Unsupervised Machine Learning Hidden Markov Models in Python (Udemy.com)

As a someone who did not have a chance to encounter with HMMs earlier, this course was a huge surprise. The way how instructor approached to the topic and explanations is for huge praise. Theory is important to understand in order to try do some practical stuffs. This course and instructor will have impact on my further research in the field of HMM implementation from both theoretical and practical aspect. Patience is also important in succeeding this course alongside good understanding. It was really above expectations and after this learning curve will definitely still be on upward path.

Hidden Markov Models introductory sections are good, the core Hidden Markov Model Discrete Observation section is just excellent. The rest of the course unfortunately doesn't have the same integrity and depth. Sections are briefly presented and not easy to understand without extensive and time consuming Internet research. TensorFlow 1.0 and Theano code looks a bit heavy-weighted. The gmm/hmm API libraries like scikit-learn, hmmlearn etc. are left ouf of the course scope. I studied them in parallel on my own and found it very useful. The minus to mention is that the author doesn't provide course slides for download as reference, at least I didn't find them. In general 4 stars.

Good

Thank you for the course. However I find some of the contents are very cryptic in nature inside the code. In some places in code please provide clear comments at a pseudologic level for instance in: (1) for loops j then another loop inside - why there are 3 loops? (i will re-run and try again to understand) and (2) np.zeros, np.ones and np.eye(D) and then divide by array - Is it normalization, what is this division for? It would be nice to explain in a slide about what np.eye(D) really means. I will review numpy again and dig this. Also there are some greek alphabets alpha, beta. here could you provide what is alphas and betas purpose. Are you aggregating, why is this required? what is mu an sigma, may be I will repeat lagrangian multiplication, however I think it doesn't hurt to cover the points like alphas, betas, mu and sigma in a greek notations appendix slide or something like that in the next slide. I think subject its my level of understanding here as its my first run of the course and HMM topic.. So subject explanation needs to be given more weight age against code. For instance, does fit coin mean..fitting the model to sequence of coin tosses? I didnt get it, I will try and revisit it. On style of code walkthrough its great from a IT perspective. However, my 2 cents, It would be great to have a tell-show-tell approach as against tell-as-you-code approach. There is no need to code along, step-1: just tell pseudo logic, then step-2: show code while walk through without typing with little explaining and finally in step-3: tell about more as needed and go ahead run code. Subject by nature is complex, In fact you could give more whitepaper references, like point to experts who have published papers Also plant some quizes at the completion of each section, every kid likes a quiz! I have never seen anyone touch the order of data science courses covered like you did. This is great and I liked it very much.

Excellent theoretical introduction!

The style of lecturing/ the git hub content managing is really poor.
Maybe he has knowledge of the topic but he needs to polish how he is managing and transferring its knowledge.

C'est tres bien de course.
J'ai compris beaucoup des choses.
Merci

Great subject. I enjoyed the mathematics.
The implementation was too confusing: the python does not match the previously derived formulas (specifically gamma and xi). The variables used become too numerous to not be named specifically. formulas with plethora of single letter variables just become too hard to make sense.
I knew what GMM are but the part on HMM+GMM still does not make sense to me (why, reasonning, when to use, what is the interpretation).
During examples, the result is only ever shown to be the loglikelihood. never actual classification/prediction/modelisation. e.g. the example on the 2 poets could be really fun but al that is shown is the final likelihood. disappointing

Great subject. I enjoyed the mathematics. The implementation was too confusing: the python does not match the previously derived formulas (specifically gamma and xi). The variables used become too numerous to not be named specifically. formulas with plethora of single letter variables just become too hard to make sense. I knew what GMM are but the part on HMM+GMM still does not make sense to me (why, reasonning, when to use, what is the interpretation). During examples, the result is only ever shown to be the loglikelihood. never actual classification/prediction/modelisation. e.g. the example on the 2 poets could be really fun but al that is shown is the final likelihood. disappointing

Great

Another high-quality course easily at par with top university grad/research level courses. The course has very dense level mathematics and technical materials, and they are explained quite well. Not suited for beginners, but advanced-level practitioners would relish it.

The content is really amazing.
I was expecting for some assignments after each section to make this more interactive