I came across two websites to improve your programming and interviews - https://cheatography.com & https://bytebytego.com/courses/system-design-interview/design-a-chat-system?fpr=javarevisited

Data science rule of thumb: Percentage changes are approximated by the difference in natural logarithms.

>>> x1, x2 = 25.1, 29.7
>>> (x2 - x1) / x1
0.18326693227091623
>>> log(x2) - log(x1)
0.16827919967091542

This is useful for interpreting log plots.

It works for percentage decreases as well:

>>> x1, x2 = 25.1, 19.9
>>> (x2 - x1) / x1
-0.20717131474103595
>>> log(x2) - log(x1)
-0.23214811440729166

https://www.manim.community/


Python 🐍 library 📚 for maths 🧮 animations. Check out

Xonsh is a Python-powered, cross-platform, Unix-gazing shell language and command prompt. The language is a superset of Python 3.6+ with additional shell primitives that you are used to from Bash and IPython. It works on all major systems including Linux, OSX, and Windows. Xonsh is meant for the daily use of experts and novices


https://xon.sh/

THE WORLD’S LARGEST SELF-DRIVING DATASET : https://doc.bdd100k.com/download.html

UC Berkeley open-sourced the largest ever self-driving dataset to the AI community in 2018. The dataset called Berkeley DeepDrive 100K (BDD100K) contains over 100,000 video sequences. Each video is 40 seconds long, shot at 30 FPS and 720p. GPS information is also provided, indicating the navigation route taken during driving.

The dataset covers a multitude of different weather and time conditions: sunny, rainy and hazy data captured both during day and at night also gives a well-balanced distribution that helps prevent overfitting.

Over 85,000 pedestrians are also present in the dataset. Therefore, the dataset also offers a reliable dataset for detecting pedestrians on the road/sidewalks. This scale of content available is quite massive too. This is 800 times bigger than Baidu’s ApolloScape dataset. https://www.bdd100k.com/challenges/eccv2022/

Analysis of various Machine Learning algorithms (Supervised and Unsupervised learning)

Lightly: A python library for self-supervised learning on images

Cool computer vision library for self-supervision and active learning.

Github: https://github.com/lightly-ai/lightly
Docs: https://docs.lightly.ai/

New version of django-upgrade released - for Django 3.2+ https://pypi.org/project/django-upgrade/

Thursday, 8 September 2022
AI Curated Latest Track
✍️✍️ Amazon Announces the Improvement of ML Models to Better Identify Sensitive Data on Amazon Macie
✍️✍️ Gartner has recognized Microsoft as a Leader in the 2022 Gartner®️ Magic Quadrant™️ for Cloud AI Developer Services,
✍️✍️ Google begins rolling out its AI Test Kitchen machine learning app
✍️✍️ Banned U.S. AI chips in high demand at Chinese state institutes
✍️✍️ PolyAI secures $40M for its AI-powered voice assistant platform
✍️✍️ The researchers using AI to analyse peer review
✍️✍️ Intel Loses Bid to Nix Rival’s Medical Machine-Learning Patent
✍️✍️ Canadian company uses machine learning to promote DEI (diversity, equity and inclusion) in the hiring process
✍️✍️ Django bugfix release: 4.1.1
✍️✍️ MIT’s new AI model can successfully detect Parkinson’s disease
✍️✍️ Stanford engineers present new chip that ramps up AI computing efficiency
✍️✍️ AI App Could Accurately Detect Covid-19 in Your Voice, Say Scientists
✍️✍️ India To create AI/ML Enabled Portal for Pensioners

Inscribed by
Raja

TEN mind-blowing AI websites that you’ve probably never heard of:

🔅 Magic Eraser. Remove unwanted things from images in seconds. Upload an image and mark the bit you want removed. Download your improved image.

magiceraser.io

🔅 Autoenhance. AI photo editor that enhances your workflow. Instantly get brighter, colourful and evenly lit images. Sky replacement and perspective correction to transform photos.

autoenhance.ai

🔅 Writesonic. AI writer that helps you write long-form blog posts & articles. Break free from writers block and scale your content production. Take content creation to the next level.

