返回到 Machine Learning: Clustering & Retrieval

4.6

1,801 個評分

•

307 個審閱

Case Studies: Finding Similar Documents
A reader is interested in a specific news article and you want to find similar articles to recommend. What is the right notion of similarity? Moreover, what if there are millions of other documents? Each time you want to a retrieve a new document, do you need to search through all other documents? How do you group similar documents together? How do you discover new, emerging topics that the documents cover?
In this third case study, finding similar documents, you will examine similarity-based algorithms for retrieval. In this course, you will also examine structured representations for describing the documents in the corpus, including clustering and mixed membership models, such as latent Dirichlet allocation (LDA). You will implement expectation maximization (EM) to learn the document clusterings, and see how to scale the methods using MapReduce.
Learning Outcomes: By the end of this course, you will be able to:
-Create a document retrieval system using k-nearest neighbors.
-Identify various similarity metrics for text data.
-Reduce computations in k-nearest neighbor search by using KD-trees.
-Produce approximate nearest neighbors using locality sensitive hashing.
-Compare and contrast supervised and unsupervised learning tasks.
-Cluster documents by topic using k-means.
-Describe how to parallelize k-means using MapReduce.
-Examine probabilistic clustering approaches using mixtures models.
-Fit a mixture of Gaussian model using expectation maximization (EM).
-Perform mixed membership modeling using latent Dirichlet allocation (LDA).
-Describe the steps of a Gibbs sampler and how to use its output to draw inferences.
-Compare and contrast initialization techniques for non-convex optimization objectives.
-Implement these techniques in Python....

Aug 25, 2016

excellent material! It would be nice, however, to mention some reading material, books or articles, for those interested in the details and the theories behind the concepts presented in the course.

Jan 17, 2017

Excellent course, well thought out lectures and problem sets. The programming assignments offer an appropriate amount of guidance that allows the students to work through the material on their own.

篩選依據：

創建者 ELINGUI P U

•Aug 20, 2016

Great course like the others

創建者 Yin X

•Nov 04, 2017

I really like the content of this course, like other courses in this specialization. However, for the assignment in module 5, one must work with GraphLab to get the correct answers in the purpose of getting a certificate. I think it is not very convenient for those who may have trouble accessing graph lab. I wonder if the instructors could provide a pandas/scikit learn version for assignment 2 in module 5. Thanks again for putting together such a great specialization.

創建者 Dony A

•Jan 05, 2017

awesome clustering course

創建者 Maria V

•Aug 02, 2016

The specialization has a good quality on average. I started doing this course immediately after it went open. I had a feeling that the quality of the course went down (questions were often unclear and it took time to figure out what is expected as an answer). However, many problems were solved quite fast and teaching stuff is really helpful.

I still would like to see more about MapReduce in-depth in this course. I did not have a feeling that it was covered sufficiently (only theory, no hands-on material). In general, hands-on material was great and useful.

創建者 Iurii S

•Nov 26, 2017

Good course overall.

Starting to get more on the side of being mostly implemented and only needing to insert a line or two.

創建者 Kartoffel

•Jul 26, 2016

Great course. Some week were tough others too easy, but general a very interesting course.

創建者 charan S

•Jul 30, 2017

Nice intuitive course with lots of understanding.

創建者 Sundar J D

•Sep 26, 2016

Great course and awesome teaching by Prof. Emily Fox. Prof. Fox did a great job of teaching some of the really tough components (GMM, LDA, etc) in simple and lucid style (like always) and that made it easy to understand and comprehend those topics.

The one thing that I felt had gone down compared to the previous 3 courses was that for some of the topics, the material felt too short and felt like it was cut down to fit within the 6 weeks course duration. I would have at least liked some extra reading material or references especially for GMMs, LDA, Gibbs Sampling, etc.

創建者 Adwait B

•Jan 26, 2018

Great Course! Tough topics well taught

創建者 Steve S

•Aug 26, 2016

Like all the courses in this specialization so far, the material has been good. The reason for only 4 stars rather than 5 is the difficulty in getting questions answered in a timely manner. There don't seem to be any active mentors for this class.

創建者 Sander v d O

•Oct 18, 2016

All the courses in this specialization are great, but compared to the other 3 i did until now, this one seemed a bit short on material. Especially week 1, and somewhat week 6 was without good material. Weeks 2, 3 and 4 were great. I got lost somewhere in week 5 on collapsed Gibbs sampling.

Still: very much recommend this course, it provides a good introduction to Nearest Neighors, K-Means, Gaussian Mixtures and LDA. Thx prof. Fox!!

創建者 Pier L L

•Aug 02, 2016

Very good course nice practical approach. I was kind of surprised that hierarchical clustering was kept at the end and discussed only marginally since it is a widely used approach.

I liked the part about LDA but IMHO I would have liked more discussion about fundamental techniques rather than such an advanced method.

Too focus on text data. Most of the application I worked on have limited textual data.

創建者 Keith D

•Jun 19, 2017

I'm disappointed that courses 5 and 6 of the specialization were cancelled. The cancelled capstone was why I purchased this specialization package.

