Skip to content

eyra/feldspar

 
 

Repository files navigation

Feldspar

Integration mechanism for developers to build an extension (flow application) that can be hosted on the Next platform. This is for example used in the Port program for data donation, as described below.

Digital Trace Data Donation (Port)

More information about the Port program can be found here.

Data donation allows researchers to invite participants to share their data download packages (DDPs). However, DDPs potentially contain very sensitive data and often not all data is needed to answer a specific research question.

Feldspar enables researchers to:

  • extract only the data of interest through local processing (on the participants device) using Python (Pyodide)
  • prompt participants for questions about the data
  • enable participants to inspect the extracted data before donation
  • enable participants to delete table rows before donation
  • consent or decline to donate the extracted data

Feldspar is open-source under the AGPL license and allows researchers to configure the frontend that guides participants through the data donation steps.

Note: Feldspar is only a frontend. In order for it to be used in a live study, it needs to be hosted on a server and connected to a storage to retrieve the donated data. To run a local instance see installation. To create a release for the Next platform or the self hosted version, see release.

Installation

In order to start a local instance of Feldspar follow these steps:

  1. Prerequisites

  2. Install dependencies & tools:

    cd ./feldspar
    npm install
    npm run prepare
  3. Start the local web server (with hot reloading enabled):

    npm run start
  4. You can now go to the browser: http://localhost:3000.

If the installation went correctly you should be greeted with a mock data donation study.

Release

  1. Create release file:
 npm run release
  1. Use release file:

The generated release.zip file can be installed on the Next platform or the self-hosted version, by adding a "Donate task" and at "Flow application" select the generated zip-file.

How to use Feldspar?

You can implement your own data donation flow by altering the Python script, which can be used to:

  1. customize the participant data donation flow in terms of screen content, type of screen (e.g. a file prompt) and screen order. You can use the Port API (props.py) for this.
  2. extract specific data from the participant DDP that is required for the research question. You can use the data extraction methods that are available in Pyodide

A typical script includes the following steps:

  1. Prompt the participant to select the DDP file
  2. Extract the data of interest from the selected DDP file. Try to aggregate and anonymize as much as possible.
  3. Present the extracted data on screen in clear tables to allow the participant to investigate the data that they are about to donate and buttons to choose to either donate or not (consent screen). If a data storage is connected, the extracted data is stored only when participants agree to donate.

Example script: script.py.

We recommend to use the example script as starting point for your own data donation study.

Port Assets

Assets needed in the script can be copied to: src/framework/processing/py/port/assets/

In your script you can access these assets as follows:

Asset path

from port.api.assets import *

def process(sessionId):
     path = asset_path("hello_world.txt")
     file = open(path, "r")
     txt = file.read()

Open asset

from port.api.assets import *

def process(sessionId):
     file = open_asset("hello_world.txt")
     txt = file.read()

Read asset

from port.api.assets import *

def process(sessionId):
     txt = read_asset("hello_world.txt")

Port API examples

Below some examples on how to use the Port API in your script.py

Main function Every `script.py` should have this function:
def process(sessionId):

This function is a generator of commands by using yield statements. No return statements should be used.

def process(sessionId):
    result1 = yield CommandUIRender(page1)
    result2 = yield CommandUIRender(page2)
    # last yield should not expect a result
    yield CommandUIRender(page3)

ScriptWrapper and py_worker using send to iterate over the commands one by one. For more information on yield and Generators visit https://realpython.com/introduction-to-python-generators.

API imports
from port.api.props as props
from port.api.commands import (CommandUIRender, CommandUIDonate)
Create file input
platform = "Twitter"
progress = 25

file_input_description = props.Translatable({
    "en": f"Please follow the download instructions and choose the file that you stored on your device.",
    "nl": f"Volg de download instructies en kies het bestand dat u opgeslagen heeft op uw apparaat."
})
allowed_extensions = "application/zip, text/plain"
file_input = props.PropsUIPromptFileInput(file_input_description, allowed_extensions)
Create consent tabels
import pandas as pd

table1_title = props.Translatable({
    "en": "Title 1",
    "de": "Titel 1",
    "nl": "Titel 1"
})
table1_data = pd.DataFrame(data, columns=["columnX", "columnY", "columnZ"])
table1 = props.PropsUIPromptConsentFormTable("table_1", table1_title, table1_data)

