Opiniones y comentarios de aprendices correspondientes a Introduction to Data Science in Python por parte de Universidad de Míchigan

good

太难了8

HARD

good

ok

ok

A

Truly horrible delivery of the material - even worse than Coursera's old Intro to Machine Learning course from Univ of Washington. This course will discourage nearly anyone from pursuing Data Science.

And it's not even an intro to data science. It's a course on Pandas for dataset manipulation. (In fairness, cleaning up ingest data is like 95% of the work in data science, but the course doesn't even tease the student with some exciting machine learning examples of where this is all headed.)

It's not delivered like you'd expect an intro course. It does an awful job of progressing the student through the Pandas toolset, building concepts incrementally. The whole topic of object types, methods, returned objects, and chaining gets barely a mention, but it's essential to the assignments. Examples are rapid-fire and sparse - very few techniques needed in the assignments can be found in the examples. The Week 2 quiz tests on techniques not introduced until Week 3, and the Week 3 and 4 assignments cite "individual study" which is academic-speak for "We didn't teach you about this - go Google it".

Then, there are errata that the student needs to pick out of the discussion forums to pass the assignments because some key questions are vague. The errata are 1-2 years old and they can't be bothered to correct errors.

The auto-grader could be the highlight of the course, but it provides limited feedback on wrong answers and no guidance toward the right answer; just "wrong". You're not allowed to post code or discuss answers in the forum - you have to go to StackOverflow to do that. (It'd be awesome if several of the exercises provided the student with the answer and challenged them to match it, but instead it's very sink-or-swim.)

Even when your answer is right, the auto-grader throws errors and warnings for, say, returning a numpy.float64 (which you should) when the grader is expecting a Python float type. Or it's expecting a float64 for a counter value (!!) when you provide an int64 (which is correct). These behaviors should have been fixed long ago.

It claimed to be a 15-hour course; I did it intensively and invested more than 30 hours before pulling the plug on the final project. That was claimed to be a 4-hour project, but experience with the rest of the course says it'd be more like another 12 hours - and that's for a guy who's not new to coding.

Bottom-line: I paid for educational material and I don't feel like this course delivers. What it does deliver is Pandas exercises and an "OK" auto-grader; truthfully, most of what I learned was via Google searches while trying to do the assignments - effective, but very slow and very frustrating. The real disappointment is seeing that the issues I encountered have been well-known for 2 years in the discussion forums; the course could be a lot better by now if they cared to nurture it.

Finally, a frustrating aside that's on Coursera, not the instructors... Coursera's online Jupyter notebook platform is really unstable and constantly drops connections even when you're actively editing and executing cells. (Including from 2 Fortune 100 companies - it's not the connection.) Once dropped, the notebook can't be re-connected, and has to be re-launched from the syllabus at the risk of losing your most recent edits. (But beware, if you run Jupyter offline for stability, this course also has defective input filenames that will cause grading to fail - read the discussion forums first.)

More or less my copy from an earlier review,

I was really excited about the this course, and was really let down. This course is really, really poorly done. I would not waste time and money on this course when there are much better options out there. I feel like I've gotten little in return for my time and money.

First, as several other students have noted, the timeframe for assignments is really unrealistic, taking much longer than projected (at least for me, and several other students). This is not acceptable when Coursera bills by the month. Coursera needs to provide a better assessment of the time commitments for the class. I took another datasciense course prior to this one (my employer wants a certificate) but still the assignments were tough, and I found it really dissappointing that I spend a lot of time solving inconsistencies in the assignments. I believe American students are in advantage here because of the Geo-American orientated datasets.

Second, the teaching is horrific. The professor is not engaging at all, but simply mechanically reads lines which often sound straight out of a user manual. The point of online videos is not to turn books into audio files- it’s to have a human talk/reason through problems with you. The teacher of the course should discuss the material, not recite a manual. In addition, the little amount of material is presented far too quickly, Also great emphasis is put on the discussion groups (which turns out to be just responded by the moderators, volunteers). In absence of a proper syllabus students are directed to Stack Overflow, a sign of the courses' weakness.

