Bead

Changing your data and code together

Overview

Questions

  • How do you ensure that your data products are reproducible?

Objectives

  • 'Use bead, a command-line tool to create, update and share data products.'

The problem

Researchers and analysts need to know the provenance of their data to trust its integrity and to quickly redo the analysis when requirements change. However, in a diverse and agile team, knowledge about data is often tacit, and is destroyed when a team member leaves. This leads to a duplication of effort. Additionally, the multitude of software tools and work methods create frictions in the analytics process. Existing solutions are either too rigid to be widely adopted or involve too much face-to-face communication, reducing researcher and analyst productivity.

bead is a lightweight software tool with which a researcher can explicitly declare dependency on other data products, referring to them with a universally unique identifier. She can see how her work fits in the bigger picture and who to work with when redoing the analysis. Bead encapsulates data, the software script that created it and references to its dependencies in a single package, making knowledge about the data explicit. Bead is platform independent and agnostic to the specific software tools and workflows used to create data. It can be adapted to any workflow that uses the file system for both scripts and data. Sample use cases include social and business datasets of a few GB each, processed and analyzed with perl, Python, R, Stata, Matlab, julia or other scripts.

Basic logic of bead

Given a discrete computation of the form

output = function(*inputs)

a BEAD captures all three named parts:

  • output - data files (results of the computation)

  • function - source code files, that when run hopefully compute output from inputs

  • inputs - are other bead' output and thus stored as references to those beads

As a special case pure data can be thought of as constant computation having only output but neither inputs nor source code.

Bead concepts

Data packages can be in one of two states. Under active development in a workspace, or packaged and stored as a bead. Beads are stored in a box, which is just a collection of completed beads.

Workspace

A workspace is a directory, where the user works. It contains a prototype of a bead - it is a computation under active development. At some time however it is supposed to have all the inputs, code and output of a computation.

The directory has special structure and content, so it must be created via the bead tool:

  • there are metadata in a hidden .bead-meta directory which .txt file enumerate all the inputs and the bead unique hash id.

  • there are 3 standard directories with special meanings:

    • input: where input data is read from. It is read only, managed by the bead tool.

    • temp: temporary outputs, this is an area that is ignored when saving the bead.

    • output: This is where results of the computation are stored.

Bead

A bead is a frozen, discrete computation, created from a workspace. It is currently a zip file.

A bead is intended to capture data with the code that produced it. The goal is transparency and semi-automatic reproducability through extra tooling. Full automatic reproducability is assumed to be inpractical/expensive, however it can be achieved by gradual process/quality improvements (learning through feedback).

The bead format is designed to be

  • resilient to change

  • decentralized

  • keep enough information to be able to get both the details and the big picture (if all relevant beads are available)

The main technology involved is a combination of different probabilistic identifiers (UUID, secure hash, Merkle-tree).

Main properties of a bead:

  • kind that is shared with other versions of a bead (book analogy: ISSN)

    • it is a technical name, whose existence allows the human name to

      change/evolve/diverge over time, while still referring to conceptually

      the same computation

  • content_id, that is unique for every bead (~version, book analogy: ISBN)

    • it is calculated, so changes in a bead makes it either invalid or a new version

  • freeze time (for ordering versions, this is fragile in theory as depends on correctly set clocks, but in practice it is expected to cause few problems)

  • freeze name

  • references to its inputs (kind, content_id)

The main changes from v. 0.0.2. to 0.8.1 that beads are referenced by names from here on.

It is important to mention that we should not create a new bead with a name already in use.

Box

A box is where beads are saved to and loaded from. It also gives names to beads and provide minimal search functionality. Currently, boxes are implemented a flat directories on the file system.

Basic workflow

Bead help

The bead help guide you through the usage of the bead.

$ bead -h

usage: bead [-h] {new,develop,save,status,nuke,web,zap,xmeta,version,input,box}

positional arguments:
  {new,develop,save,status,nuke,web,zap,xmeta,version,input,box}
    new                 Create and initialize new workspace directory with a new bead.
    develop             Create workspace from specified bead.
    save                Save workspace in a box.
    status              Show workspace information.
    nuke                No operation, you probably want zap, to delete the workspace.
    web                 Manage/visualize the big picture - connections between beads.
    zap                 Delete workspace.
    xmeta               eXport eXtended meta attributes to a file next to zip archive.
    version             Show program version.
    input               Manage data loaded from other beads...
    box                 Manage bead boxes...

optional arguments:
  -h, --help            show this help message and exit

All the positional arguments have own subcommands with complete help.

$ bead new -h

usage: bead new [-h] DIRECTORY

Create and initialize new workspace directory with a new bead.

positional arguments:
  DIRECTORY   bead and directory to create

optional arguments:
  -h, --help  show this help message and exit

Create a new bead

Initial setup. The latest bead-box already made on the haflinger.

