In January 2022, I experimented here on Twitter.
Instead of my usual overthinking, I went full steam ahead and shared fun stuff and created deep dive threads.
In case you missed them, here are the popular
- 5๏ธโฃ Tweets &
- 5๏ธโฃ Threads
in machine learning, Python and careers
๐ค Starting out strong on some machine learning shenanigans!
Huggingface has an amazing library of models.
In this tweet, I shared three models for three different applications.
Have fun & experiment
Pretrained models are the most fun you can have in machine learning.
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 30, 2022
Go to @huggingface models and play.
Computer Vision?
vit-base-patch16-224
Text generation?
distilgpt2
Text-to-speech?
fastspeech2-en-ljspeech
Build something and break it.
Explore, have fun & test limits.
๐ฅผ Some good ol' advice for researchers
Install the Unpaywall Chrome extension to get papers for free legally!
(Bonus tipp: You can always email scientists, we love it when you're interested and nice!)
Researchers almost always have a right to give their papers away for $000.
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 25, 2022
And there's a tool to find those free versions of the expensive ones.
No piracy,
actually completely legal.
Unpaywall just makes it super easy for you to access science. pic.twitter.com/kdWbUmc5FH
๐ Jack-o-... all trades?
Had some nice chats over on Linkedin under this one. I love busting a common quote, because a lot of them are misused.
Go be a jack of all trades:
Jack of all Trades,
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 28, 2022
Master of None.
Do you know the full quote?
The third line is:
But better still than Master of One.
Go explore that niche interest.
Take a few courses in something non-monetizable.
Pursue that hunch, even if it's just for the joy of learning and growing.
๐จ Matplotlib can be hard to work with.
Proplot attempts to make matplotlib easier to wrangle.
Get some sane defaults &
reduce the boilerplate code to a minimum.
(This one's from my newsletter actually! ๐)
Scientific plots in Python?
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 28, 2022
Probably will dust off good ol' Matplotlib.
Check out Proplot to get some sane defaults to make your life a bit easier.https://t.co/KuB5lqw2L9 pic.twitter.com/zS9XF1VobV
Subscribe to receive insights from Late to the Party on machine learning, data science, and Python every Friday.
๐ Back to matplotlib!
Get a completely free book on scientific visualizations. A lot of these, are absolutely next-level. The book is good for inspiration and application!
A Cookbook for Scientific Visualizations in Matplot?!
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 29, 2022
For $000.00?
My yes, thank you I'll take that.
Matplotlib can be rough to wrangle.
This is the manual you never knew you needed!
It has plots you never knew you needed!
I at least didn't...https://t.co/eBnTeWVXdP pic.twitter.com/EGGHpoDO5H
๐ฉโฌ๐จ Cassidy Williams asked us to build a Wordle validator in her newsletter!
I used a thread to go through my approach how to solve these types of problems.
Honestly, I hope this helps some junior devs on whiteboard interviews
Solving coding challenges is hard but it's great practice for tech interviews.@cassidoo posts a solving challenge in her newsletter every Monday.
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 20, 2022
Let's work through this week's Wordle guess checker.
๐ Take a page from my book!
๐ ๐งต
๐ My posts about being selected as a @SoftwareSaved fellow were super supportive!
That was just lovely seeing so many people reach out.
I have been selected as a @SoftwareSaved fellow ๐
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 13, 2022
I'm so stoked to finally share this with you!
With the help of this fellowship, I will create content around
๐ค sustainable machine learning applications in research!
Here's what that means!
๐ ๐งต https://t.co/5OmsAXNTsP
๐ Reproducible code in research helped my paper get over 50 citations.
I go through some advice how anyone can reproduce this huge success (on my scales)
My first first-author paper broke 50 citations. I think I know why and you can do it too.
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 27, 2022
I shared my code and made it ridiculously easy to use.
I'm lazy myself.
If I find a paper that gives me all its ingredients ready to go,
I'm happy.
But how do you do this?
๐ ๐งต
๐ง My first thread in January did really well.
People are missing the best parts about Kaggle, in fact Anthony Goldbloom shared some cool data that validated my intuition on the same day!
I added it to the thread:
You're using @Kaggle wrong.
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 8, 2022
Get the maximum out of the platform.
Learn from a bunch of data science and machine learning professionals. Explore free applied expert courses for all levels and explore tricks for winning comps.
Here's how you use #Kaggle like a pro!
๐ ๐งต pic.twitter.com/q9srNG58GA
๐ธ Trillions of Dollars are lost in PDFs worldwide
This must've been my most popular thread this month.
I thought it would, because the hook is pretty interesting and it's a super common problem.
Also good advice and tools.
A trifecta of goodness.
Worldwide Trillions of Dollars in value are lost in PDFs ๐ธ
โ Jesper Dr.amsch @[email protected] (@JesperDramsch) January 11, 2022
We pay thousands for a 300-page report that is at best skimmed.
Every company does it internally or externally!
Here's how you unlock that $$$ for free.
(With a sprinkle of Python, @huggingface and @spacy_io)
๐ ๐งต pic.twitter.com/7Ar3ciHPwY