I had started with the UW ML specialization in Coursera. Although the course is good, it takes a more theoretical approach. The concepts are explained nicely but by the time, you put it to use, you forget what has been taught and you need to re-read the lessons.
The question i have is, in terms of the topics and the breadth of the material covered, how different are these 2 courses?
Does Fast.ai ML course cover topics which are entirely different from what is covered in UW ML specialization in Coursera?
âThe question i have is, in terms of the topics and the breadth of the material covered, how different are these 2 courses?â
I personally did this Specialization about two years ago. This course on coursera is more like Bottom Up instead of Top Down as the Fast.Ai is, however many details from the botton up approach are not covered, because is a introduction course.
âDoes Fast.ai ML course cover topics which are entirely different from what is covered in UW ML specialization in Coursera?â
There is some differences because ML in UW Coursera cover a broader range of machine learning algorithms but the fast.ai course covers the most practical of it these days. The Specialization on Coursera was developed in 2015 to 2016 while the Fast.Ai is from 2017 until now.
Also the UW ML never touched some context promised like Neural networks and on fast.ai âMLâ you will be dealling 50% practical ML skills and 50% Neural Network introduction and you will have more two courses to watch, DL1, and DL2 that will be entirely focused in Deep Learning.
Hello @willismar Thanks for the response.
So, if i were to totally skip the UW ML course and instead just focus on the fast.ai one(ML + DL1 + DL2), do you think i would be in good shape to begin with? I want to supplement my learning with the material in the book: Hands on Machine Learning with Scikit-learn and Tensor flow.
What are your thoughts? Do you think i can totally skip the UW Coursera course?
Do you think i will benefit more by taking all of the courses on Fast.ai than the Coursera one?
Well I am suspect to tell, because I did the UW ML and I loved it, but I donât suggest you drop the specialization either .
âWhat are your thoughts?â
In my time UW ML used a package called GraphLab Create that was awesone, but Apple bought it and opensourced it recently with another name: Turi Create and almost nobody knows or use that package because lacked maintenance for a long time until apple opensource it.
Fast.ai course you will use fastai library that is built on top of Scikit-learn and Pytorch (beloved by researchers) and others packages too and also has a lot of maintenance because was born on the concept of opensource.
If you want finish the course and put your skills in a good use I say that fast.ai is the course that you need to give more attention in your life.
If you want to know more about other general Machine Learning Algorithms for Supervised and Unsupervised Learning I recommend you also watch too the course on UW ML but as a secondary course.
âDo you think i can totally skip the UW Coursera course?â
I absolutely donât recommend you skip fast.ai because is growing fast and changing fast as new algorithms and techniques came along while UW ML is stuck on the past. You can UW ML and know things, but may not be applicable today (at some degree) while 100% of you learn from fast.ai is applicable today.
âDo you think i will benefit more by taking all of the courses on Fast.ai than the Coursera one?â
Absolutely, If I was to start again, I would definitely start from fast.ai because the practical skills you get is like you get ready to go. UW ML approach is more like: âI finished the course and now what ?â
I hope this help you decide where invest more of your time wiselly
@willismar Thank you very much. One last question. Can the book Hands on Machine Learning with Scikit-learn and Tensor flow. act as a replacement for what is taught in the UW Coursera course for supervised and unsupervised learning?
Does the book have the same depth?
Doesnât the fast.ai course cover supervised and unsupervised learning?
" Can the book Hands on Machine Learning with Scikit-learn and Tensor flow. act as a replacement for what is taught in the UW Coursera course for supervised and unsupervised learning?"
Definetly I extremely recommend any book. I Forgot to mension the intent of the Specialization of UW ML couse. They was planning to teach supervised and unsupervised learning during the following modules:
Course 1- Machine Learning Foundations: A Case Study Approach (recorded)
Course 2- Machine Learning: Regression (recorded)
Course 3- Machine Learning: Classification (recorded) They canceled SVM lessons to extends the talk about Decision Trees and Boosting Algorithms (AdaBoost)
Course 4 - Machine Learning: Clustering and Retrieval (recorded)
Course 5 - Recomender Systems (canceled)
Course 6 - Capstone project, Neural Networks, Deep Learning and Transfer Learning (canceled)
The Time the Courses 5 and 6 was cancelled, Apple was announcing the acquisition of Turi Company. From there you can imagine what may have happened that the professors couldnât even explain about on the forums.
âDoes the book have the same depth?â
I believe so.
âDoesnât the fast.ai course cover supervised and unsupervised learning?â
Just supervised learning⌠I believe you can verify by professor Jeremy pod cast interview what he say about his course and library.
