In this part of the article, I am going to share some ways to keep on remaining updated with new concepts, state-of-the-
It’s totally wrong! There is dense literature on this topic: how to train a neural network, which type of networks works in a particular contest, what is the hardware architecture required to train a neural network of a certain size. To start, you can take the courses from the Deep Learning Specialization while as a textbook check Deep Learning. Two of the most used frameworks for building artificial neural networks are TensorFlow and PyTorch. There are many materials online starting from their official websites.
So if you arrived here, you already know a bunch of algorithms to build models, you are confident to develop a data science project with well-known packages and you already have an idea of how neural networks work. It’s time to have a full immersion into deep learning! The common misconception about deep learning is that it is just a series of layers of neurons that can be fit to predict anything, as magic.
Colab on steroids: free GPU instances with SSH access and Visual Studio Code Server A step-by-step guide to SSH into a free GPU instance provided by Google Colab and install Visual Studio Code Server.towardsdatascience.com
As the last point of this guide to become a great data scientist, I would like to point out the difference between the concepts and the tools. The concepts remain after you have learned them but the tools could change in the future. How a neural network works is a concept but its implementation can be done in Tensorflow, PyTorch, or any other framework. A data scientist should be open to innovation. If a new language is more suitable for your project, learn it, and use it! Don’t be anchored to your tools! Try new stuff!
Basic mathematical courses include linear algebra, real analysis, and numerical analysis. Basic statistics courses include descriptive statistics, inferential statistics, bayesian statistics. Basic computer science concepts consist of: time and space complexity, data structure, sorting and searching algorithms, algorithms on graphs, and algorithm design. You can find all these arguments in any undergraduate coursebook. And of course, Google is your friend!
I want to dedicate a section to big data projects. Often, to train machine learning models in a big data context, basic packages for traditional data science projects are not enough!
Hopefully, you are already working as a data scientist when you reach this section. By working in a real-world context you will find out that technical knowledge is not the only part of a data science project. Soft skills and experience are the main ingredients to conclude a successful project! As a data scientist, you will interact with the business to transform the business needs into technical requirements, you will work in a team together with other professionals (for instance, data engineers, dashboard designers, etc.) and be able to share your findings both in a technical and non-technical way. Also, as you have already realized, for each problem setting the number of approaches to test is always larger than the time you have to conclude the project.
If you got here, you are almost ready! You should have filled up your GitHub repository with reusable code, interesting use cases, and application examples! Participate in an online competition on Kaggle! You could win money! :) Most importantly, on Kaggle you can learn from others on a variety of problem settings! You will realize that what you know is only a minimal part of the tools, methodologies, and algorithms that the data science community has developed.
Most of the time, the key component of a data science project is data visualization. A data scientist should be able to communicate its insights with effective plots or make the results usable by the business through a dashboard. If you are using Python, matplotlib, plotly, seaborn and streamlit are common choices. In R there are ggplot2 and Shiny Dashboard. If you like the hard way, you can try D3.js. You will find out that there are several data visualization solutions out there and it is impossible to learn to use all of them! My philosophy here is to learn just a couple of them to be ready to create a prototype to show the results. In many projects, the final data visualization solution is already chosen because part of an existent workflow and you only need to create endpoints to deliver the output of your models.
Custom neural networks in Keras: a street fighter’s guide to build a graphCNN How to build neural networks with custom structure and layers: Graph Convolutional Neural Network (GCNN) in Keras.towardsdatascience.com
http://live-stream.munich.es/twr/video-Al-Rayyan-Al-Gharafa-v-en-gb-1ykx30122020-24.php
http://live07.colomboserboli.com/niy/video-Al-Rayyan-Al-Gharafa-v-en-gb-1khz-23.php
http://news24.gruposio.es/ktn/Video-Qatar-SC-Al-Kharaitiyat-v-en-gb-1rxf-4.php
http://news24.gruposio.es/ktn/video-Al-Rayyan-Al-Gharafa-v-en-gb-1ztg-.php
probability. In one conversation, I mentioned to him that probability riddles would be very useful for people who want to get into the most scientifically applicable scientific subject in the world (my very, very biased opinion). What I said earlier about play is even more applicable to probability, a field that really started with gamblers, used by traders and adventurers, and perfected by finance and insurance mathematicians. Probability applies to all empirical fields: gambling, finance, medicine, engineering, social science, risk, linguistics, genetics, car accidents. Let’s play with it by adding to his feed some probability riddles.You will face different technologies to ingest and manipulate big data. We are at the border between data scientists and data engineers here. Usually, the big data development environment is set up by engineers but data scientists should know how to use these new tools! Hence, learn to use Kafka, Spark, Scala, and Neo4j!
Until here, we have seen a learning path to become a great data scientist. The list is far from complete and it is based on my personal experience. If you have any suggestions, please drop me a comment! Of course, the order of this list is only recommended, feel free to adapt it to your needs!
Until now, you have studied to become a general data scientist. You know a little bit about everything but you are not an expert on anything! And while you are learning all the stuff to be here, other thousand people are doing the same! So if you want to be different from the rest, you need to specialize in something! You could be an NLP specialist, a CV specialist, an expert in network science, or a data visualizer… the list could be very long but remember: the specialization shouldn’t be only technical but also for the business field! It is easy to find a data scientist expert in time series forecasting, less easy if the data scientist should also be an expert in the stock markets.
A growing trend in recent years is the online platforms for the developments of data science projects offered by Google, Microsoft, and other players. These platforms mainly offer a development environment and already trained models as services. Again, the setup of these environments are mostly done by data engineers but knowing that certain type of services already exists online and which types of pre-trained models you can use as a service can boost your productivity and shorten the delivery time! Thus, pay attention to these cloud platforms because many companies are already using them!