Third, the title of this course is a misnomer: an introduction to data science would provide an overview of the tools, techniques and scope of the field. An extremely detailed introduction to Pandas, which is essentially what most of this course is, is useful if well executed (which it is not here), but it is not an introduction to data science.

A more minor complaint is the absolutely horrendous choice of the background. Showing different permutations of lifeless office drones is not exactly inspiring material for aspiring data scientists, even if this the reality of office life- it’s distracting at best, and at worst, deeply disparaging. Why not have just a plain colored background? Or anything else?

The only positive thing besides some of the misleading assignemnts are the rest of the assignments. In general I had fun solving them, and althoug I've had my share of Jupyter Notebook and Grader's issues I was able to complete the course. I will not reconsider any online course from Michigan University again.

Issues:

1. The lectures are a chore to sit through. Dry, slow, and unorganized. The lecture on pivot tables was not used in any assignment. So why have it?

2. Assignments. Ooof. Lets break this down.

a. Autograder is poor. Aside from the oddities that just break it sometimes, the hidden test feedback is lacking. If my answer is off at the 15th decimal point because my dataframe is 226 rows and not 227. I need another assert and feedback telling me that. Instead of a lesson on logic a lot of these problems became frustrating cases of github searching for other peoples passing code and then working backwards

b. Assignments felt rushed. Each assignment had poorly written questions that frequently popped up in the forums asking for clarification. The assignments themselves had odd jumps in difficulty and assumptions. Some would build upon the lectures but other times they would jump and assume that we would figure out the middle. Oftentimes we did, but imperfectly, and the assignments penalized us for that imperfection. For example, if question says clean the data and we do using one of a dozen different ways why are we penalize if we have 224 clean rows, and the answer requires 227. If that level of end accuracy is required, then we need more guidance to achieve exactly that.

3. Forums. Useless. Filled with garbage, and the useful ones are unstructured mess. For one, the autograders output is small grey typwriter text which is undecipherable and the TAs always wanted it posted. This lead to long chains of code blocks and one line responses. I also think the TAs emphasis on posting zero code is wrong. The entire web is built on Stack Overflow, so why not allow code snippets in the forum?

It pains me greatly to give just 2 stars to a course from UofM, since it is my alma mater, but I will be honest. I would like to echo the sentiment of the majority of my fellow learners that the course needs to be structured better. Instructor needs to take more time to explain some of the concepts in greater detail. It seems like the instructor and his assistants are always trying to rush things and cover too much material in tool little time . I had to pause and replay lecture videos to completely grasp what was being conveyed. I also adjusted my playback speed to 0.75x to keep up with the instructions. I will admit that I had to heavily rely on the pseudo codes posted on the forums to answer assignment questions and even though I answered them correctly, I did not completely grasp the reasoning behind lot of them, which I think defeats the purpose of learning a programming based course.

Suggestions for improvement: Upgrade the autograder, because it is frustrating to keep rectifying the answers to make them acceptable for the autograder. Completely overhaul the assignments so that they are more in-line with what is being taught in the lectures. Students should not have to figure out everything from the online forums. If not for the pseudo-codes, algorithms and explanations from mentors, this course would have been an impossible one to finish. Assignments and exams need to be designed such that learners don't have to treat forums and stack overflow as a primary vehicle for getting successful with the course, but more like a helper tool.

The course content is very good. The videos are very good. Unfortunately this course is severely hurt by a very high ratio of non-learning work to learning work. This is due to some issues that could be easily addressed. The questions are poorly worded or ambiguous about critical details. Some of these details are hidden in the forums, but that's a waste of time. Some of the assignments do not directly bear on the course content and involves much "self-learning". Unfortunately this means I do not know if my self-taught methods are optimal - there is no feed back or checking. So you can do very poor coding but still pass in scoring and never get any feedback to improve your coding skills. All along, some very simple hints about what libraries and methods to use for each question would prevent lots of blind searching on the web. There are some helpful instructors and helpers haunting the forums, but they are not always around, and they are not always implementing permanent fixes to the problems that are frustrating students. One shouldn't have to hunt around forums to find out about broken pieces of the application or other errors in the course. Finally, the grading system is unstable and the Jupyter Notebook system is also not very stable, leading to many submissions and resubmissions just to make sure it got through for grading. For these reasons it took much more time than three weeks for me personally. I would not have signed up had I known.

While my Python chops definitely improved as a results of the course, the homework was extremely frustrating. The requirements of the questions are not communicated in a consistently clear way.

What was more irritating, though, was that the auto-grader is extremely picky. There is very little room to solve the problem in your own way, and more of my time was spent trying to contort my code to fit what the auto-grader wanted than spent actually solving the problem and applying the course concepts.

I also was disappointed with how much we were expected to manually clean data. One of the questions even explicitly says that the answer will require students to "hand-code" the answer. This strikes me as an extremely poor habit to instill in students--combing through data manually to strong-arm the data into the formatting conventions won't cut it when tackling a dataset that is millions of lines long. For a computer science course, this is not a scientific, or even a programmatic, approach to solving problems.

I give the course 2 stars because I felt that only 40% of what I learned was data science and/or Python. The other 60% of what I learned was how to smash my code until it conformed to the auto-grader, how to bother the TAs in the forum when it wasn't obvious how to do so, and how to write translation dictionaries with the "wrong" format as the key and the "right" format as the value and then apply it to DataFrames.

The topics covered in the course are certainly important for anyone interested in using Python for Data Science, but sadly, there is only really basic information about each topic taught in the videos. In the labs, there is heavy focus on "self learning" (basically, the instructor encourages you to use StackOverflow and other resources to figure out things that were intentionally not taught in the course).

While it's interesting that the problems have a very real-world nature to them (including searching the web to help you find answers to things), if I wanted to learn the tools taught in this course that way, I wouldn't have enrolled in this, and just got a good book and practiced myself independently.

Also puzzling is that there are weird "discussion" segments that have no relevance to any of the topics taught. One moment you are learning about pandas dataframes, and the next, it's asking you for a 90 word opinion on data science ethics. This is somewhat ironic given the otherwise very practical/applied nature of the course. Not relevant at all to a working professional.

One other note: as I write this, for the past few days, no students were able to get their assignments graded, since the auto-grader was broken, and there are multiple reports of this in the forums. It's puzzling the staff let this problem persist for days with no ETA or acknowledgement of the issue.

Overall, this course is tough for me. In my opinion, the course title is not suitable, which is "introduction". The word "introduction" generally means something basic and fundamental. However, the level programming assignments in this course is between slightly medium and very hard.

Furthermore, the assignment instructions are not clear enough, i.e. what steps need to be followed to complete each question. So, it needs time for me to find the way or solution in the discussion forum. I know in each week forum, there is one discussion specifically for assignment (i.e. Assignment X Tips), which is good. But, I think the forum is not a good place, since we have to find it meticulously the information and what we want to find is spread out, not in an easy to follow order.

Moreover, I do not think the lectures and the assignments are well synchronized as there are many things in assignments that are outside from the lectures. I mean, it is "too many" new things in the assignment. Although it is a great way for the students to search and enhance their skills by themselves, in general, the assignments should at least reflect the lectures.

Lastly, in my humble opinion, this course is not appropriate for someone who is new to programming and has not been familiar or mastered any programming languages, though the title is "introduction".

