposes and don't have a robust coding to validate unexpected cases or handling exceptions. For this reason it is not distributed as a Python package. It is however used among multiple Dow Jones DNA examples.

To use these methods, clone this and the djdna_common repository to the same base directory, and (if necessary) create a symbolic link or copy the folder content. A sample sequence looks like this:

$ git clone https://github.com/miballe/djdna-snapshot2elasticsearch.git
$ git clone https://github.com/miballe/djdna_common
$ cd djdna-snapshot2elasticsearch
$ ln -s ../djdna_common/ djdna_common

dna-es-mappings.json

Mappings file to create a new index in Elasticsearch. This file contains a slightly modified field structure with some additional fields for embeddings (title and body). The following excerpt shows some of the highlights of this file.

"body": {
    "type": "text"
},
"body_dv": {
    "type": "dense_vector",
    "dims": 512
    },

A new index matching the DNA schema can be created by using this file and the following code. This code uses the Python Elasticsearch library.

import json
from elasticsearch import Elasticsearch
from elasticsearch.client import IndicesClient

es_client = Elasticsearch(['http://localhost:9200'])
ix_name = 'dnaarticles'

with open('./dna-es-mappings.json') as ix_file:
    ix_map = json.load(ix_file)
es_client.indices.create(index=ix_name, body=ix_map)

load-dna-data.py

This file has the main logic to read a Snapshot AVRO files and load their content to a pandas DataFrame. Then, it enriches the dataset with some embedding to the title and body fields, and finally, the final outcome is loaded to Elasticsearch.