# Index

# How this guide is designed

This guide is designed in the same fashion as the Purgatorio structure, so you find a subsection for each phase of the Data Science process.

For each section, we list for you different libraries and the best resources to learn them.

We'll give you some general tips to learn effectively and develop rock-solid foundations, that you can rely on to address and solve Data Science problems in the complexity of the real world (which is messy by definition).

In this guide, you'll find a ton of different small tools, big libraries, and even entire frameworks, but be aware!

As stated in the Virgilio's Teaching Strategy (opens new window) Guide, the best way to deal with the learning process is to leverage the "Divide et Impera (opens new window)" principle, which states that every time you're tackling a problem (in this case, the learning problem) you should split it into smaller parts, easier to handle and understand.

# Solving the Learning Problem

The process of learning a tool and use it effectively is equal to solving a problem.

We consider "solved" the learning problem when you can use the tool comfortably for the task at hand.

So, you shouldn't try to learn all the features of a library in the same run, or you're going to be overwhelmed!

Every rich Python library contains thousands of different functions and utilities, but you'll need a small subset to get the things done!

So, Virgilio suggests the following approach:

  • Choose a library from the ones that are widely used and feature-rich.
  • Get a general introduction through to Virgilio's resources listed in this guide.
  • Find a problem (similar but not equal to the introductory ones).
  • Dig into the problem, and get your hands dirty with the chosen library.
  • Learn on the way to the problem solution.
  • Solve the problem, trying to do it yourself.

If you follow this approach, you take home the following 3 things:

  • real problem experience
  • you feel comfortable coding with a subset of the functions of the tool

Believe it or not, it works like magic!

Know you know how to solve a new learning problem.

Bonus tip:

When trying to know about a library or framework or tool, use these two methods:

  • search for them on Medium.com
  • search for them on YouTube or another video/learning channel

Both these will show how others are using it or how else it can be used, often better than learning from the documentations initially.

Now, how to choose from the myriad of tools and libraries available?

# Navigating the Ocean

It's very easy to find out articles, blog posts, and YouTube videos telling you "the 20 libraries you need to learn in 2020!".

What? 20 libraries?

Come on, this is ridiculous and impossible, for a beginner!

Don't even try to do that, or you will miserably get stuck, because it is impossible to learn 20 programming libraries together, even for an expert.

Learning tools and libraries will be a thing most of your life if you want to study or work in the Data Science field.

But, as a beginner:

  • is better to start learning very well some tools/libraries, and meanwhile becoming "aware" of the similar ones (and their pro/cons).

  • For doing that, it's useful to read the homepage of the official documentation (God bless good documentation!).

  • It's way better to learn one (max 2) libraries for each step of the Data Science process, as a starting point!

  • You'll always be capable of learning more libraries when you reach the capabilities limits of the one that you already know or you have different needs.

Now you know what is the best approach to learn tools and libraries and frameworks, and how to do it using wisely your time (opens new window)!

The following links expand these points and you should read them if you are a beginner, some answers are illuminating:

Having that said, let's jump in!

# The Data Science Process Toolkit

# Data Collection

Data collection is one of the most crucial parts of a Data Science project (you don't say?), and often a tedious one if you have to label your data on your own!

Luckily, good researchers and institutions often release their datasets, so you don't have to collect the data (or label it), just download!

But if you can't find the dataset that you need?

Let's see at some tools that can help you in this task:

See also:

Remember that a lot of free access datasets are available for free, for example in the UCI Dataset (opens new window) website or at the Data science Challenge platform Kaggle (opens new window).

Check also this repo:

# Data Preparation

Data preparation is often the most time-consuming part of the Data Science project (opens new window), so it's crucial to make sure that the right tools are used from you or your team.

Often the right tools in this phase are the programming languages and their libraries.

The most popular and widely-used libraries are:

# Data Visualization

As you learned in the Data Visualization Guide (opens new window), the human race evolved in the African savannah through thousands of years, and the sight is most advanced surviving technology that he has developed.

More than 60 % of the actual computation power of our brains is all-time processing visual information, and using it to create an internal representation of the world!

So, it's easy to understand why to visualize data and their relationships it's compelling to get rich insights from them.

Also, remember that data visualizations are the way to communicate your results to other people, including the non-technical audience.

You can get an overview of the most popular and wide-used data visualization libraries in the Data Visualization Guide (opens new window).

One thing to note Data Visualization isn't properly a separate step or stage, it is done at every step and stage before and after doing a step.

Check also this repo for a very exhaustive list of Data Visualization tools:

# Feature Engineering

The feature engineering is the process of transforming your data into useful features, that can be consumed by ML and DL algorithms during training.

It's hard to suggest (opens new window) a single tool for feature engineering: currently, feature engineering is still largely a laborious and manual process, and it heavily depends on your data.

By the way, you can consider the way Scikit-Learn applies transformations to data, as a good starting point.

Scikit-learn provides a library of transformers, which may clean (see Preprocessing data), reduce (see Unsupervised dimensionality reduction), expand (see Kernel Approximation) or generate (see Feature extraction) feature representations.


# Machine Learning

Fortunately for all aspiring Data Science practitioners, there are now mature frameworks and libraries that implement most of the best algorithms to train models from data.

Like any branch of computer science, you can find dozens of similar libraries that do more or less the same things, and in the case of Machine Learning model building, it's no different.

Virgilio suggests you focus on one particular library in order not to risk wasting your time and especially because it is the most widespread and supported.

Scikit-Learn (opens new window) is one of the most complete, mature and well-documented libraries for Machine Learning tasks. It comes out-of-the-box with powerful and advanced models that offer facility functions for the data science process.

Also, they have a wonderful map (opens new window) to guide you in finding the right ML model for the problem you want to solve.

Scikit-Learn resources:

Also, Scikit-Learn is the library chosen by the awesome book Hands-On Machine Learning with Scikit-Learn, Keras, and Tensorflow (opens new window), which is highly recommended to have in your pocket as a learning companion.

Additional Machine Learning libraries and tools:

# Deep Learning

When it comes to Deep Learning it is not easy to decide which framework to recommend to a beginner.

Out there, on the Internet, at the query "which Deep Learning framework to learn first?", you quickly understand that the two most complete frameworks are Tensorflow (opens new window) (from Google) and Pytorch (opens new window) (from Facebook), both open-source and full of useful features to work with Deep Learning.

After evaluating several factors as:

  • completeness and maturity of the framework
  • usability and learning curve
  • how easy to understant is the official documentation
  • number of official and unofficial examples, guides and tutorials
  • industry-side diffusion

Virgilio suggests you start with learning TensorFlow and try Pytorch later.

Brief argument: Tensorflow, especially from 2.0 version and with the Keras (opens new window) Interface (native in TF from 2.0), is a better choice for a beginner. Pytorch has different pros but it's more suitable for research and high flexibility needs. TF 2.0 is the tools chosen in the up-to-date version of the mythical (and "mandatory") book Hands-On Machine Learning with Scikit-Learn, Keras, and Tensorflow (opens new window).

To learn TensorFlow, start with the Keras wrapper!

# Keras

Keras is an Open Source Neural Network library written in Python that runs on top of TensorFlow.

Easy to use and widely supported, Keras (opens new window) makes deep learning about as simple as deep learning can be!

There is a reason why Google engineers decided to go integrate Keras natively in TF 2.0: because it's awesome! Most of the time you don't need to dive so deep inside a neural network architecture to make it work!

Start from the official Keras guide integrated in the TF Docs:

# Tensorflow 2.0

To learn TensorFlow you'll need mostly a lot of practice, starting from toy examples to most complex projects, but Virgilio collected for you the most important resources out there:

You should be able to find different frameworks and dozens of other exotic neural network libraries on the Internet, but as usual, Virgilio highly recommends to use your time wisely (one enemy at time). It's not always true that the most used thing is the best, but for sure you can be certain that Tensorflow is here to stay and to remain one (or the most) widely used deep learning framework.