table2_title = props.Translatable({
    "en": "Title 2",
    "de": "Titel 2",
    "nl": "Titel 2"
})
table2_data = pd.DataFrame(data, columns=["columnA", "columnB", "columnC", "columnD"])
table2 = props.PropsUIPromptConsentFormTable("table_2", table2_title, table2_data)

tables = [table1, table1]
# Meta tables currently not supported
meta_tables = []

consent_form = props.PropsUIPromptConsentForm(tables, meta_tables)
Create donation screens
header = props.PropsUIHeader(title)
footer = props.PropsUIFooter(progress)
body = props.PropsUIPromptFileInput(file_input_description, allowed_extensions)
page = props.PropsUIPageDonation(platform, header, body, footer)
Create user input screen with radio buttons
header = props.PropsUIHeader(title)
footer = props.PropsUIFooter(progress)
body = props.PropsUIPromptRadioInput(title, description, [{"id": 0, "value": "Selection 1"}, {"id": 1, "value": "Selection 2"}])
page = props.PropsUIPageDonation(platform, header, body, footer)
Extract data from input file
page = props.PropsUIPageDonation(platform, header, file_input, footer)
result = yield CommandUIRender(page)

# Result is a dictionary (Payload)
if result.__type__ == "PayloadString":
    # File selected
    filename = result.value
    zipfile = zipfile.ZipFile(filename)

    # Extract the data of interest from the selected file
    # Write your own functions for data extraction
    ...
else:
    # No file selected
Handle user consent input
platform = "Twitter"
donation_key = f"{sessionId}-{platform}"
page = props.PropsUIPageDonation(platform, header, consent_form, footer)
result = yield CommandUIRender(page)

# Response is a dictionary (Payload)
if result.__type__ == "PayloadJSON":
    # User gave consent
    yield CommandSystemDonate(donation_key, result.value)
else:
    # User declined
Track user behaviour
tracking_key = f"{sessionId}-tracking"
data = "any json string"

# Use the donate command to store tracking data
yield CommandSystemDonate(tracking_key, data)

Use Feldspar in a data donation study

Feldspar serves as the frontend, providing the application with which participants engage. It facilitates the flow and logic for data donation. To utilize Feldspar in a data donation study, it must be hosted on a server capable of storing the donated data.

You can host Feldspar on the Next platform or the self-hosted version as explained here.

Alternatively, you can host Feldspar by embedding it in an iframe. After the iframe loads, send a message that includes a channel. The Feldspar application will use this channel to relay messages with data ready for storage. Here's a JavaScript example:

// ... wait until the iframe is loaded
const channel = new MessageChannel();
channel.port1.onmessage = (e) => {
  console.log("Message receive from Feldspar app", e);
};
// get the iframe via querySelector or another method
iframe.contentWindow.postMessage("init", "*", [this.channel.port2]);

Data donation as a service

Would you like to get support with setting up your data donation study or host your data donation study on the Next platform? Reach out to Eyra for custom pricing: connect@eyra.co.

Technical specifications of Feldspar

If your study requires specific adjustments (new interactive elements etc.), you have the flexibility to modify the Feldspar functionalities. Leverage the following technical insights to suit your needs.

Data model

Feldspar uses the following data model (also see: src/framework/types)

  • Modules

    Module Description
    ProcessingEngine Responsible for processing donation flows
    VisualizationEngine Responsible for presenting the UI and accepting user input
    CommandHandler Decoupling of ProcessingEngine and VisualizationEngine
    Bridge Callback interface for Bridge Commands (e.g. Donation)
  • Pages

    Page Description
    SplashScreen First page that is rendered before the Python script is loaded with GDPR consent logic
    Donation Page that uses several prompts to get a file from the user and consent to donate the extracted data
    End Final page with instructions on how to continue
  • Prompts

    Prompt Description
    FileInput File selection
    RadioInput Multiple choice question
    ConsentForm Displays extracted data in tables and asks for user consent
    Confirm General dialog to ask for extra confirmation
  • Commands

    Command Description
    Render Render the page
    Donate Save the extracted data

    Commands can be send from the Python script using the yield keyword.

