Ad

Thursday, December 2, 2021

Newsletter November Part 2

Exclusive right has been retained to repost this on the internet. Uniqtech only posts on uniqtech.medium.com, ml.learn-to-code.co, and here occasionally. All other posts are not authorized, may contain malicious links. In this special edition, we will talk about both machine learning (our usual data) as well as web3 - the next evolution of web. In the next newsletter, in early December, we will also analyze the job postings of a few web3 and blockchain companies such as NFT marketplace OpenSea. Our staff recently attended their recruiting event. Blockchain curriculum and flash cards are available for pro members at NO additional charge. It's included in your $5 / month membership! Yay. 


Blockchain and web3 seem to be the future, though Elon Musk surely disagrees (he tweeted today), though he seems to support Dogecoin. Art Basel, a premium fine art, NFT art, and blockchain art fair is happening today in Miami, the city of crypto.


Because blockchain is a public, distributed database, we data people, machine learning practitioners won't be in foreign territory. What we know have a lot of synergy! If you know web development, even better. 


A reminder : you can log into the website with either gmail or github! 😸 The technology is Google Firebase Authentication. cheers

https://ml.learn-to-code.co/firebase_auth.html


Any feedback, questions are welcome hi@uniqtech.co or through the message tab on the site. 


You can become a pro Medium member via our uniqtech referral link. A portion of your membership fee will automatically be given to us and help us build tutorials on medium! Join our 1100+ subscribers to the data science bootcamp.  https://uniqtech.medium.com/membership 


Did you know that you can request content at hi@uniqtech.co? Pro substack members get all the content. If you are a medium pro, clap for our stores and subscribe, we can release some premium tutorials to you too.


You can now tip us dogecoin! 🐕 or any crypto 🚀 🌕

Like our tutorials? You can donate/tip us! Help us make more tutorials. Send us $DOGE any #crypto Like our tutorials? You can donate/tip us! Help us make more tutorials. Send us $DOGE any #crypto https://commerce.coinbase.com/checkout/e2175af7-7b63-4d5a-a418-79edf6e2f397 powered by @coinbase not tax deductible. #learntocode #100DaysOfCode #learnprogramming tutorials on medium: uniqtech.medium.com #machinelearning #data 

This is a part of the @coinbase #weacceptcrypto campaign!

You can tip us or pay your membership $5/month payment using Crypto. 🐕


In the past, we discussed the preprocessing, vector representation aspect of deep learning models. Data cleaning and processing should take up to 80% of machine learning practitioners' time, but today we need to address the elephant in the room : deep learning training. This is both the art and math of AI. In the machine learning section, we will discuss training loops.


Easter egg for subscribers :

Best resources for understanding deep learning math and formula [flash card, pro, easter egg, PRO only]

https://ml.learn-to-code.co/skillView.html?skill=PQhmnCII2bh6oClghsCT 🥚


DISCLAIMER WARNING

We have updated our disclaimer and terms of use to include domain info and blockchain, crypto related info. Please review.  Link here. By reading, using our content, you agree with what's outlined in the disclaimer article.



BLOCKCHAIN

Did you know the hottest crypto and NFT trends are based on the blockchain technology, a public ledger system which is also a database! And that's why our paid substack subscribers / website members will get free blockchain learning materials. We publish tutorials great for the Data Science Bootcamp (publication). 


What is EVM? [flash card]

https://ml.learn-to-code.co/skillView.html?skill=Q9Yigd3XYtfckbilhrOM


Premium Medium members, comment on our posts, clap, follow us on twitter @siliconlikes send screenshots to hi@uniqtech.co to win blockchain tutorial screenshots. We won't be able to send u the entire package, but we will send u as much tutorial "candies" as possible~ Active members will automatically be entered into the next free study material raffle, no re-entry necessary! New graduate, students without earnings can apply for scholarships 3 months pro member on our LTC site. hi@uniqtech.co

https://ml.learn-to-code.co/topics.html


Kaggle (data science competition site, a part of google) is hosting a crypto forecasting competition right now. Check it out and learn - https://www.kaggle.com/c/g-research-crypto-forecasting Talking about Google, Google also hosts digital skills workshops and some of those are free - educating the employees of tomorrow's job market! http://g.co/AppliedDigitalSkills Talking about free education, pro members, check the previous newsletter on substack for free premium coursera based training (easter eggs).


This week the majority of our engineering time is spent on chainlink hackathon. The sponsors are impressive : polygon, avalanche, moralis, alchemy.io - huge names in blockchain tech. We also noticed that blockchain gives almost 10x higher prize pool than conventional hackathons. 


We may generate NFTs for our subscribers. Any coding NFTs you would like ? Message us with your ideas hi@uniqtech.co


How to prepare for hackathons and competitions? We have lots of tips for you. Our staff has won Facebook HQ Pytorch deep learning, Udacity Angular JavaScript hackathons.


The blockchain glossary, vocabulary list landing page is live! More construction is needed, but it's available as a smart tutorial. Anyone who log in can view. Doesn't have to be a member. Member can view premium flash cards. 

https://ml.learn-to-code.co/path3.html?tutorial=p_smart_tutorial_blockchain_NFT_crypto_glossary

Previously we launched NFT smart tutorial landing page

https://ml.learn-to-code.co/path3.html?tutorial=p_smart_tutorial_NFT_explained


Recently our staff made a first purchase on NFTs and cryptos. We learned a lot about the initial minting, reveal, what can go wrong in code, deployment and how to track on blockchain. Will share more soon. Want to know more, message us hi@uniqtech.co We are happy to share drafts with members. 


Web3 technology explained

https://ml.learn-to-code.co/skillView.html?skill=wwJPc0Tix4CF7kez2m6V

 [pro, high quality, pro tip, web3] 


Everyone is talking about web3.This is what Robinhood newsletter said:

https://ml.learn-to-code.co/skillView.html?skill=UZXIPGTD2SLvF77dNRU6


This crypto punk explains web3 with just a tweet!

https://ml.learn-to-code.co/skillView.html?skill=2Q350uSUEQcj5GGQ61FT



METAVERSE

Metaverse also include gamefi, requires developer to have skills such as : in-game economy and game algorithms / design

Metaverse gainers blockchain


MACHINE LEARNING + DEEP LEARNING

This newsletter's machine learning section is dense!  Don't worry, we will resurface these flash cards and incorporate them into lessons, so we can slowly memorize these flash cards through repetition. 


MACHINE LEARNING FLASH CARDS + DEEP LEARNING

rule of thumb: how to calculate dot product [PRO member]

https://ml.learn-to-code.co/skillView.html?skill=gq5Ij2ph4JYCy1AzNnaB


rule of thumb, in_features, out_features [PRO member]

https://ml.learn-to-code.co/skillView.html?skill=TnUpKKcfdUfS1VlkaUPq


Why did OpenAI researcher build codex after GPT-3

https://ml.learn-to-code.co/skillView.html?skill=mJx50k8UEdyTfyJYdpnp


UTC timestamp [blockchain, time series, data, data type, data preprocessing, date time]

https://ml.learn-to-code.co/skillView.html?skill=DsmMg0xKUqSBS5yk5TUh


Pytorch hook

https://ml.learn-to-code.co/skillView.html?skill=ny7iqbAAiNOZQX7lOltt


Weakness of GPT-3 model

https://ml.learn-to-code.co/skillView.html?skill=WS1ff6M6T2rHQoi81bJk


This newsletter's mega topic


TRAINING LOOP


Today's main focus in Machine Learning training is : training loop and gradient

What is training? Training is the learning part of "machine learning". In iterative processes, your models compare predictions (y_hat) with ground truth (y) in supervised learning. Then it will compute a heuristic of distance, which represents how good or far off is the prediction, also represents how distant is the prediction from the ground truth. Your models' goal is to minimize the distance using math and algorithms tricks, educated guess, even randomization. 


Depends on the metrics chosen, the distance result in very different numbers. There are strengths and weaknesses of each distance function, making each of them suitable for different tasks. There's no one magical distance function. An example of distance function in regression is euclidean distance. An example of distance function in classification is multi-class cross-entropy. In some scenarios, when comparing two vectors, their dot product can be used as the distance function. The choice of distance function takes trial and error. And can be an art. 


Check out the machine learning flash card landing page for more information on training APIs how the .fit function in scikit-learn different from pytorch custom loop.

https://ml.learn-to-code.co/topic_end_to_end_machine_learning_course_flash_cards.html#training_uniqtech_guide


Because of the interdisciplinary nature of machine learning (a perfect partnership of math, statistics, computer science, probabilities, even biology, neuroscience and more), there are often different versions and names for the same terminology. It's confusing for beginners, even experts. Other names for distance function are similarity measures, loss function, cost function. 


We won't be able to explain and teach the training loop in one newsletter. We will need to create a series of youtube videos for our subscribers. But we want to show you the codes, line by line, and what's the purpose of each line. How does pytorch training loop differs from Tensorflow? 


On the machine learning flash card landing page, we explained tensorflow and scikit-learn has a high level API .fit() To train a model you just have to call the fit API on X(input) and y(output) in supervised learning. The pytorch custom training loop is a bit more involved. Just a few more lines of code. You can also write custom loops in Tensorflow. In this newsletter we just want to link to the source code. Don't worry about understand every little detail. We will want to record videos that explain these algorithms line-by-line. But if you take a close look, the training loop is under 50 lines of code. That's pretty manageable! Don't be intimated. If we learn one section a day, we will be done in a week. 


Pro members check out the pytorch custom training loop, annotated here, in this flash card. 

Annotated flash card for pytorch training loop source code

https://ml.learn-to-code.co/skillView.html?skill=AOUkVyCzxd3shMR2YbY7


"sklearn (scikitlearn) and keras just call .fit() to train. The format is model.fit(X_train, y_train). In Pytorch you will have to do a bit DIY your own custom training loop."


Tensorflow also supports custom training loop [pro, tensorflow, training, ]

https://ml.learn-to-code.co/skillView.html?skill=mMoge83m7UhGzyuzogcN


Like our tutorials? You can donate/tip us! Help us make more tutorials. Send us $DOGE any #crypto https://commerce.coinbase.com/checkout/e2175af7-7b63-4d5a-a418-79edf6e2f397 powered by @coinbase not tax deductible. #learntocode #100DaysOfCode #learnprogramming tutorials on medium: uniqtech.medium.com #machinelearning #data


Chain rule, review [flash card, gradient, public]

https://ml.learn-to-code.co/skillView.html?skill=Q71hxQGPCtEB9xUY7LM1


Best resources for understanding deep learning math and formula [flash card, pro, easter egg, pro only]

https://ml.learn-to-code.co/skillView.html?skill=PQhmnCII2bh6oClghsCT 🥚


Understand matplotlib in one minutes. Data visualization cheat sheet [pro tip, high quality, data viz, cheat sheet]

https://ml.learn-to-code.co/skillView.html?skill=XTuqJlHv6KwP3uVlIl2r




INTERVIEWs


More contents regarding interview coming in the next newsletter. We will be analyzing web3 blockchain job posts such as from work-for-startup expo at Y Combinator and NFT market place open sea. As you know we are huge fans of hackathons! Recently we participated in Chainlink smart contract hackathons. We have a lot of tips to offer. The prizes in blockchain hackathons are insane they are 10x higher than what traditional hackathons offer. It's truly a blockchain renaissance. We will offer more information in the next newsletter.


onsite interview basics

https://ml.learn-to-code.co/skillView.html?skill=lTq6Gijln6wHbblG52QB


Like our tutorials? You can donate/tip us! Help us make more tutorials. Send us $DOGE any #crypto https://commerce.coinbase.com/checkout/e2175af7-7b63-4d5a-a418-79edf6e2f397 powered by @coinbase not tax deductible. #learntocode #100DaysOfCode #learnprogramming tutorials on medium: uniqtech.medium.com #machinelearning #data


No comments:

Post a Comment

Flask 2022

 increase jinja2 template cache limit from 50 to above. to speed up flask app significantly source: One line of code cut our Flask page load...