Help me improve our Data Science instruction! | Coursera Community
Coursera Header

Help me improve our Data Science instruction!

  • 2 December 2018
  • 16 replies
  • 928 views

Userlevel 1
Badge +1
Hi all,

With such positive feedback on our Applied Data Science with Python specialization we've decided to launch a full on Master's degree in applied data science. We're working hard on content, instruction, and assignments, and now would be a great time to help us shape both our degree program as well as additional other open content on the Coursera platform! What would you like to see in the world of applied data science? Thoughts could be topics, teaching strategies, technologies, or outcomes. What have you seen that was awesome, and what did you find that was frustrating?

I welcome any thoughts on how to make our online instruction better! Keep on learning,

Chris

16 replies

Userlevel 7
Badge +13
@Liz – I thought you might like this invitation!
Userlevel 5
Badge +5
@Laura thanks for the tag!

I haven't done the Applied Data Science with Python specialization so I can't offer any feedback on that one.

The best resource I have found on Applied Data Science is Andrew Ng's Machine Learning Yearning, definitely worth a read! It is written in such a way that you do not need to have a mathematical background to follow the material and has very short 1/2 page chapters. This is great because it means I can print off one page and show it to others in my team when trying to explain new ideas to them. It also focuses much more on how to apply the ML rather than the algorithms, which is really important. It's great to say look at all these models I've trained, but if you can't interpret the results correctly what is the point?

I work with a lot of time series data, so if you include any case studies it would be nice to see some forecasting or anomaly detection for time series data. A lot of ML buzz is always about image classification, object detection, etc. So it is easy to find examples of how to solve those types of problems. It would be good if you can include some of the less exciting problems as well, because they are still important even if they are not quite as fun to demo!

I think it is important to mention things like how does ML work in production? What do we do after the model is trained? How to deploy models?
Badge
Thank you so much for the specialization. I really enjoyed the entire specialization a lot. I like the programming assignments. They were in the right amount of difficulties and the discussion forum provides really good community when you have a question. I also like the scientific articles provided during the courses. They are very relevant although some of them are a bit harder to understand.

One thing I would suggest to open a course on Kaggle competitions or anything that deals with real life data. I believe it is a great way for students to see real life data and how to apply knowledge we learned from the specialization to the competition/data. Thank you again for organizing the specialization.
Badge
Hello! I have completed the wonderful specialization Applied Data Science with Python. The reason I can work as a freelancer is this specialization. I refer to this material frequently and I have used most of the knowledge I have acquired.
Now as far as it concerns my knowledge and experience, what the market looks for is:
a) data mining - data scraping - some XML ! still
b) machine learning, mostly classification, less Regression
but
most clients ask for deep learning-neural networks, with keras-tensorflow-flask-Azure, many tasks of image recognition, open cv.
So a more andvanced specialization should contain image processing, deep-learning, use of keras-tensorflow-Azure-flask, some guidance to GPU using, and advanced NLP with deep learning.
Perhaps open cv should be a distinct course. These, are things which I have not learned and I lose many jobs.
Badge
I wish I had taken this course to provide constructive feedback. So far I have taken only one course in Python and really enjoyed it.

I would like to pursue MS degree in Data Science offered by Michigan University if cost and time permits and eventually become full time paid TA or virtual instructor for the students around the world to bridge the gap between students and university and/or professor.
Badge
Hello


I think the MS should have a variety of examples from various fields of study.
Badge
Hello Christopher and thanks for reaching out to the community.

What would you like to see in the world of applied data science?

Free, actionable data insights on issues that apply directly to the community of interest. Content should be relevant to local contexts, yet broad enough to teach the latest data science ideas and methodologies. For me specifically, I am interested in equity in educational opportunities for our young people and the power of data to tell stories.

What have you seen that was awesome

I have been asleep on Coursera since Charles Severance PY4E and Andrew Ng's older Octave Machine Learning Course - both of which I was able to complete assignments for and receive feedback on for free... amazing content.

What did you find that was frustrating?

It has become more difficult to learn without paying. Paywalls sometimes come after you've completed a quiz and realize it won't be graded. When the coding assignments can't be autograded it is a huge bummer.

I wrote an article at the start of 2018 about Data Science learning materials, https://evanrushton.blogspot.com/2018/01/learn-data-science-2018.html

if I were to add anything now it would be the Tidy Tuesday R community. @drob's weekly screencast is absolutely phenomenal. https://www.youtube.com/watch?v=pBGMt28xgvk&feature=push-fr&attr_tag=oC9Eamg2fT2eMfZ5%3A6
Badge
I did another specialization in coursera, it was about Python.
I began to freelancing this year. I wasnt lucky in the beginning, but after time I notice that to code with Python is the best. You can find work from simple data problems to very complex about deep learning. For example, many people asks about scrap data to automatic process for their pages, other for companies, and so on.
All the skills I was able to learn are needed now.
I suggest to include scrapping data, use of API, neural networks, use tensorflow, image recognition, basic php.
Many people use different kind of software, jupyter notebooks are required, in the same way, some coding with Python 2 can be good too. I like Python 3 and Spyder.
Good luck with the Master degree chance, it is the best to implement right now.
Badge
There are many resources related to Python, machine learning, and so on. And it is not a problem to hone these skills yourself. It depends on your motivation. But in all materials / courses on ML used in advance prepared data sets. Because of this, you are not ready for a real project.
In a real life you need to see a business problem, convert it into an ML \ DL problem and prepare a dataset for it. And it's really not easy for anyone who has just completed a course like Applied Data Science / Machine Learning. It is not enough to be prepared for uncertainty in a real project.
Another key point is the interpretability of the results. To make correct inference of obtaining result is really important. And it would be cool if material how to improve this skill is appeared.
And thirdly for a real-life business project, model accuracy is not the goal. Business needs a turnkey solution that can build-in enterprise IT-landscape, and it doesn't matter how it works inside. It's good to make accent on how to embed ML\DL solution in existed of multilayer architecture.
Badge
Hi Mr. Brooks

I am taking the Applied Data Science with Python specialization, I really like it!! But, if you are going to made a Master's degree, I suggest to

1. Add mathematics in the courses, just to understand how it works everything in the practice, or additional to this, add some documentation where to find the mathematics and the process to get the final equation used to transform in a program (Python, or other). With this, the studen will be able to use any programming languaje.

