Nowadays every programming language has one or more package managers, but first let's explain this. It's a command line tool that allows us to add software packages to a project, that are public and high quality. Also we can track versions of packages and install all necessary dependencies. Programming languages such as PHP has composer, Rust has package manager cargo and for nodejs and javascript source code are npm and yarn.

What is NPM

Npm is the default package manager for nodejs, we can search and browse available packages in the official website npmjs.com. When we use npm in a project first we have to create an important file  that is named package.json, this will contain the list of packages that we will use in json data format, the source code always wei be stored in a folder called node_modules. Creating this file is easy just by typing the command.

npm init

When we push new code to the repository of our project, usually the we do not save the files from node_modules folder there, we only add main file package.json. In case we want to install the project to another computer we just cloning the repository and then reinstalling packages by typing in cli the following.

npm install

Some of the popular packages of npm is the express a web framework for nodejs, another is the React Native from facebook which is a tool for building native mobile applications.

For checking if some of the modules are old and there are newer version we do it using the command, this check the versions from package.json file against the online registry of the package manager

npm outdated

After that we can update all modules but this it's best to do it only when we properly have setup software testing procedures. Also we can make individual updates per package by using their name e.g. for express package

npm update
npm update express

What is Yarn

First few years after npm was created there were many setbacks and bottlenecks like low security, not deterministic installation and slow download and installation speed for packages. So facebook created Yarn as an alternative package manager for nodejs to overcome all these problems.

When you do an installation through yarn for a new package the exact version and related dependencies and metadata are saved to lock file yarn.lock . After 2018 npm also improved and add support of auto-generation of a lock file, so now can support deterministic installation, the lock file is named package-lock.json.

The official site is yarnpkg.com and  the logic is the same for installing and updated software and from package.json, to create one type

yarn init

for installing packages

yarn

The last years npm overcome previous problem so npm and yarn are almost the same, so it's matter of preference.

Installing Tensorflow for machine learning

As mentioned previously, package managers can simplify the use and maintenance of external resources like libraries and APIs. Here an example, let's install the javascript library for Tensorflow, it is a ML library, that you can use it in custom web based project, installation is easy with the following command.

npm:

yarn add @tensorflow/tfjs

yarn:

yarn add @tensorflow/tfjs

So now we can easily build and train machine learning models using a purely javascript library directly in any browser. Also there are other versions that of tensorflow if you want run as a backend in server side through nodejs.

Setting up environment for ML is easy, now if we to use predefined deep learning models to implement directly data from to a web application, using keras. A detailed guide to import python keras models in Tensorflow.js there is the official website.