writesonic.com

🔅 Your own writing assistant and personal collaborator. Powered by AI to write better & faster. Create viral hooks and headlines in seconds.

tribescaler.com

🔅 Rytr - An AI writing assistant to help you write 10x faster.Enter your post topic, select the variants and creativity level. Choose the variants that are best suited to your style.

rytr.me

🔅 Namelix. Generate a short, catchy business name using AI. Decide using key words or domain extensions. Short names are unique, memorable and affordable.

namelix.com

🔅 Replika. Your AI companion who cares. There to listen and talk, always on your side.Ready to chat when you need an empathetic friend.

replika.com

🔅 Your AI assistant for meetings. Instantly record meetings across any web-conferencing platform.Transcribe, and search across your voice conversations.

fireflies.ai
🔅 Excel Formula Bot. Stop wasting hours creating Excel formulas. Turn your spreadsheet problem into a formula in seconds. Experience the full power of Excel & Google Sheets AI.

excelformulabot.com

🔅 Talk To Books. Get quotes from books that respond to your question. A creativity tool by Google to explore new ideas. Explore an index of > 100,000 books.

books.google.com/talktobooks/

—————🤷🤷Tensor Parallelism

There are >= 3 1/2 paradigms for training deep neural nets on multiple GPUs:

1) Data parallelism
2) Model parallelism
3) Pipeline parallelism
( 4) Tensor parallelism)

Which one(s) are you usually using; and anything missing?

1) Data parallelism: splits batches & distributes those across several GPUs. Here, in each iteration, the gradient estimate (for the model update) is computed from as a weighted avg over sub-batches.
Eg via DataParallel or DistributedDataParallel (recommended) in PyTorch

2) Model parallelism: divide model onto separate GPUs; usually to deal with limited VRAM. Note it doesn't imply that the training happens in parallel!

How? L1. to('cuda:0'), L2. to('cuda:1') etc.

Tutorial: pytorch.org/tutorials/inte…

3) Pipeline parallelism: related to model parallelism where you put different blocks of the model onto different GPUs(, and data parallelism where you split batches). But here you make those blocks run (somewhat) in parallel.

4) Tensor parallelism. It's basically a flavor of model parallelism, but instead of dividing the neural network by layer (horizontally), you divide the tensors themselves (vertically). E.g., put half of a weight layer on one GPU, and the other onto another GPU


—🪡excellent info ℹ️————

The AI innovations on the Hype Cycle reflect complementary and sometimes conflicting priorities across four main categories:
🔅Data-centric AI
🔅Model-centric AI
🔅Applications-centric AI
🔅Human-centric AI

DevOps - Tools 🪡🪡

The difference between coding interviews and a tech job is you’d be fired if you actually checked in leetcode-style solutions instead of stuff like this. Think!!

BoonDock Tip : { Section - Python}

Do not initialise the empty String with quotes. Try to use 'None' in your projects. It is the best practices though.

sqlConnection=“” ——- Not good practice
sqlConnection=None —- Best practice

15 Websites To Follow As A Developer

1. Stackoverflow
2. Google
3. YouTube
4. DevDocs. io
5. Github
6. Freecodecamp
7. LeetCode
8. IndieHackers
9. Udemy
10. Hashnode
11. Medium
12. Dev. to
13. W3Schools
14. Codecademy
15. Hacker News

May be you have another list 😇😃

🐸💬 - a deep learning toolkit for Text-to-Speech, battle-tested in research and production

• High-performance Deep Learning models for Text2Speech tasks.
◦ Text2Spec models (Tacotron, Tacotron2, Glow-TTS, SpeedySpeech).
◦ Speaker Encoder to compute speaker embeddings efficiently.
◦ Vocoder models (MelGAN, Multiband-MelGAN, GAN-TTS, ParallelWaveGAN, WaveGrad, WaveRNN)
• Fast and efficient model training.
• Detailed training logs on the terminal and Tensorboard.
• Support for Multi-speaker TTS.

https://github.com/TheAlgorithms/Python/blob/master/DIRECTORY.md


One place where you can find hundreds of algorithms.
Even you can find them in another languages as well.

https://github.com/TheAlgorithms