muktabh

That is so relatable. I use Tomboy Notes instead of just plain text file as it is stuck on my desktop and keeps reminding me that downloading and trying out a new github project is something I can spend next 5 minutes on, but not the next hour doing.

I also somehow feel that just listing down tasks on Tomboy in terms of achievable chunks helps my focus.

I absolutely love working in FastAPI and pydantic. So much so that the backend of a new app I just launched in private beta on is written in FastAPI. My startup has traditionally been a flask workshop but FastAPI is just so much more productive due to the documentation and pydantic type checks.

I seriously hope Python developers understand that apart from the superb programming language it is, package ecosystem also is what makes people prefer Python as the language to invest their hard labor into. Small teams need productive libraries like FastAPI to develop systems. Please don't kill FastAPI and pydantic. I don't want to go back to writing flask again after working in FastAPI.

Huge FastAPI fan here. I have used FastAPI and friends (tortoise ORM, fastAPI-sqlalchemy and arq) in my recent projects and it has been the most enjoyable web services writing experience in my life. I never had thought python optional typing had any use before I used fastapi.

Slightly out of context here, I find the entire stack of bsb, bsb-native, ocaml and esy pretty cool. However, I just dont find enough resources, good tutorials etc on Google search. Is there a good set of beginner tutorials anyone can point to ? Thanks in advance.

I dont disagree with the premise that credential based tertiary education credits for jobs are pointless and a genius can come out of any university. You can do well despite of where you studied. There are people who succeed in life due to no talent and just connections.

However, this analysis is just based on outright flawed data. The author has no idea about how education system works in India, or deliberately chose to ignore it because this is what could be done on the data they had accumulated. And I am not talking about the disclaimer about data sample being small, the entire model is wrong for majority of Indian students.

Couple of things I want to comment on when you look at the article and the original paper :

1. The study treats few public institutions as "elite" and private institution as for non-elite people. That is wrong. There are private institutions which are elite too. My alma-mater is one of the elite ones and it was not a public university. This is the least problematic error in analysis.

2. The assumption is a common academic score being used to score university students. For college entrants in early 2000s (the time period when the survey respondents applied to colleges), people had to literally give 50 different entrance exams to get into universities. The score which is used to evaluate who was marginally better to get into elite university and who was rejected (the entire premise of the article), IS NOT what is used to decide entrance into universities in India. Its getting slightly standardized now, but was really chaotic 15-20 years back. There is a common exam that people give at secondary / senior secondary level but very few universities entertain them at entrance. Actually there are many "boards" to give Senior Secondary exams as well. Net net, there is very large variety of ways by which universities in India take entrance. Even the elite ones. I think, given the author is from the field of economics, they are overgeneralizing. A few (not all) economics elite universities do treat senior secondary tests as entrance criterion.

3. Exit exams used to compare academic credentials of students to see if they are academically better than the other DO NOT EXIST ! They dont even exist now and there is no point of them existing. Every public/private university in India (at least the elite ones), have their own grading/curriculum. So the academic credentials cannot be compared directly. With different universities having different grading method, people having somewhat similar academic distribution is a known thing (I dont recollect the name of the phenomenon exactly, but its studied and published). That said, it doesn't go with how the data has been modeled here.

Overall, the author's model is wrong for many (50%+) of Indian students. They might be referring to economics colleges (given they come from field of economics) Delhi University, Mumbai University and Kolkata University type of institutions where their model of tertiary education fits correctly, but in that case her analysis doesn't work well for most of "India". Totally wrong messaging.

If someone finds it helpful, here is a list of 25 free Machine Learning books available online : https://blog.paralleldots.com/data-science/24-best-and-free-... and another list of over 50 free Data Science books : https://blog.paralleldots.com/data-science/50-must-read-free... we compiled. Pick your poison.

This is so well written. I remember being taught these in my first year engineering and I really never understood a lot intuitively. (I could solve numericals and clear the course but I did not get it then. Being a CS student, I never had to care about it ever since). I wonder how the world would have been different if all courses were taught this way.

Just checked, works for me on 2 different ISPs. This looks like fake news.

Apart from the huge human misery this technique can avoid, also hope it can avoid the pandemic of avian malaria killing many birds.