This is a pretty awful course, as of the time of writing this review in July of 2019. Let me preface this by saying that the material you learn is very helpful. Pandas is a great library to learn for loading, cleaning, and manipulating large amounts of data. But the real problem with this course isn't the material, it's the lectures and the autograder. The lectures are very short. They don't cover the concepts well enough, and some material is blatantly skipped and you have to learn it yourself through google. Then comes the video quizzes which test you on functions and concepts that haven't even been introduced yet. It's like the quizzes were put at timestamps randomly. Then comes the worst part of the course: the Autograder. I'm not sure how old this course is but the autograder is running on an outdated version of both python and pandas. What does this mean for you? Well if you want to code on your computer instead of the course's broken online coding notebook, you will run into severe code-breaking bugs between versions. It really ruins the course. I learn the material but then spend hours trying to please the broken autograder. Most of the time in this course isn't spent learning, it's spent fixing code that the autograder rejects even though it runs perfectly on your machine locally. Have fun!

When ones motivation for taking a course switches from learning as much as possible to wanting to finish the course in order to leave a review warning others not to take the course until the numerous structural issues with the course are resolved then something has gone very wrong. The course materials are okay for an intermediate course. Just 'okay'. Not good. Not great. Certainly a substantial step down from the wonderful 'Python for Everybody' courses which, by the way, are inadequate preparation for this course despite the Dr. Brooks' claim. That said, the main issue with this course lies in its incredibly vague and poorly thought out assignments. If you are actually decent with Python you will, in all likelihood, spend more time fighting with the Jupyter notebooks and auto-grader than you will actually completing the assignments. If you're newer to programming expect to spend at least five times as much time on each assignment as the estimated completion time suggests. Also, good luck if you actually need help, as this course has the most aggressive enforcement of Coursera's honor code that I have seen on this site which means that you are SOL if you need help on a problem. In short, I would recommend that no one take this course until the numerous issues with it are provably fixed.

Thanks for the course. A few things that can be improved:

1- The video material was very short. I expected same amount of teaching like homework, but it's more like 30 minutes versus 7 hours every week. As a result it's mostly googling and copy-pasting code, which I'm very sceptical if solving issues this way will enter my long-term memory. Probably next time I will need to look it up again. I'd prefer if the videos were the source of knowledge instead of stack overflow and the forum.

2 - As an experienced programmer but being new to Python I found it difficult to load the data. I was not aware that the files are available on the server and can just be loaded by read_csv(filename.csv). Instead I tried to load the files from my local drive, which worked, but not for the grader. Then I tried to submit it offline which also failed. I wasted half a day on this. I suggest to mention quickly how the online and offline assignments work, in particular in how to load data.

3 - The feedback from the grader is usually not telling much. It was often unclear to me what was the expected outcome. I think a screenshot of the expected answer for the more advanced questions (first few rows+columns) would help a lot and save us a lot of suffering.

Four year old instructional content. Four year old versions of pandas and other libraries. $50/month for 4 year old content??

The course relied on very basic functions and libraries in pandas and numpy. I doubt that any of the specific skills and content taught here would transition very well to a professional or academic context.

Why is it so hard to find a real practical Python data science course?

I'm also pretty sure there were errors some of in-video quizzes. And there exists a broken link to Chris Anderson's Wired Article entitled "The End of Theory..." that has been broken for the past 3 years at least when I checked on discussions for the article.

This course in general has sat here for 3-4 years seemingly unmaintained (see the broken quizzes and links above) and unchanged (see the 4 year old videos and libraries above) just yearning money for Coursera and UMich with little to no evidence of improvements or basic maintenance. I think it's shameful that I am being charged $50/month for access to these materials and the grading system which quite frankly has stymied and stunted the growth and improvements to the course material overall.

Content is actually very good, I can feel the content creator and staffs emphasis on real world problems. Projects and Assignments are quite useful, and I can expect to use skills I acquired in my day to day work. However, the most agonizing parts of this course is it requires tremendous of time to do self-study. I spend tons of time on google search, reading docs and go over other people's post on various forums in order to find the right way(sometimes optimal way) to finish the assignments. I understand and accept that self-study is very import to master things nowadays, but I really think gap is too big between what was taught in class and what you need to complete this course, especially for people that has zero data science programming experience like me.

I'd rather spend time watching 2 more hours of teaching videos per week that can cover more aspects/topics/tips/tricks than go over lots of docs and posts. I won't deny I learned a lot through this course, but I believe my learning curve can be flatten significantly if there is more class materials available. If the majority of course is spent on googling, what's the point to take it?

I wasted hours learning outdated code. I read on the course forums that the course is soon to be updated (July 2020?), but the course I did (May-June 2020) was developed in 2016, and quite a bit of the code that was taught was out of date. These courses should be reviewed for currency at least every year, and the failure to do so is reflective of poor quality control and also not the highest regard for students' time. My other criticisms are that the pace of the course was too fast - too much was covered in four weeks - and that the assignments made me rely too much on teaching myself. Some teaching of oneself is fine - we learn how to teach ourselves - but in this course it was just too much. I spent far more time on the course than the course overview suggested. Having said all this, I did learn quite a bit during the course, and Professor Brooks seems like a genuinely nice guy. As well as that, the readings on topics to do with data science were really interesting. So - overall, a frustrating course that teaches a decent amount inefficiently.

Why are these people in a hurry???

I had to reduce the video speed to 0.75x to get through all the material, because they do not give you time to process anything. There is a video of a guy 'teaching' the numpy module where he spits out 185 numpy functions in less than 8 minutes, and that means 1 FUNCTION EVERY 2.5 SECONDS!! Who in the world can learn like this? I can only assume they are making these videos to show off how much they know instead of actually teaching. They should have given numpy a whole week instead of just one hasty video. And I feel that way about many of the other python modules they cover.

The projects are indeed challenging and that is fine. However, they are often putting questions on the graded assignments for things they didn't actually cover yet. They show you how to do A and then expect you to do ABCDEF on the graded assignments. Furthermore, the assignment questions are just badly worded, creating a lot of confusion and unnecessary stress (as seen in the course discussion forums).

This course needs to be more user-friendly. I observed that video explanations way go too fast when they are typing code at the same time, so I felt like I had to simultaneously digest the explanation while also learning the code syntax, that is not friendly at all. I found myself many time re-watching and pausing the videos so that I could understand each step of the code. The course content is good, but I would compare this course to one of those classes were the content is really good but is not properly conveyed to the audience, specially when the audience is people with no previous experience with Pandas. Also, I get the point of going to Stack Overflow and consulting the Pandas the documentation to complete each weekly assignment, but what is the point on doing all that when I am taking a paid class precisely to be explained those concepts, I do not get it, if that is the philosophy behind the course, then it is flawed, it is just better to tell students ''go learn Pandas on your own".

The course gives practice in Python, but programming assignments were loaded with "busy work" that added little value and often had only loose ties to the material discussed in class. While this forces more independent learning, it also makes it so that this course isn't really much better than free online tutorials. I found myself learning almost everything for the programming assignments from the pandas package documentation and from Google searches that led to stackoverflow. For an instructor from the University of Michigan whose bio says "I work with colleagues to design tools to better the teaching and learning experience in higher education," I was expecting a lot more value for the time spent in this class. I suspect that this course is being used to provide data for furthering his own research. It's often more difficult to maximize value to customers when personal incentives may not be perfectly aligned with customer (student) objectives.

This is not an organized course by any means. The lectures do not cover much and assignments are incredibly tough to the point that each requires more than a week by themselves.

For context, I have completed the Python Specialization by Dr.Chuck (Which was incredible!) and have an Fulltime-Masters in Engineering. One of my main motivation for taking this course is to learn all the topics in well structured manner but I don't understand what Dr. Brooks trying to do with his bizarre way of 'teaching'. Unfortunately, Dr. Brooks just points to StackOverFlow for everything rather than explaining the concepts by himself.

Currently, I am just learning through various Pandas Lectures in youtube and attempting the assignment (which is harder than most assignments I have completed in my college!!). Unless you have great motivation to learn by yourself, this course would be very hard to complete.