2. Add some research topics, in order to let to know the student if wants to continue to a PhD. I mean, if we looks for research in Data Science over Internet, it is a little difficult to find a topic. I tryied because I want to made a PhD.

3. Use several not commmercial programming languajes, because the data is increasing and some time the commercial package will be necessary to be updated. Moreover, the data changes their format with the time, so the commercial packages maybe will be not adapted faster.

4. Give a UMich certificate, because sometimes the enterprises didn't likes a third party certificate, they mention such certificate says "an online non-credit course authorized by" or something similar. In the other hand, for the formal student, I think is more a matter of pride to have one from UMich, without underestimating the third party provider.

5. Think to prepare also a PhD, if it's possible remote, but with real instructors, face to face interaction by using Skype or other technology, etc. In my case, I would like to made a PhD, but I don't have the resources to move from my country. (For technological items, we can remotely access servers to manage high computing resources, etc.)

Well, regarding the course, the tasks are very challenge (we need to investigate more than the course itself and that give more learning), and that's is good for students, so this will be another recommendation for the Master's degree.

Thks for request our comments, this is another kind of consideration from you.

Regards
Adolfo
Badge
Hello Christopher

I've done 3 courses out of 5 from this specialisation and I surely say this has been the best online course I've taken so far. So, well-done; and hopefully I will finish the specialisation in the next two months.

What I probably would like to see in such courses is more about assignments. I think it's great to have a sequence of assignments which are built on top of each other. This helps students to understand the process of finishing a project and coming to the outcome rather than only solving a problem which might be just a small step of a whole. As an example, if I've learned to obtain and clean up the data in a week and I've passed the assignment successfully, the next topic's assignment which can be an EDA assignment would be great to continue the work on the same data set as the previous assignment. At the end of the program or the course students have completed a work which they can use in their portfolios.

Lower assignment scores in the beginning of courses may end up in failing in final weeks because the whole project is not well handled, but this is exactly what happens in industry. I'm sure you will find a solution for that in a learning system not to demotivate students from continuing. Maybe giving students personal feedback on their assignments after they passed! This helps them to prepare their data set for the next topic.

Thanks again for your perfectly structured course, and for asking for our recommendations.

Regards,
Mehran
Badge
Hi,
Here are some elements of answer to your questions.

What would you like to see in the world of applied data science?
  • Ability to work at scale in an autonomous manner after the course
  • business dimension of the use cases
  • end-to end approach : from the first step of defining a use case to deployment and maintenance
  • courses oriented on families of uses cases (eg behaviour identification, ...) rather than algorithms
  • understanding of the evolution trends, so that we can optimize the tools and languages to learn
What have you seen that was awesome ?
Andrew Nh course
I also enjoy quite a lot the series ML with Tensorflow on Google Cloud mainly for the broad view, the simplicity of the approach and the support we get

What did you find that was frustrating?
I always have many difficulties with features preparation, probably because of a lack of knowledge of Python and graphics.
Fundamentally i would like to use more advanced tools for data preparation and vizualization, which seem to be a trend in the market.
I find frustrating to spend 2 or 3 times the indicated time on a course and to look for information in the forums (very time consuming, with poor search tools).
I would like too to be able to stop and resume later a courser without paying during the interval.

Thank you for helping us !

Regards,

Jean-Luc
Userlevel 1
Badge +1
Hi all,

Thanks for all of the insight! I really appreciate the thoughtful comments, and a lot of these resonate well with what we are doing.

I should emphasize that we are both expanding our free online offerings (or low-cost if you get the certificates) in this space with more MOOCs and refinements of MOOCs, as well as launching the online degree. We're also planning to add more practice content (so you can verify how you are doing for free in the MOOCs) as well as try and make it easy to move from the MOOCs to the online degree. For instance, we just launched a new Python 3 specialization with the intent of helping people get the prerequisites for the online degree.

Thanks again for sharing your thoughts, both on the MOOCs in this space and on the online degree we're designing. Looking forward to launching in 2019!

Chris
Badge
Hi Chris
I have completed only the first course of the specialization so far. It is a great course but I found frustrating the fact that we needed to browse the forums to get some information in order to complete the assignments. In other words, we need to teach ourselves things that are critical for the assignments.
On the other hand, the deep learning specialization with Andrew NG had the right amount of research.
I also, think that there were not enough notes provided, even if I would have taken my own notes anyway.
Those two points were offputting , but the teaching and the structure of the course was definitely awesome
Regards
William
Badge
Each domain area of knowledge where DS is to be applied should attend specialized or advanced courses. So, for example, someone working in marketing will need to understand things like web hits and sessions, pages and screens, how user IDs are managed and the rules of PII.

Prior to this, there should be some level of understanding about the workflow for how data is captured, stored, transformed, analyzed and activated. This could be done generically to be applied to different cloud environments whether Google Cloud, AWS or Azure.
Badge
@Christopher Brooks Just tried the link to the "Master's Degree in Applied Science" included in your post. It appears as if that link is broken, leading to "Page not Found".

Do you happen to have an updated link? Thank you!

Reply