$ mkdir /somepath/bead-box/latest
$ bead box add latest /somepath/bead-box/latest
Will remember box latest

This is where completed beads will be stored. Create an empty bead with name name:

/somepath$ bead new name
Created name

Add some data to the output of this new bead which we can use later. This bead has no computation, only data.

/somepath$ cd name/
/somepath/name$ echo World > output/who-do-i-greet
/somepath/name$ bead save latest
Successfully stored bead.
cd ..
/somepath/$ bead zap name
Deleted workspace /somepath/name

Working with inputs in a new bead

Create a new data package:

/somepath$ bead new hello
Created hello
/somepath$ cd hello/

Add data from an existing bead at input/<input-name>/:

/somepath/hello$ bead input add name
Loading new data to name ... Done

Create a program greet that produces a greeting, using input/name as an input:

read name < input/name/who-do-i-greet
echo "Hello $name!" > output/greeting

Run the program:

/somepath/hello$ bash greet

This script has create a text file in output/greeting. Let us verify its content:

/somepath/hello$ cat output/greeting
Hello World!

Load specific version of an input in bead

I want to roll back analysis. I have cherry picked by bead meta, but how do I load the exact bead versions into the input folder?

You need to bead input delete it first, and then bead input add the version you would like to use. bead input load and bead input unload do not touch the bead meta, they only load the data specified in the meta to the input folder or delete them from the input folder (to free up disk).

Bead input have a time option. For example: you have two versions with different time stamps. address_balance_20180823T120149294589+0200.zip address_balance_20200317T111341998516+0100.zip

If you want to use version 20180823 you can type:

bead input add address_balance -t 20180823

You have to use the bead name and the yyyymmdd part of the timestamp with the -t option.

(source: https://stackoverflow.com/c/ceu-microdata/questions/40)

Visually display the bead chain

Bead web is a new feature of version 0.8.1. You can check the details with bead web -h

$ bead web color auto-rewire heads / source-bead target-bead / png filename.png

Auto-rewire is required for the new bead.

Color is optional.

Heads are optional: if loaded they will only plot the latest version of each bead plus what is referenced by another bead.

If you change the source bead to .. it plots the entire bead structure leading to the target bead. If you change the target bead to .. it plots the entire structure starting from the source bead. It is very important that before and after / you need a space character.

Instead of png it can be svg filename.svg if you prefer that format.

Package the data and send it to an outside collaborator

Save our new bead:

/somepath/hello$ bead save latest
Successfully stored bead.

This stores output, computation and references to inputs. Now the content of /somepath/BeadBox is

/somepath$ ls -1 BeadBox/
hello_20160527T130218513418+0200.zip
name_20160527T113419427017+0200.zip

These are regular (and, in this case, small) zip files, which can be transferred by usual means (e.g. emailed) to collaborators. The recipient can process them via the bead tool, keep the integrity of provenance information, and adding further dependencies as needed. Even withouth the tool, she can access the data by directly unzipping the file and inspecting its content.

The output of the computation is stored under data/*. An outide collaborator without access to bead can just ignore the computation and all other metadata.

/somepath$ unzip -p BeadBox/hello_20160527T130218513418+0200.zip data/greeting
Hello World!

/somepath$ unzip -v BeadBox/hello_20160527T130218513418+0200.zip
Archive:  BeadBox/hello_20160527T130218513418+0200.zip

    This file is a BEAD zip archive.

    It is a normal zip file that stores a discrete computation of the form

    output = code(*inputs)

    The archive contains

    - inputs as part of metadata file: references (content_id) to other BEADs
    - code as files
    - output as files
    - extra metadata to support
    - linking different versions of the same computation
    - determining the newest version
    - reproducing multi-BEAD computation sequences built by a distributed team

    There {is,will be,was} more info about BEADs at
    - https://unknot.io
    - https://github.com/ceumicrodata/bead
    - https://github.com/e3krisztian/bead

    ----

    Length    Method    Size  Cmpr    Date    Time   CRC-32   Name
    --------  ------  ------- ---- ---------- ----- --------  ----
    13        Defl:N       15 -15% 2016-05-27 13:01 7d14dddd  data/greeting
    66        Defl:N       58  12% 2016-05-27 13:01 753b9d15  code/greet
    742       Defl:N      378  49% 2016-05-27 13:02 a4eb5de9  meta/bead
    456       Defl:N      281  38% 2016-05-27 13:02 9a206f53  meta/manifest
    --------          -------  ---                            -------
    1277                  732  43%                            4 files

The following graph summarizes the internal structure of a workspace and the logical links to other beads.

Bead boxes

Key Points

  • Keep and share data together with the code that produced it.

  • When sharing your data, always do it in a bead.

  • Never refer to external data from a bead, use bead inputs.

PreviousServerNextWorking with data

Last updated