  • Payloads

    Payload Description
    Void Command without user input as a response
    True Positive user input (e.g. Ok button in confirm prompt)
    False Negative user input (e.g. Cancel button in confirm prompt)
    Error Unexpected problem when handling command
    String String result
    File Only used in Javascript. This is intercepted in py_worker.js and translated into a String (filename), while the bytes of the file are written to the Pyodide file system
    JSON User input structured as JSON, used to return the consent data from the consent form

    Payloads are part of a Response back to the Python script after sending commands:

    export interface Response {
        __type__: 'Response'
        command: Command
        payload: Payload
    }

    Responses are intercepted in py_worker.js and only the payload is returned to the Python script. Payloads don't have a Python representation in the API yet. They are translated into a dictionary (default Pyodide behaviour).

Python-Javascript interoperability

See: src/framework/processing/py/port

  • ScriptWrapper

    This object is used in main to wrap the process generator function in your script. It translates incoming Javascript and outgoing Python commands.

  • API

    • commands.py: Defines commands, pages and prompts that are used to communicate from the Python script to the VisualisationEngine and Bridge.
    • props.py: Defines property objects for pages and prompts

Code instructions

These instructions give you some pointers on things you might like to change or add to Feldspar.

Change copy (texts shown on the web pages)
The app has two types of copy:

Currently two languages are supported (Dutch and English). The Translatable object plays a central role and has a Python and a Typescript implementation

From Python code copy can be used as follows:

from port.api.props import Translatable

copy = Translatable({
    "en": "English text",
    "de": "Deutscher Text",
    "nl": "Nederlandse tekst"
})

In React components copy is handled as follows:

import TextBundle from '../../../../text_bundle'
import { Translator } from '../../../../translator'
import { Translatable } from '../../../../types/elements'

interface Props {
    dynamicCopy: Translatable // from Python script
    locale: string
}

export const MyComponent = ({ dynamicCopy, locale }: Props): JSX.Element => {
    const dynamicText = Translator.translate(dynamicCopy, locale)
    const staticText = Translator.translate(staticCopy(), locale)

    return (
        <>
            <div>{dynamicText}</div>
            <div>{staticText}</div>
        </>
    )
}

const staticCopy = (): Translatable => {
    return new TextBundle()
        .add('en', 'English')
        .add('de', 'Deutsch')
        .add('nl', 'Nederlands')
}
Add new prompt
Add the properties of the prompt in [src/framework/types/prompts.ts](src/framework/types/prompts.ts) with the following template:
export type PropsUIPrompt =
    PropsUIPromptNew |
    ...

export interface PropsUIPromptNew {
    __type__: 'PropsUIPromptNew'
    title: Text
    description: Text
    ...
}
export function isPropsUIPromptNew (arg: any): arg is PropsUIPromptNew {
    return isInstanceOf<PropsUIPromptNew>(arg, 'PropsUIPromptNew', ['title', 'description', ... ])
}

Add the prompt component to src/framework/visualisation/react/ui/prompts with the following template:

import { Weak } from '../../../../helpers'
import { ReactFactoryContext } from '../../factory'
import { PropsUIPromptNew } from '../../../../types/prompts'
import { Translator } from '../../../../translator'
import { Title2, BodyLarge } from '../elements/text'
import { PrimaryButton } from '../elements/button'

type Props = Weak<PropsUIPromptNew> & ReactFactoryContext

export const New = (props: Props): JSX.Element => {
    const { resolve } = props
    const { title, description, continueButton } = prepareCopy(props)

    function handleContinue (): void {
        // Send payload back to script
        resolve?.({ __type__: 'PayloadTrue', value: true })
    }

    return (
        <>
            <Title2 text={title} />
            <BodyLarge text={description} />
            <PrimaryButton label={continueButton} onClick={handleContinue} />
        </>
    )
}

interface Copy {
    title: string
    description: string
    continueButton: string
}

function prepareCopy ({ title, locale }: Props): Copy {
    return {
        title: Translator.translate(title, locale),
        description: Translator.translate(description, locale),
        continueButton: Translator.translate(continueButtonLabel(), locale),
    }
}

const continueButtonLabel = (): Translatable => {
    return new TextBundle()
        .add('en', 'Continue')
        .add('nl', 'Verder')
}
Use external Python libraries
Python packages are loaded using micropip:
await micropip.install("https://domain.com/path/to/python.whl", deps=False)

Add the above statement to the py_worker.js file as follows:

function installPortPackage() {
    console.log('[ProcessingWorker] load port package')
    return self.pyodide.runPythonAsync(`
        import micropip
        await micropip.install("https://domain.com/path/to/python.whl", deps=False)
        await micropip.install("/port-0.0.0-py3-none-any.whl", deps=False)

        import port
    `);
}

The standard library is available by default. Please check The Pyodide docs for other packages you can use.

