Chevron Left
Retour à Statistiques bayésiennes

Statistiques bayésiennes, Université Duke

502 notes
151 avis

À propos de ce cours

This course describes Bayesian statistics, in which one's inferences about parameters or hypotheses are updated as evidence accumulates. You will learn to use Bayes’ rule to transform prior probabilities into posterior probabilities, and be introduced to the underlying theory and perspective of the Bayesian paradigm. The course will apply Bayesian methods to several practical problems, to show end-to-end Bayesian analyses that move from framing the question to building models to eliciting prior probabilities to implementing in R (free statistical software) the final posterior distribution. Additionally, the course will introduce credible regions, Bayesian comparisons of means and proportions, Bayesian regression and inference using multiple models, and discussion of Bayesian prediction. We assume learners in this course have background knowledge equivalent to what is covered in the earlier three courses in this specialization: "Introduction to Probability and Data," "Inferential Statistics," and "Linear Regression and Modeling."...

Meilleurs avis

par RR

Sep 21, 2017

Great course. Difficult to apprehend sometimes as the Frequentist paradigm is learned first but once you get it, it is really amazing to see the believe update in action with data.

par GH

Apr 10, 2018

I like this course a lot. Explanations are clear and much of the (unnecessarily heavyweight) maths is glossed over. I particularly liked the sections on Bayesian model selection.

Filtrer par :

145 avis

par De'Varus May

Feb 15, 2019

Though this section in the specialization is a little more difficult than the other sections. The supplemental material provided is helpful in navigating through the course. I will continue to read through this material to further my understanding of the material.

par Toan Thien Le

Jan 26, 2019

Good for reviewing Bayesian Statistic. But not for new learners.

The quality is below the previous courses in the same Specialization. The contents are rushed. The labs are impractical and sometimes confusing.

And beware of the final assignment. Since the number of students is low, the grading takes lots of days. And you might miss the enrollment window for the Capstone course.

par Richard Millington

Jan 24, 2019

While the other modules so far have been terrific with good levels of support and clear explanations, this module is pretty terrible for a few reasons.

1) The level of support.

Your chances of getting a response to any question are slim - which means you're pretty much on your own here. Don't understand anything? Go find the answer elsewhere.

2) The tutors.

Mine Çetinkaya-Rundel has generally been terrific so far. Speaks slowly, repeats what variours terms mean (instead of assuming we memorize them the moment we hear them) and provides good clear examples to work from.

Sadly both Merlise and David are the opposite. They whiz through the material uncomfortably reading from a telepromter often assuming we instantly grasp every possible concept. It's almost impossible to follow most of the sessions they present. Most of the time there aren't even any exercises or opportunities to check we've understood the material correctly. They would both be 100% better if they frequently reminded us of the definitions of the concepts they use.

3) The material. There is FAR too much here to be covered in a single module. This is an entire course on its own (or a much bigger module).

4) Assumptions we know things which are never taught. I've lost track the number of times a word or concept sneaks into a quiz, into a lecture, or into an R package without explaining what it means. At times it feels this material was pulled from 2 or more sources and this has created gaps in understanding.

Sorry guys, I've really enjoyed the first three modules...but this one was a bit of a disaster.

Provide better support, shrink the material, create a better lecture experience and I'll happily revise this.

par Liew Hoe Peng

Jan 17, 2019

This course is challenging and well-presented!

par Sara Melvin

Dec 24, 2018

Starts out good in the first week and then ramps up to graduate level statistics without really a lot of notation explanation. Week 3 with the silver haired lady as the teacher was the WORST. nothing made sense when she taught.

par Pedro Guilherme Frade Moro

Dec 20, 2018

great course!

par Tulio Rodrigues Carreira

Dec 11, 2018

The last two weeks are way too hard to follow and could provide more practical examples instead of focusing on mathematical theory and formulas. That would make more sense to this course when compared to the content of the previous ones in this specialization.

par Wei Chun Chang

Dec 06, 2018

The materials and response from the organization are unavailable for a while and never get an answer

par Tansel Tanner Arif

Dec 05, 2018

Unfortunately, for me, this course did not live up to my expectations in comparison to the previous 3 courses I took as part of the Statistics with R specialisation. I gave the previous 3 a full 5 stars each.

The problems I had with this course was not that my statistics knowledge was lacking or that I found it difficult. The problems were due to the robotic delivery of the material. Specially towards the end of the course. It is very clear that the instructors have a great depth of knowledge which is incompatible with the robotic delivery structure currently in place.

For example, if you use a particular technique, even if it was introduced earlier, all it takes is a brief 2-10 second statement to re-iterate. This encourages the delivery of the material to be a hybrid of both written text the instructors are reading from, as well as a more informal aura of discussion. A guideline is: 'Can you get someone off the street to read the material you wrote to the screen?'. The more this statement is false, the more amazing your course is.

Another issue I had was that the accompanying material was immense. Am I paying a subscription to read books and passages in order to understand the material? This point is also prevalent in the forums where it was raised multiple times. These books and supplementary material would be largely not required if simple commentary was in place in the videos.

E.g. We are applying a 'BIC' prior. What does this mean? Up to this point we are used to applying priors that are distributions. This means that we are approximating the posterior probabilities of the models using the maximisation of their log-likelihoods which turns out to be easier to calculate than the posterior distributions. However, as the model space grows (>25 parameters), we may need to rely on a sampling technique, these techniques which rely on posterior probabilities to traverse the model space. If I were to say this, it would take me 10 seconds but would provide so much information to the learner.

In summary, I could have read a lot of the presented material here from a text book and found it clearer which wasn't the case in the previous 3 courses. The resources were helpful and focused on interesting points. I loved the interviews at the end with experts in the field. The statsR package is great and this is a great way to showcase its capabilities. This course is OK but I think the delivery could be improved upon.


par Zhen XIE

Nov 26, 2018

Provide bunches of intuition of bayesian statistics. Worthwhile to enroll!