#python #sqlalchemy

Хоть и старая, но очень крутая преза. Обязательна к просмотру всем, кто работает с БД

"How do you take the big step from casual SQLAlchemy user, who treats your database as a mysterious object store, to advanced power user, who optimizes critical queries, plans indexing and migrations, and generates efficient reports?"

https://www.youtube.com/watch?v=399c-ycBvo4

#oauth2.0

A well written article about OAuth. Recommended

https://gravitational.com/blog/everything-you-need-to-know-about-oauth/

#dl #ml

To their point, a Synced report estimated that the University of Washington’s Grover fake news detection model cost $25,000 to train in about two weeks. OpenAI reportedly racked up a whopping $12 million to train its GPT-3 language model, and Google spent an estimated $6,912 training BERT, a bidirectional transformer model that redefined the state of the art for 11 natural language processing tasks.
In a separate report last June, researchers at the University of Massachusetts at Amherst concluded that the amount of power required for training and searching a certain model involves the emissions of roughly 626,000 pounds of carbon dioxide. That’s equivalent to nearly five times the lifetime emissions of the average U.S. car.


https://venturebeat.com/2020/07/15/mit-researchers-warn-that-deep-learning-is-approaching-computational-limits/

#Math #ML #MIT

This course introduces the Dynamic Distributed Dimensional Data Model (D4M)

Goal
To dramatically reduce the time to develop complex algorithms for analyzing large data sets.

https://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-042j-mathematics-for-computer-science-fall-2010/

#ML #AutoML #dabl

Один из контрибюторов в scikit-learn и по совместительству Принципал из Майкрософта дает крутоую презу на тему автоматизации МЛ процесса.

Он написал либу dabl.

The dabl library is a first step in this direction. It provides automatic visualization routines and model inspection capabilities while automating away model selection.

https://www.youtube.com/watch?v=TQnxH90PqFc&list=PLYx7XA2nY5GejOB1lsvriFeMytD1-VS1B

#SOLID #Software #Dev

A good and simple talk on SOLID principles:

https://www.youtube.com/watch?v=rtmFCcjEgEw

This guy is simply awesome! ☝Lessons learned in a hard work

https://www.youtube.com/watch?v=bEg5ySTUGxE

#distributed #systems #articles

The past week, I've been reading and watching a lot about Distributed Systems. If you are fine with reading, then this one is for you. A list of plenty of useful articles to read covering basics

https://www.the-paper-trail.org/post/2014-08-09-distributed-systems-theory-for-the-distributed-systems-engineer/

digraph s3-website.eu-central-1.amazonaws.com

#FTS #Go

Let's build a Full-Text Search engine in Go

Full-Text Search (FTS) is a technique for searching text in a collection of documents. A document can refer to a web page, a newspaper article, an email message, or any structured text.

Today we are going to build our own FTS engine. By the end of this post, we'll be able to search across millions of documents in less than a millisecond. We'll start with simple search queries like "give me all documents that contain the word cat" and we'll extend the engine to support more sophisticated boolean queries.

https://artem.krylysov.com/blog/2020/07/28/lets-build-a-full-text-search-engine/

#aws #spot #instances #cost #optimization

https://habr.com/ru/post/509790

Если вы думаете о зарубежной магистратуре, то почему бы не рассмотреть теплую южную страну? Например Италию. Ловите подборку англоязычных ИТ программ в этой прекрасной стране.

Politecnico di Milano:
📌MSc in CS and Engineering
📌MSc in Geoinformatics Engineering

Politecnico di Torino:
📌MSc in Data Science and Engineering
📌MSc in Computer Engineering

Sapienza Università di Roma:
📌MSc in Artificial Intelligence and Robotics
📌MSc in Computer Science
📌MSc in Cybersecurity
📌MSc in Data Science
📌MSc in Engineering in CS

University of Bologna:
📌MSc in AI
📌MSc in bioinformatics

University of Trento:
📌MSc in AI systems
📌MSc in human-computer interaction
📌MSc in Computer Science

University of Padua:
📌MSc in cybersecurity
📌MSc in Data Science
📌MSc in Physics of data

Free University of Bozen-Bolzano
📌MSc in Computational Data Science
📌MSc in Software Engineering

#DL

A really cool specialisation on deep learning from amazing instructors

If you want to master it and have a portfolio by the end then you should for for it.

The price is quite affordable. You'll pay it back very soon, I believe

Deep Learning
https://www.coursera.org/specializations/deep-learning

Recently, I've been looking for resources on DL/ML and production. Here's a good resource that addresses the full cycle starting from setting/formulation of a problem to deployment

https://course.fullstackdeeplearning.com/

​​В США протестировали сеть спутникового интернета от SpaceX

Микроспутники Starlink от компании SpaceX Илона Маска раздают интернет со скоростью до 60 мегабит. Исследование спутникового проекта показало, что для потребительских целей средние показатели достаточно высокие.

Пока компания ведет работу над запуском на орбиту новых спутников для раздачи интернета, возник вопрос о том, какова будет скорости связи в реальности. Пробные замеры показали, что 11—60 мегабит. Средняя скорость — чуть несколько больше 40 мегабит.

Значительно ниже скорость для выгрузки данных – 4-17 мегабит. Замеры делались в Лос-Анджелесе, Сиэтле и Вашингтоне.

Таким образом скорость загрузки и выгрузки ниже желаемой руководством SpaceX — 1 гигабит. Хотя для пользователей достаточно и 15 мегабит, поскольку стандартная скорость «земного» широкополосного интернета в США такая же, а в сельской местности и того ниже.

Сейчас сеть спутников Starlink еще тестируется и, не исключено, что в конце концов скорость соединения приблизится к желанной для SpaceX. В настоящее время на орбите находятся примерно 600 спутников. Их количество планируется расширить до несколько тысяч, чтобы обеспечить глобальное покрытие.

Для тех кто изучает МЛ и хочет лучше понять что такое AUC, ROC

https://dyakonov.org/2017/07/28/auc-roc-%D0%BF%D0%BB%D0%BE%D1%89%D0%B0%D0%B4%D1%8C-%D0%BF%D0%BE%D0%B4-%D0%BA%D1%80%D0%B8%D0%B2%D0%BE%D0%B9-%D0%BE%D1%88%D0%B8%D0%B1%D0%BE%D0%BA/

#ML, #softmax
I've asked my self why Softmax uses exponent. Here's a good response:

- Monotonically increasing — To ensure that larger inputs are mapped to larger outputs.
- Non-negative outputs — Because probability values must be non-negative.
- The outputs should sum to one — This can be achieved by simply dividing each element of the output by the sum of all elements of the output.

https://www.quora.com/Why-is-exponential-function-used-in-softmax-function-in-machine-learning

Here's another good question, that Daniil you may ask shortly: Why logit is slower than relu

https://www.coursera.org/learn/neural-networks-deep-learning/discussions/weeks/1/threads/OC2Bh8ahEeetfwrt7EZTMA

We've been discussing it with Agerke last week.