Topics: Coronavirus | AI | Data Science | Deep Learning | Machine Learning | Python | R | Statistics

KDnuggets Home » News » 2020 » Oct » Tutorials, Overviews » 10 Days With “Deep Learning for Coders” ( 20:n38 )

10 Days With “Deep Learning for Coders”


Read about the author's experience with the course and the book from fast.ai.



By Arnuld on Data, Freelance Data Scientist

Figure

Image by Ryan McGuire from Pixabay

 

The Background

 
I started Practical Deep Learning for Coders 10 days ago. I am compelled to say their pragmatic approach is exactly what I needed.

I started data science by learning Python, Pandas, NumPy, and whatever I needed in a short few months. I did whatever courses I need to do (e.g. Kaggle micro-courses) and whatever books I needed to read (e.g. Python for Data Analysis). All of this I did as a part of a 90-day MOOC’athlon learning challenge started back in April this year. It was one of the greatest learning periods of my life. After this, I completed both Iris data and Boston house price prediction projects at Kaggle. Then I scraped data from the internet and used my Pandas skills to clean them out. And implementing research papers was still confusing and frustrating. I struggled with research papers and Kaggle projects. I am good at C language. You can give me any programming language and I can learn it pretty quickly than most because programming is my 2nd nature. Computer Programming just flows in my thoughts. And this wasn’t happening with data science. Even though there were learning and a lot of work but in the end, I couldn’t remember any of what I did (yes, data wrangling is a hell lot of work). I was in a constant state of frustration. I started getting irritated at little things and it spilled into my personal life. I thought of quitting data science and machine learning. It had been 4 months and I was nowhere near having the capability to accomplish anything deemed usable for business. I felt trapped in a 4x4 feet concrete box.

 

End of Frustration

 
I am always in constant search for different approaches to problems. I consider a different point of view as a blessing from the universe. Yes, we all have our own “point of view” and it is always refreshing when someone approaches the same problem from a different and strange angle you never thought of. What changed my situation was a blog-post by Caleb Kaiser:

Don’t learn machine learning
Learn how to build software with ML models

 

His blog-post made me understand that this frustration doesn’t belong just to me, it is what every coder/programmer can get, that I am not alone. This completely changed my perspective and I decided to use his approach. I decided to do both: read the book and watch the videos.

In just one week with this course, I was able to comprehend this research paper on weight-poisoning by Appsilon Data Science.

Another few days and after many failures, I successfully implemented the bear detection model as a web-app on binder. You can try this web-app. Just pass it an image of any of:

  1. American black bear
  2. Grizzly
  3. Polar bear
  4. Teddy bear

and it will detect which one it is. Pass it anything else, a shoe e.g., and it will try to match it with 150x4 images of bear pics in the model and do its best. It is not human =:o) :
https://mybinder.org/v2/gh/ArnuldOnData/fastai-projects/master?urlpath=%2Fvoila%2Frender%2FCh02%2Fbear-detection.ipynb

Figure

Photo by Ben White on Unsplash

 

And I seem to remember most of what I did. Data science is finally becoming my 2nd nature. This fastai course has given me more confidence to pursue more and higher machine learning knowledge. I think if some learning approach doesn’t work for you then you need to give the pragmatic approach of this course a try.

 

The Why

 
It is not to say that traditional approaches of learning like doing MOOCs and then Kaggle projects or the university route of graduating in data science don’t work. They might or they might not. People are different and we all have different backgrounds and we can’t all be following traditional or hybrid approaches. Times have changed. Thanks to the internet, its decentralized nature, and net neutrality, the world is far more connected than it used to be. The world has become a smaller place where everyone can connect to everyone. This has resulted in the 4th industrial revolution:

Figure

Image by Christoph Roser at http://AllAboutLean.com

 

Figure

Klaus Schwab, Founder and Executive Chairman of World Economic Forum, Text from Wikipedia

 

4th industrial revolution will lead to:

  • Lights out manufacturing (fully automated factory which requires no human presence on-site)
  • Smart factories will monitor physical processes and create a virtual copy of the physical world and make decentralized decisions.
  • Over the Internet of Things, cyber-physical systems communicate and cooperate with each other and with humans in real-time both internally and across organizational services used by participants of the value chain

As per my understanding, The 4th industrial revolution way says we need to learn tools fast, we need to adapt to new technologies and systems quickly. Instead of taking 2-4 years to learn and prepare for work, you need to have the aptitude and attitude to jump right into the work and build something usable, something that adds value to the business in the next few months. We need to learn differently from the ways we used to. I think Jeremy Howard, Rachel Thomas, and Sylvain Gugger’s uncool approach to learning is what is needed in this 21st-century industry.

 
Bio: Arnuld (@ArnuldOnData) is an industrial software developer with 5 years of experience working in C, C++, Linux, and UNIX. After shifting to Data Science and working as data science content writer for over a year, Arnuld currently works as a freelance data scientist.

Original. Reposted with permission.

Related:


Sign Up

By subscribing you accept KDnuggets Privacy Policy