In some cases, a data science mentor might suggest you take certain online classes, undertake particular project for your portfolio, or develop specific skills. But what if you’ve started your educational journey in the middle and don’t know where to go next?
When I completed my data science course Malaysia at Mobius, I was in the middle. I’d established a strong foundation in data science areas like machine learning, deep neural networks, natural language processing, and more — and my acquisition curve had been rather steep. So I thought that taking another online course wouldn’t teach me as many things per dollar as I’d gotten used to during my Data Science course.
I want to reveal the 5 things I concentrated on that provided me with more knowledge and abilities in data science. To create successful projects and get outcomes, I improved my technical as well as non-technical skills. If you’re at this point, learning more will be best achieved by focusing your efforts on those five items than by taking another online course or undertaking another project.
Implement large projects from start to finish
The majority of people learn most effectively by doing. That is why, in many cases, online courses will first teach you a skill or idea and then ask you to apply it in a little exercise or project. As a result, implementing significant projects from beginning to end with the potential for numerous disasters will educate you even more.
If you do research for an online course, the research question, data set, and sometimes even the models and evaluation metrics are assigned to you. So all you have to worry about is the code and implementation. However, choosing a research question, dataset, model, and assessment metric are the tough parts! If you’ve ever had to create a thesis
Whether you’re a data analyst, BI analyst, or data scientist, one part of your job is to find patterns in large amounts of data without being told what to search for. In other situations, you may be assigned to investigate a certain question but no dataset exists, so you must consider what could be used to answer it and how to acquire it
From the research question to deployment, you should practice your abilities beyond implementation by creating your own project from the ground up. The following sites can assist with this:
- Creating project ideas: If you’re looking for fresh ideas for data science projects, check out my article on how to come up with innovative and implementable concepts.
- 10 distinct methods are covered in this blog article to help you plan and manage your data science project. They include the CRISP-DM, agile, ad hoc, and waterfall methodologies.
- The data science project structure: Cookiecutter Data Science provides a standard project structure that makes it easy to create repeatable, self-documenting projects. It considers all phases of a project, from collecting and transforming data to generating reports with the outcomes.
Aside from the project’s start-to-finish execution, the focus on larger projects will increase your learning. More models, more data sets, and more questions to solve will lead to more issues and difficulties along the road. While struggling may be aggravating, it actually teaches you valuable information and abilities.
Create your own datasets
For many data scientists, modeling is the most fascinating aspect of their job – deciding on the ideal algorithms to utilize, putting them into practice, fine-tuning them, and assessing their results. However, as a professional data scientist, you must spend at least 80% of your time dealing with data collection and cleaning.
If you work for a firm that does not have a data engineer, you will almost certainly be in charge of data collection. So knowing what data is essential to answer a certain research question, where and how to get it, and what pre-processing steps are necessary is critical.You should practice web scraping (but do it lawfully and ethically), get acquainted with sources that offer up pre-existing data and APIs (which you can combine and expand on), and modify the data for further study and modeling.
Many portfolio projects, in contrast, need a one-time acquisition of data. Real-world applications frequently require ETL pipelines that extract, transform, and load fresh information on an ongoing basis. Write a script that continuously extracts fresh data, transforms it, and saves it to a database as an ETL process instead of collecting new data for each project
The following sites will assist you with establishing your own databases and ETL processes.
- Users on Kaggle have shared hundreds of datasets that they produced. Many of them explain the sources and procedures used to gather the data and provide you a hint about where to look for more information.
- Data science techniques by extracting information from websites: Kerry Parker
- I highly recommend checking it out. He wrote an excellent guide on web-scraping for data scientists, which you can find here:
- 22 APIs — including IBM Watson, Spotify, and the US Census.gov portal — are featured in this section to provide data for data science and machine learning.
- Python: A list of resources for creating ETL pipelines in different languages, as well as an explanation of the term “ETL.”
Read academic papers
My favorite resource for getting a high-level view of a topic or grasping the fundamentals of an algorithm is blog articles on Towards Data Science and other websites. A higher level of understanding, on the other hand, will only get you so far.
Reading academic papers that discuss, compare, and contrast algorithms and machine learning techniques will give you a deeper understanding of these topics than any blog post could. For example, you learn why a certain method was developed, how it works on a mathematical level, what other research and models exist to address the same problem, and what future research needs to answer.
Furthermore, reading academic papers aids in keeping up with new advancements in your field. Random forests, XGBoost, BERT, and GPT-3 are just a few of the ML algorithms and NLP models that were created by researchers and published in research papers — which is exactly why they’re so effective.
You’ll be better prepared to explain the inner workings of algorithms, select the correct models for your use case, and defend your choice once you’ve read academic papers on a regular basis. It may be tough and tiring to read scientific writings; however it is worth your time and effort. The following resources can help you get started with academic papers:
- Kyle M Shannon’s Guide to Reading Academic Papers
- This book provides an engaging and entertaining approach to help you understand how your reading habits may be affecting your writing skills.
- Robert Lange’s papers to read
- He runs a popular blog on deep learning and machine intelligence, which he updates monthly with summaries of his favorite new research papers.
- Use an RSS feed to keep up with the newest scientific research. The arXiv feed gives access to studies that are relevant to specific subject areas, such as computer science, statistics, or machine learning.
This article is posted on CoffeeChat.
