search

buccaneerai / bottlenose

:x: Deprecated. High-level RxJS operators for data wrangling
Other
6 stars 0 forks source link
experiment opensource package

readme

CircleCI CircleCI License: MIT

⚠️ Although most of the operators are functional, this monorepo is no longer maintained. (The maintainers found Lerna to be too unwieldy to maintain and it generated more overhead than it was worth.) Instead, the RxJS utilities have been moved to separate npm packages starting with @buccaneerai/rxjs-<package-name>.

Description

🐬 RxJS offers a great toolkit for reactive functional programming. But if you use RxJS in a larger software application, you may often find yourself handcrafting your own functions to handle common workflows and wrap them in RxJS observables (like streaming file data or connecting to a WebSocket). Bottlenose is a collection of npm packages that provide industrial-strength RxJS operators and utilities for handling common data streams (filesystems, CSV, AWS S3, Websockets, socket.io), analyzing data (statistics, NLP) and implementing machine learning models (like SVM).

Features:

You can use Bottlenose to...

Bottlenose is a young project started in late 2019 so it has a limited but fast-growing feature set. It is maintained by an opensource community. It is also actively used and improved by a healthcare AI startup called Buccaneer, which is adding new modules frequently as their team develops them for enterprise SaaS data pipelines.

Documentation & Guides

Documentation

Installation

Each Bottlenose module is namespaced under @bottlenose and installed separately. See the docs for a list of all currently available modules. For example, this would install Bottlenose's statistics module:

npm i --save @bottlenose/rxstats

Or

yarn add @bottlenose/rxstats

Basic Usage

For a full list of operators and modules, see the documentation.

Easily generate input streams

Bottlenose has modules for handling common data input sources like CSV, AWS S3, local file system and websockets. For example, this would parse CSV input:

import { from } from 'rxjs';
import { map  } from 'rxjs/operators';
import { parse } from '@bottlenose/rxcsv';

// Create a stream of raw CSV data
const csvString$ = from([
  'name,systolicBp,dialostilicBp,message\n', 
  'Blackbeard,140,91,Yarr\nCrunch,120,', 
  ',180,Arr\nSparrow,110,70,Savvy\n',
]);

// Stream the CSV data into an RxJS Subject
const row$ = csvString$.pipe(parse());
row$.subscribe();
// {name: "Blackbeard", systolicBp: 140, diastolicBp: 91, isAngry: true},
// {name: "Crunch", systolicBp: 120, diastolicBp: 80, isAngry: false},
// {name: "Sparrow", systolicBp: 110, diastolicBp: 70, isAngry: false},

Analyze data reactively

Bottlenose also has modules for common data analysis tasks (like descriptive statistics and NLP). For example, this would calculate the mean on the csv stream from the previous example:

import { map } from 'rxjs/operators';
import { mean } from '@bottlenose/rxstats';

// using the row$ csv from the prior example
const mean$ = row$.pipe(
  map(row => row.systolicBp), 
  mean() // calculate the mean on one of the columns
);
mean$.subscribe(console.log);
// 140
// 130
// 123.33333333333333

Train machine learning models reactively

Bottlenose is also adding modules for common machine learning algorithms (like SGD). For example, this would train an SGD classifier:

import {classifier} from '@bottlenose/rxsgd';

// train an SGD model on the example above
const sgd$ = row$.pipe(
  map(row => [[row.systolicBp, row.diastolicBp], row.isAngry]),
  classifier()
);
// train the classifier and emit it its new trained parameters each time 
// a new item is ingested
sgd$.subscribe();

Community

If you share the goal of creating amazing data science tools for the Javascript community, then here are some ways to help: