Research
Security News
Quasar RAT Disguised as an npm Package for Detecting Vulnerabilities in Ethereum Smart Contracts
Socket researchers uncover a malicious npm package posing as a tool for detecting vulnerabilities in Etherium smart contracts.
data-juggler
Advanced tools
all life is fermentation - Feynman
This library serves little purpose, like all of us and everything we do, but is a bit covered and tested, as you can see.
This libary was __init__.py
iated with typescript starter and uses the devil 👹 tslint.
In the context, the idea is to generalize and abstract the data propagation within an application, that we magically create in all of our projects. This library abstracts that, normalizes, counts, calculates and does many inefficient things that feed into our laziness.
yarn add data-juggler
The usage can be easily seen from the test, assume you are fetching some (csv like for now) data from an API and the columns have the following "type":
data = [
{ height: 190, gender: 'male', timeOfMeasure: 1552397833139 },
{ height: 170, gender: 'female', age: 22, timeOfMeasure: 1552397832139 },
{ height: 164, gender: 'female', age: 20, timeOfMeasure: 15523912333139 },
{ height: 176, gender: 'female', age: 12 }
];
types = {
height: 'continuous',
gender: 'categorical',
age: 'continuous',
timeOfMeasure: 'date'
};
const { data, moments } = dataJuggler(data, { types });
Launch the dataJuggler
function with the sample data and instance types, and enjoy a beutiful dataset full of getters and stuff with everything that you need in it (this is, at least for now, a lie).
You'll get your data back (don't worry) with added properties!
const instance = data[0]
t.deepEqual(instance, {
height: {
raw: 190,
scaled: 1
},
gender: {
raw: 'male',
},
timeOfMeasure: {
dateTime: // the day js instance of the dataset,
isValid: true,
iso: '2019-03-12',
raw: 1552397832139,
scaled: 1
}
})
// true
On top of that you also get some getters for each variable that return the whole column, this could be cut eventually in a censorship attempt by my boss. Edit: the censorship did happen, this is not there anymore.
It could be annoying having to rescale all your functions because some time having them fitted from min-max to 0-1 is not what you need! We got your back. Another entry in the magical return object of the juggler is scalers
! Imaging the classical situation below:
const d = [
{x: 0},
{x: 1},
{x: 2}
]
You can simply apply dataJuggler
:
const { data, scalers } = dataJuggler(d, { types: infer})
Now if you need to rescale the data you can simply do this:
const newExtent = [0, 4]
const rescaleV = scalers.x(...newExtent)
const rescaledX = data.map(datum => rescaleX(datum.v.scaled)) // [0, 0.25, 0.5]
The library accepts as a second parameter a config object of the form
const config = {
types,
formatter: [{ 'height': [/*stuff*/]}],
parser: { 'height': (cm: number) => /*more stuff*/ }
}
const { data, moments, types } = dataJuggler(data, {...config});
As in the example above the types
object is rather explanatory! Let's look at the other ones, but first...
If you feel adventurous you can only pass the data without any types
in the config
object and our advanced (read naive) detecting system will try and determine, mostly leveraging the awesome dayjs ⏰ library, the type for you and then pass it as key types
in the object returned by the function.
You can also pass a custom formatter for each column type as follow.
const formatter = {
height: [
{
property: 'feet',
compute: (datum) => datum * 0.0328084
},
{
property: 'rescaled',
compute: (datum, min, max) => datum / max
}
],
timeOfMeasure: [
{
property: 'year',
compute: (day) => day.format('YYYY')
}
]
}
// if we look for the same instance as before
const dataStore = dataJuggler(data, { types, formatter });
t.deepEqual(istance, {
height: {
raw: 190,
scaled: 1,
// newly added
rescale: 1,
feet: 6.233596,
},
timeOfMeasure: {
dateTime: // the day js instance of the dataset,
isValid: true,
raw: 1552397832139,
scaled: 1,
// newly added
year: '2019',
}
})
// true
What if the data you are passing is not parsed correctly by the library. Once again, no worries! Just pass the parser yourself, the typechecker should prevent you from breaking everything, but hey, what do I know?
const { data: correctlyParsedData } = dataJuggler(datasetWithDates, {
parser: {
d: (day: string) => dayjs(day, 'YYYY-MM-DD').unix()
}
})
const { data: wronglyParsedData } = dataJuggler(datasetWithDates, {
parser: {
d: (day: string) => dayjs(day, 'MM-DD-YYYY').unix()
}
})
You can of course use other functions from within the library. For example, let's say you want to use the autoinference function and do some operations on it you can simply import it and use it
import { autoInferenceType } from 'data-juggler/build/main/lib/utils/dataInference';
const inferredType = autoInferenceType(yourData, yourTypeObject || {}, yourParser || {})
const { data, momemnts } = dataJuggler(yourData, { types: inferredType })
The types you passed in the config will not be overwritten by the library! The same can be done with other functions like computeMoments
and parseDates
. Not sure why you would do that but knock yourself out.
So far we have been talking about cases where you know your data at "compile time" and neither does Typescript. This hinders the ability of the library to help you structure stuff correctly but all the core functionalities are intact and shining bright. Let's assume you need to fetch some data, of which you know the structure but you don't "have" it, you need not to worry! The function dataJuggler
accepts one generic which is the type of the datum so you can do something like this:
interface ExpectedDatum {
height: number,
name: string
}
const raw = ky.get('this.is.a.fancy.end.point.with.many.points.com/get').json()
const juggled = dataJuggler<ExpectedDatum>(raw)
If, for example, you use Mobx state tree you could do the following:
import {} from ''
export const YourFancyMSTModel = t
.model('YourFancyMSTModel', {
juggleType: t.frozen<InferObject<YourDataStructure || Record<string, unknown>(),
juggledData: t.frozen<JuggledData<YourDataStructure || Record<string, unknown>(),
moments: t.frozen<MomentsObject<YourDataStructure || Record<string, unknown>(),
})
0.5.10 (2020-05-20)
<a name="0.5.8"></a>
FAQs
_all life is fermentation - Feynman_
The npm package data-juggler receives a total of 3 weekly downloads. As such, data-juggler popularity was classified as not popular.
We found that data-juggler demonstrated a not healthy version release cadence and project activity because the last version was released a year ago. It has 10 open source maintainers collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Research
Security News
Socket researchers uncover a malicious npm package posing as a tool for detecting vulnerabilities in Etherium smart contracts.
Security News
Research
A supply chain attack on Rspack's npm packages injected cryptomining malware, potentially impacting thousands of developers.
Research
Security News
Socket researchers discovered a malware campaign on npm delivering the Skuld infostealer via typosquatted packages, exposing sensitive data.