I have read the book and can totally recommend it as a beautiful intro in the world of machine learning as well as a wonderful tensorflow tutorial. It has much more depth than most ML/DL courses out there. I would certainly recommend the book as a supplement for fast.ai course: sometimes you really want to get to the theoretical background of some techniques (say, gradient descent with momentum) without waiting for the next lecture. In this case, it is a wonderful companion.
Hey @willismar thanks for your great answers. I have a few questions of my own.
I am starting fastai ML course . I guess, this is based on the old fastai library and not on the recently released v1. So, is there a disadvantage bcoz of that? I think they arenât compatible, so is there a way I can migrate my learnings from the old library to the new one because i think the new version of the ML course (with fastai v1) will take some time to happen.
I am doing Andrewâs Coursera course. But, I read on reddit, that the version hosted on Stanfordâs site is much more in-depth than the coursera version. Since, I am a CS student, I want to do both application( via fastai) and theory. But, the in-depth version is of 2009 and @jeremy has said that Andrewâs courses are somewhat outdated. So, any suggestions in this area. Should i just do the Coursera version ?
Hey @noisefield , i have a few Qs regarding the Hands-On book. Tensorflow will be releasing v2 by the end of the year,so, I am wondering whether this book would become outdated?
Does the Hands-on book have complete mathematical explanations(i want that)
Also, I am doing Andrew Ngâs Coursera course,so, I assume a lot of material in that book would become redundant, right?
Also, what do you think about this book : https://www.amazon.com/Introduction-Machine-Learning-Python-Scientists/dp/1449369413
I donât know ⌠I donât get a notebook and just play it and finish. I donât know if was that you meant. But I get it and I retype everything on my own. And as I get any trouble I fix it to reflect the same from the video material or the old version. I donât know how you plan to organize your self , but I can even use the older and the newer version at the same time. For that you can just set it on your system as please. I donât follow manuals to setup my learning box because I have a long journey with linux and computer science so I donât bother to change things. For real I enjoy it.
True, both of the materials from Coursea and Stanford version are in Stanford Site. One is CS229 (more theoretical ) and the other is CS229A (Application of machine learning - like a introduction without requirements)
About doing both together with fastai, I believe you will have to run a lot. Remember that learn itâs not about a competition to anybody.
If you just started to learn Machine Learning try easy classes and keep increasing the level. Start from Fastai, then jump to Coursera , and fly out to others. There are many other courses out there including Master Level classes, with very high material, here is some channels to you look at coursera, edx, udacity, ocw. Take a few minutes to find them on google and the courses related inside them.
Also keep in mind that learn takes time. There is a saying :
Who hears, remember.
Who sees, understand.
Who practice, learn.
Donât get stuck by the needs of doing everything and watch everything because knowledge fades away in few weeks without continuous practice. Eventually you will be tired and will be bored and will procrastinate. So doing one course at time you will not get stressed by needs of doing everything. I personally started in 2012 and never stopped. Every opportunity to get a new vision from other people I stop to watch and learn what he / her can say about it that I can understand better. There is no rush.
A last note the scenario keeps changing year by year. When I started I thought that Deep Learning was a monster of five heads (Dungeons and Dragons - Tiamat) but after I started to learn I realized that is not.
Deep learning is moving fast and kicking machine learning scene too fast and many frameworks been released to support many and more techniques. If you get frozen by some course you will get stuck in the past. Fast.ai and Pytorch keeps bringing what are the top techniques to your hands by this course. So you can decide if you want spend time learning fundamentals first or have practice first. When I started the frameworks available was very limited and we had to build things from scratch. Now itâs a paradise the amazing work that is shared everywhere.
You will get more out of doing projects and coding. UW course is good but very basic. I did the first part of it but did not complete. I would recommend to focus on coding over courses and unless you like the math and theoretical aspects not to take most coursera/udactiy courses. I have started and drop many of those because they promise that no math requisite and then they straight into the math.
The problem with most courses is that either they are very basic and do not deal with the interesting and important aspects regarding validation, deployment, etc. or they go deep into the mathematical and theoretical aspects.
I think the âtop downâ approach is better for beginners. There is value in learning the âbottom upâ too. I like Jason Brownleeâs machinelearningmastery.com website and books too.
If you are new to ML it helps to have a sort of map. The map is basically supervised, unsupervised, and deep learning. But, I would try to figure out what you want to work on and work on that aspect more directly as well.
Practically, you need to get Jupyter environment up and running and you need to get comfortable with manipulating the data. Most of the work in building ML models or at least a lot of it is actually manipulating the data.
If you have a programming background, I would not try to learn Python like a software engineer either (as a software engineer). Instead create lots of notebooks with ârecipesâ and open them up to pull and extract. First notebooks should be on basic python operations and then build up from there.
wiselly