Practical Machine Learning

This course sucks. This is about machine learning. not about students learning. students don't learn anything with this course. apart from typing a one-liner code and pressing return.

This was supposed to be the last course of their data analytics specialization program. There is very little mathematical explanations, proofs, and the whole thing lacks a lot of rigour.

Worse, the data in the final data assignment has a lot of flaws, the provided testing data was actually (almost) a subset of the training data. Some people also figured out there were mistakes in the protocol leading to the data, with some instruments not being reset, and identifying indirectly the subjects. etc...

in one word, the data was garbage. garbage in garbage out.

But, when correcting other people's work for peer-grading, of course, I found out almost NO student was able to figure out that by himself, not having done enough exploratory data exploration and not expressing enough judgment. That was not so difficult in this case, but students wrote a few line of codes and were happy with it.

I don't think 99% of people are stupid. it just reflects the fact that the material was very badly taught, and despite messages on the forum, there was no one on the staff willing to deal with the problem in time, not even a word from them, whereas people actually PAID money to follow this course.

the lectures are a bit messy. in 4 weeks they want to present a lot and lot of algos but with so little time, students get only a glimpse of the ideas behind it, and finally all they learn is to type a code in R like

train( , method="xxx") and know the available methods "xxx" available.

and just print the results, without learning to interpret anything or make sense of anything.

I'd say go away from JH courses, there are much better courses around. if they're not available, you can still try them for free. but be aware that you'll have only a very superficial and "practical" only

This is the second-to-last course in the Data Science specialization from Johns Hopkins, and the final of three courses covering actual data analysis techniques (preceded by Statistical Inference and Regression Models).

This was one of the better courses in the series, and I thought it lived up to it's name. This was certainly a practical overview of machine learning techniques. There was very little discussion of the algorithms behind these techniques, certainly much less than even in Andrew Ng's Coursera course, which is itself supposedly fairly watered-down compared to many university courses on the subject.

I was taking Analytics Edge on EdX at the same time as this (and still am, actually, at the time of this writing), and I found them to be fairly similar in depth in areas where they overlapped (e.g. random forests). PML also covered boosting, bagging, and regularized regression among other things. The final project I thought was fairly easy, and my random forest-based model correctly predicted all 20 test cases on the first try.

Overall, four stars. This is a nice overview course if your goal is to understand a bit about how to implement machine learning procedures in R. You won't gain much of a deep understanding of these techniques from this course, but it's enough to get you started.

I found this course very valuable. It isn't realistic to expect to become an expert in machine learning in the 4-5 days you might spend on studying the materials and, if you do, you'll be disappointed. However, it is a pretty good practical introductory course for those, like me, who start off knowing nothing about the subject.

If you're not too clear what machine learning is, what problems it can solve, and what the principles and procedures involved are, you'll find out in week 1. Weeks 2 & 3 are practical - you end up being able to use powerful tools like random forests for prediction. Week 4 is light-weight, as throughout the specialisation, but provides a quick overview of important specialised sub-fields like forecasting and unsupervised learning.

There are some glitches in the materials and I got the wrong end of the stick a few times and became very confused because important ideas hadn't been flagged up enough for me to spot their significance. However, I don't recognise the problems with the assessment that one of these reviews complains about.

Don't underestimate the background that you'll need. There are no revision classes or recaps here. You'll have to have fully mastered the material in the earlier modules in the specialisation to survive, especially the statistics, but also cleaning data, exploring data, producing reproducible reports and putting them on github and, of course, R.