Unlike what this article tries to subtly hint, there is no massive conspiracy theory stuff going on here. There are too many people in India (with smartphones and cameras and 3G/4G internet) to pull off anything covert. No one here views this as mystery but as a result of country's discipline in lockdown till now. The country imposed an early screening and tight lockdown and people followed it well till now.

Farmers are financially not that well off and a big share of population in India is farmers. There is already provision of prosecuting farmers in law, but no administration wants to punish the relatively poor + electorally important farmers by enforcing the laws.

The windows build isn't working for me. Still stuck on 0.4 .

I bought a copy of the book a few months back (and haven't yet found time to read it fully, I can write a lot of programs just after reading a first few chapters), will Manning upgrade it for a newer version for Nim 1.0 ?

I am data scientist at ParallelDots , the company that has launched this Excel plugin. We realised that a lot of people want to use Machine Learning solutions like keyword extraction, sentiment, emotion etc. on Excel itself where they collect data. Here we have made such facilities accessible as Excel formulae where you need zero programming skills to use them. We have provided a free tier where you can try it out. I am happy to answer any questions about the functionality.

A lot of universities are open to give a separate commercial license when contacted. They charge for their efforts, which is fair. Whether public universities should charge given we already pay them from our taxes is a different issue. source: I am also cofounder at an AI startup, we often buy licenses to use academic datasets for commercial usage.

To add a bit to Parth's answer it has been trained using metric learning type losses and normal softmax combined. There is a next version of combined low/high level features combined algorithms we are slowly going to update into.

Hi, That can be done, but image generation algorithms require a bit more time to be good enough for such a thing. Two major bottlenecks (I am assuming you are talking about GANs) are: 1. The algorithms are still not too good for hi res images. 2. The algorithms work very well on images where fine grained features are not that important. (For example images of nature), but they introduce unwanted features otherwise (Eyes looking other ways etc.). We might be all ready by next ICLR/NIPS though, cannot say.

Google has its own huge internal datasets for image classification. You can check for its mention in Chollet's ExceptionNet paper. That may be the reason why they are not really interested in working on imagenet.

Most data is proprietary and out of reach of researchers, so we don't have examples. I am pretty sure manufacturing companies will have internal teams which will be working on their internal data.

This needs to be updated everyday seeing the rate at which new achievements coming up. Just yesterday, an algorithm outpaced gynocologists in IVF.

Fake News is a harder problem due to two reasons:

1. Fresh News constantly keeps cropping up. Its a dynamic problem, not something that can be done on a static dataset. 2. Verification is a harder problem as compared to clustering. One of the problems is what data to treat as ground truth for verification. Technologically it is composed of hard problems : finding out each individual fact in a document and then verifying them from close facts from ground truth article.

Yes. We love PyManOpt here at ParallelDots (the startup which wrote the blog).

This looks like a good addition. Will suggest to add it.

Thanks for the suggestions. I don't think PyMC3 and Stan are less known, hence did not suggest to add them in the list. Dimple and Church look good additions. Will get them appended.

That's a grammatical mistake in the title. All the libraries mentioned have a of work behind them. [disclaimer: I helped compile the list]

Here is something you should check out in that case https://github.com/eriklindernoren/ML-From-Scratch . Although be warned, everything from scratch approach l, despite being addictive, is harder than standing on shoulders of giants and takes too much resilience. If you are even a bit of procrastinater like me, it's tend to stay a sweet fantasy a lot of times. It's only when I was half working on a project and had no other option, I could go for implementing my first algorithm from scratch.

Thanks for the hard work sublime text team for making programming so enjoyable. I already find ST3 so flawless that to think it still can be improved is beyond me. Once again, great going.

I have been asked to write books too, but I go with a no response always. My reasoning is generally that my I have read free books and learnt a lot of what I have ( was not really rich to buy all trendy books being a middle class Indian student) , my book whenever it is written would be open and free access. That will be my way to pay back. If I have to make money , I will rather do it through my work, not letting publishers have control over the knowledge source I write.

This was in 2014. Here is a more recent perspective (from a different blogger who used Julia in production) https://blog.staffjoy.com/retro-on-the-julia-programming-lan...

We make a contextual recommendation engine as a service for online publishers at our startup ParallelDots. We discovered the problem of tags not really working well for recommendations on our clients websites too. We ended up using unsupervised word embeddings and auto encoders on top of them to solve the problem. We dont still use it for personalization though, just contextually similar articles. Great seeing some of similar problems being solved at New York Times too. :)