Implement support for alternative web framework
Create a new folder in [src/framework/visualisation](src/framework/visualisation) with a file inside called `engine.ts` to add support for your web framework of choice.
import { VisualisationEngine } from '../../types/modules'
import { Response, CommandUIRender } from '../../types/commands'

export default class MyEngine implements VisualisationEngine {
    locale!: string
    root!: HTMLElement

    start (root: HTMLElement, locale: string): void {
        this.root = root
        this.locale = locale
    }

    async render (command: CommandUIRender): Promise<Response> {
        // Render page and return user input as a response
        ...
    }

    terminate (): void {
        ...
    }
}

Change implementation of assembly.ts to support your new VisualisationEngine:

import MyEngine from './visualisation/my/engine'
import WorkerProcessingEngine from './processing/worker_engine'
import { VisualisationEngine, ProcessingEngine, Bridge } from './types/modules'
import CommandRouter from './command_router'

export default class Assembly {
    visualisationEngine: VisualisationEngine
    processingEngine: ProcessingEngine
    router: CommandRouter

    constructor (worker: Worker, bridge: Bridge) {
        const sessionId = String(Date.now())
        this.visualisationEngine = new MyEngine()
        this.router = new CommandRouter(system, this.visualisationEngine)
        this.processingEngine = new WorkerProcessingEngine(sessionId, worker, this.router)
    }
}
Implement support for alternative script language
To support an alternative for Python scripts, create a Javascript file (eg: `my_worker.js`) in [src/framework/processing](src/framework/processing) with the following template:
onmessage = (event) => {
    const { eventType } = event.data
    switch (eventType) {
        case 'initialise':
            // Insert initialisation code here
            self.postMessage({ eventType: 'initialiseDone' })
            break

        case 'firstRunCycle':
            runCycle(null)
            break

        case 'nextRunCycle':
            const { response } = event.data
            runCycle(response.payload)
            break

        default:
            console.log('[ProcessingWorker] Received unsupported event: ', eventType)
    }
}

function runCycle (payload) {
    console.log('[ProcessingWorker] runCycle ' + JSON.stringify(payload))
    // Insert script code here:
    // 1. Handle the payload
    // 2. Create next command, eg:
    nextCommand = new CommandUIRender(new PropsUIPageDonation(...))
    self.postMessage({
        eventType: 'runCycleDone',
        scriptEvent: nextCommand
    })
}

Change the implementation of index.tsx to support your new worker file:

const workerFile = new URL('./framework/processing/my_worker.js', import.meta.url)

Make sure to add the worker to the ts-standard ignore list in package.json:

"ts-standard": {
    "ignore": [
        "src/framework/processing/my_worker.js"
    ]
}

Note: don't forget to import this new worker file in your server code

Testing

  1. Automatic

    Jest is used as a testing framework. Tests can be found here: src/test.

    Run all unit tests:

    npm run dev:test
  2. Manual

    Start the local web server (with hotloading enabled):

    npm run dev:start
  3. Integration with Next

    To run the Port app on top of Next locally see: https://github.com/eyra/mono/blob/d3i/latest/PORT.md

Technical notes

Code generation

Code in Javascript types and Python api are currently created by hand. Since both of them are implementions of the same model we will seek the opportunity in the future to define this model in a more abstract way and generate the needed Javascript and Python code accordingly.

React

The project is a react app created by create-react-app. This is not set in stone for the future but it was a nice way to speed up the development process in the beginning. Using this strongly opinionated setup hides most of the configuration. It uses webpack to bundle and serve the app.

Code style

The project uses ts-standard for managing the code style. This is a TypeScript Style Guide, with linter and automatic code fixer based on StandardJS.

Pre-commit hooks

Before committing to github Husky runs all the necessary scripts to make sure the code conforms to ts-standard, all the tests run green, and the dist folder is up-to-date.

Funding

Feldspar is part of the Port program for data donation and has been funded by the UU, PDI-SSH (D3i project), and Eyra.

Contributing

We want to make contributing to this project as easy and transparent as possible, whether it's:

  • Reporting a bug
  • Discussing the current state of the code
  • Submitting a fix
  • Proposing new features

If you have any questions, find any bugs, or have any ideas, read how to contribute here.