ProjetJS

Authors :

• Tristan CHEVET
• Emilie ROUX
• Norbet ASTIER
• Massimo BACCI

Description

In this project :

• Reading a csv file and putting it in a dataset
• Shuffling of the dataset
• Creation of a trainset and testset
• Chi² test on the test dataset
• Principal Component Analysis of the dataset

About Chi² test

The Chi² test is to make sure our test set is not biased.
In order to have a good accuracy on a model we need to have a good equirepartition of data in the test set.
Therefore we run the Chi² test on 2 list, OBSERVED and EXPECTED.

Chi² needs a null hypotesis, here we used « The EXPECTED and OBSERVED values are independant ».
If the null hypotesis is rejected it means the test set is biased.
If the Chi² test fails to reject null hypotesis it means the test set has a good equirepartition.

Example:
We have a dataset of 30 DOG, 30 CAT and 30 GIRAFFE.

The null hypotesis is « The EXPECTED and OBSERVED values are independant »

OBSERVED is the real quantities of values in the test set so [5,15,10]
EXPECTED is the quantities of values we want in the test set so [10,10,10]\

Here you can see that the precision is not going to be good because we have more CAT than DOG.
So Chi² test will reject the null hypotesis.

Read more here https://towardsdatascience.com/chi-square-statistic-chi-squared-distribution-2499084b5da8

About PCA (Principal Component Analysis)

PCA is a statistical procedure that uses an orthogonal transformation that converts a set of correlated variables to a set of uncorrelated variables.

In this project we used the famous Iris-dataset.
We took 4 columns to analyse

• sepal length in cm
• sepal width in cm
• petal length in cm
• petal width in cm

With the help of the librairie pca-js, we found we only needed 2 variables to analyse the dataset and thus reducing the dimension of 4 to 2.

Installation

In order to run the project, you need to install Node JS : https://nodejs.org/en/download/

You then ned to install all of the necessaries libraries with the following command :

With npm :

npm install

Start the project

To start the project you just need to execute the following command :

With npm :

npm start

Test the project

To test the project you just need to execute the following command :

With npm :

npm test

Test the code-style of the code

To test the code-style of the code you just need to execute the following command :

With npm :

npm run code-style

Github Actions

When we have a pull request, we use Github Actions (the «pull_request_git.yml» file) to make sure the code:

• Doesn't merge directly
• Passes all required test (the style-code and the npm test) to see if there is any mistake in the code

Libraries

ramda :

https://ramdajs.com

@stdlib/stats-chi2test (For the equirepartition on a dataset test) :

https://www.npmjs.com/package/@stdlib/stats-chi2test

pca-js :

https://www.npmjs.com/package/pca-js

chai :

https://www.chaijs.com

gulp :

https://gulpjs.com

mocha :

https://mochajs.org

xo :

https://www.npmjs.com/package/xo