Conditional Random Fields : Data Science Concepts

[@spiderkent]

I like your narrative description of the topic. It is good that you have written everything down beforehand so that you can refer to any part of the formulas to emphasize their relationship. Thank you for the effort, job well done!

[@huyhoannguyen9913]

Your explain is much easier to understand than the course that I attended. Keep doing the great job RitVikMath

[@jordanblatter1595]

I have an assignment on segmenting chinese words with crfs due tonight. Perfect timing!

[@ritvikmath]

Best of luck!

[@prodbyryshy]

this is the best video ive seen on this topic (for beginners) so far

[@erickleuro6159]

Thank you, great video! I used your other Time-Series video series (not pun intended) to help me with my final project, and they were super helpful!

[@ritvikmath]

Good to hear!

[@karunesharora3302]

It is a wonderful explanation for HMM and CRF. It would be great if you could post a separate video dedicated to generative vs discriminative models, as this becomes basis for various NLP models.

[@allendark2982]

The best video about crf ever!

[@bilalbayrakdar7100]

you are true pioneer of data science, you make everything understandable . keep it up

[@johnathancorgan3994]

I like the whiteboard presentation style, and your audio was fine.

[@ritvikmath]

Thanks!

[@WBPCS]

Thank you for the great explanations! I have watched several videos in different languages trying to get an intuitive idea of CRF, but unfortunately they all focused on symbolic maths. I do understand the maths, but I just couldn't reach an intuitive understanding from the maths. The comparison with HMM you make helped me a lot, and I have a much clearer picture of what CRF is doing after watching this video. Thanks a lot!

[@宋子阳-u4e]

Thanks Ritvik! The video is so clear and i've learned a lot!

[@nikhildharap4514]

Superb! Just can't thank you enough for these videos. You make the concepts so easy to understand.

[@nisharathod2945]

You make it sound so easy! Thanks dude

[@zhenwang5872]

Really good work! I found it inspiring to look at.

[@anoop8753]

Brilliantly explained

[@CarlosSoto-rn7jc]

truly amazing explanation! thanks!

[@muhammadal-qurishi7110]

Thank you for this video. I have to add something here and correct me if I am wrong: HMM is a general form of Naive Bayes whereas CRF is a general form of Logistic Regression.

[@uansholanbayev5670]

thanks man, finally got it clear

[@blairt8101]

saved my life again!

[@dragolov]

Bravo, Master!

[@sergioserino1823]

At last, I get it! Thank you!

[@zhiyili6707]

Thank you for the video. It is really helpful.

[@Giovanni-em7ny]

You are truly amazing!

[@DPCoder]

that was awesome explaination. Thanks alot.

[@BiggestITDisasters-br4jy]

Thanks for this video!

[@SEOTADEO]

Thanks a lot! Helps so much.

[@kunalnarang1912]

Hey Ritvik, great stuff! I have a question: How exactly does one define a different feature function for each timestamp in the sequence. Let's say that the X, Yi-1 and Yi are the same, but the only difference is i. Will that mean we have to define a different feature function every time we see that combination in the sequence. Is there an easier way to do this? Is that something we have to define before training the CRF?

[@mikewood8175]

Hey, such a great person you are at explaining. I just want you to make video on why LSTM backprop solves vanishing gradient intuition and also backprop of CNN model! I really have hard time understanding gradient flow of both these models. Just the intuition will work too.

[@n1984ster]

This video talks a lot about feature functions in CRF but HMM video doesn't elaborate on the feature functions concept as related to HMM. Like what feature function could be used in HMM. The HMM video talks about probabilities, but I couldn't find any mention of feature functions. @ritvik

[@cm-a-jivheshchoudhari9418]

what i dont understand is that we use conditional probabilities in HMM as well? P(Y|X) then how is it not discriminative but cRF is?

[@j.b.7237]

Hi Riktiv, what a great video. In my opinion the best understandable video on youtube. I still have a question, are the observed states X_i the respective segmented elements of our data (e.g. words or chars for textual data) or are these already the feautures? I found in the paper "An Introduction to Conditional Random Fields" by McCallum (the inventor of CRFs) a graph example of a CRF, where each Y_i had three connections to observations, but the observation states had only the connection to Y_i in each timestep.

[@namratanath7564]

Why would you want to use crfs instead of lstm s?

[@Hermioneswand1]

Thank you for this video, really helped me out!! The audio could be a little louder though

[@n1984ster]

The drawback of HMM having static transmission and emission probabilities I couldn't understand very well. Please if someone could elaborate a bit more.

[@raghavamorusupalli7557]

Hi Rishi, What about Z?

[@lexisense]

Awesome. It is a bit too technical for a linguist. Could you make it more easy please by adding some examples from the English corpus. Thank you in advance

[@ChocolateMilkCultLeader]

When you talk about generative vs discriminative model, please make sure to include a section talking about how these models can be combined. Their being exclusive is a huge misunderstanding in Machine Learning and something I've covered in my videos and articles. Hope you can cover that idea too

[@chidam333]

20:09

[@cherryfan9987]

thanks

[@lipe5331]

I love you

[@antrasen77]

why this accent though?😑