創建者 Asifur R M

•Mar 19, 2017

For me, this was the toughest of the first four courses in this specialization (now that the last two are cancelled, these are the only four courses in the specialization). I'm satisfied with what I gained in the process of completing these four courses. While I've forgotten most of the details, especially those in the earlier courses, I now have a clearer picture of the lay of the land and am reasonably confident that I can use some of these concepts in my work. I also recognize that learning of this kind is a life-long process. My plan next is to go through [https://www.amazon.com/Introduction-Statistical-Learning-Applications-Statistics/dp/1461471370], which, based on my reading of the first chapter, promises to be an excellent way to review and clarify the concepts taught in these courses.

What I liked most about the courses in this specializations are: good use of visualization to explain challenging concepts and use of programming exercises to connect abstract discussions with real-world data. What I'd have liked to have more of is exercises that serve as building blocks -- these are short and simple exercises (can be programming or otherwise) that progressively build one's understanding of a concept before tackling real-world data problems. edX does a good job in this respect.

My greatest difficulty was in keeping the matrix notations straight. I don't have any linear algebra background beyond some matrix mathematics at the high school level. That hasn't been much of a problem in the earlier three courses, but in this one I really started to feel the need to gain some fluency in linear algebra. [There's an excellent course on the subject at edX: https://courses.edx.org/courses/course-v1%3AUTAustinX%2BUT.5.05x%2B1T2017/ and I'm currently working through it.]

Regardless of what various machine learning course mention as prerequisites, I think students would benefit from first developing a strong foundation in programming (in this case Python), calculus, probability, and linear algebra. That doesn't mean one needs to know these subjects at an advanced level (of course, the more the better), but rather that the foundational concepts are absolutely clear. I'm hoping this course at Coursera would be helpful in this regard: https://www.coursera.org/learn/datasciencemathskills/

創建者 Kostyantyn B

•Nov 07, 2017

A high quality, intermediate difficulty level course. The instructors are obviously very knowledgeable in this field and strive to pass their knowledge and skills onto the students. One of the major advantages in my opinion, is the fact that the authors decided to include a number of advanced topics, which you normally don't find in an introductory level course on the Unsupervised Learning. The exercises seem to revolve mainly around the Natural Language Processing, which is fine by me, for two reasons. First, it is a very challenging part of the Machine Learning. Second, NLP is in high demand in the industry. So, I see no downsides here. Plus, there is only so much one can squeeze in a 6-week course...

I would however like to mention that I wasn't entirely happy with the way the Latent Dirichlet Allocation and the Gibbs Sampling were explained. This was the first time I heard about these techniques and I found them fascinating. I understand that these are challenging topics that require a more advanced math for a serious discussion. But I still think it would be worth including perhaps an optional video and/or exercise to go deeper into this subject. I am sure some students would appreciate it; I know I would...

In summary, it is a great course to take. It will help you better understand the theoretical foundations and boost your practical skills in the Unsupervised Learning.

創建者 Michael B

•Sep 04, 2016

Good survey of the material, but assignments are superficial and don't test thorough understanding.

創建者 Matt S

•Nov 10, 2016

I found that Week 4, Assignment 2 was testing our knowledge in ways that are in opposition to the general ethos of the course.

I mean that this course is about gaining insight, intuition and the practical tools for ML, where the smaller details like - knowing which numpy function gives you a random univariate normal distribution - are normally provided for us, so that we can focus on the aforementioned broader, more useful aspects of ML/Clustering.

The assignment had good content on the whole but the parts which were chosen for "write code here" could certainly be improved.

I hope this is useful feedback and that the assignment is reviewed so that it doesn't needlessly discourage people.

創建者 Michele P

•Sep 02, 2017

Advanced course. The material taught in this course is more advanced compared to Regression and Classification courses. You have to invest more time in respect to the previous courses. For some topics (LDA and hierarchical clustering) I had to look for other sources in order to understand the concepts properly. However, this course is a good introduction to clustering and retrieval.

創建者 Maxence L

•Dec 15, 2016

Comme les précédents dans de cette spécialisation, ce cours est très riche et donne les clés pour utiliser des outils complexes et puissants. Toutefois, un peu plus de détails sur certains aspects, notamment théoriques, pourraient améliorer la compréhension de certains chapitres plus techniques.

創建者 Jack B

•Mar 04, 2017

Should use pandas instead of Graph Lab Create

創建者 Mehul P

•Sep 11, 2017

Nice explanation on clustering methods.

創建者 Galen S

•May 08, 2017

I liked the slides.

創建者 Abhishek S

•Feb 10, 2018

Till Expectation Maximization, the learning is tremendous. However, once past that, everything would feel incomplete since most assignments are spoon fed after that. Rating it four stars because of initial lectures.

創建者 Bob v d H

•Oct 02, 2016

Some of the interesting topics discussed in this course could be treated substantially more extensive and detailed in order to get a better grip and understanding on them (e.g. Gibbs sampling). After this course, it is a bit dazzling how much different algorithms and methods are available for clustering and retrieval tasks and this course easily could have been subdivided into two or three separate courses on the same topic with a more detailed treatment. Still, about many interesting subjects a tip of the iceberg has been brought to you ... it tastes so good that you would like to have much more!

創建者 Marcin W

•Aug 09, 2016

Very good course. Too long interval between modules make hard for non-Python developers. Easy to forget some of the Python structures.