Don't try to learn the old TF1 version, hadn't aged well, and Google learned a lot from that experience: go for 2.0.

# Hyperparameter Tuning

If you’ve ever tried to tune hyperparameters for a machine learning model, you know that it can be a very painful process!

To fine-tune your models you can check a variety of tools, check this article:

In particular, if you are learning Keras (opens new window) as a top-level framework on the top of Tensorflow, Virgilio suggests you check Hyperas (opens new window).

A very simple convenience wrapper around Hyperopt (opens new window) for fast prototyping with Keras models. Hyperas lets you use the power of hyperopt without having to learn the syntax of it. Instead, just define your Keras model as you are used to, but use a simple template notation to define hyper-parameter ranges to tune.

# Serving Models

After training Machine Learning models, you may want to make them available as a service, to your users, or to other computer systems that may use them. The most general way to serve a trained ML model is to use a library like Flask (opens new window) API.

Here are some resources to get started with this framework:

Then follow these tutorials to learn how to use Flask to serve your trained models:

The best way to understand something is to do it! So, check your knowledge with this 2-hours guided project from Coursera.

Also noteworthy is the Streamli project:

Streamlit is an open-source Python library that makes it easy to build beautiful custom web-apps for machine learning and data science.

# Automation Tools

There's an old rule that says:

If you have to do a job more than 3 times, it's time to automate it.

Read this intro to Automation in the Data Science process:

Here you can find a very comprehensive list of automation tools, targeting different parts of the Data Science project.

# Miscellaneous

Virgilio lists here for you a bunch of incredible good GitHub pages with lists of hundreds of tools and libraries you can find useful when working in the Data Science environment. What is the best way to use these pages? The answer is that every problem has different characteristics and caveats, and you should be able to understand if you need to learn a new tool, or you can get the things done with those that you already know! So, use these pages to compare the pros and cons of the hundreds of tools that exist!

# Production Ready Tools

In this section, Virgilio listed for you some "Production-Ready" tools that can speed up the process of creating value with data. You can use these tools along with those listed in the previous sections.

For a comprehensive list of production-ready tools, be sure to take a tour at:

Virgilio has chosen for you the best tools of this type for 3 problem categories:

# Computer Vision

OpenCV (Open Source Computer Vision Library) is an open-source computer vision and machine learning software library. OpenCV was built to provide a common infrastructure for computer vision applications and to accelerate the use of machine perception in commercial products. Being a BSD-licensed product, OpenCV makes it easy for businesses to utilize and modify the code.

Learn how to use OpenCV here:

Check also Virgilio's tutorials in the Inferno section, They all use OpenCV!

# Natural Language Processing

SpaCy is a free, open-source library for advanced Natural Language Processing (NLP) in Python.

If you’re working with a lot of text, you’ll eventually want to know more about it. For example, what’s it about? What do the words mean in context? Who is doing what to whom? What companies and products are mentioned? Which texts are similar to each other?

spaCy is designed specifically for production use and helps you build applications that process and “understand” large volumes of text. It can be used to build information extraction or natural language understanding systems or to pre-process text for deep learning.

They offer a free course on their website:

# Time Series Forecasting

Prophet is a procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects. It works best with time series that have strong seasonal effects and several seasons of historical data. Prophet is robust to missing data and shifts in the trend, and typically handles outliers well.

Prophet is open source software released by Facebook’s Core Data Science team. It is available for download on CRAN and PyPI.

Get the tutorials here:

# Conclusions

In this guide, we listed A LOT of tools, libraries, and a project that can help you be faster and more effective during the Data Science process.

The amount of tools listed here is enormous, so choose wisely what you really need for your problem!

And remember, no one is going to ask you to know 10 different Computer Vision tools, but they are going to ask you to know one of them very well!