In this new Notebook I explore several techniques for doing convolutions and go on a hunt for the fastest conv implementation!
Along the way you'll learn about Fourier transforms, Toeplitz matrices, , JAX and JIT compilation.
kaggle.com/code/carlolepe (1/2)
Follow
Click to Follow carlolepelaars
Carlo Lepelaars
@carlolepelaars
Carlo Lepelaars’s Tweets
#Python gives you the ability to get your MAC address using the `uuid` module.
Here's an example:
4
26
144
What an amazing resource on quantum computing by and ! If you've ever been curious about quantum computing, definitely check out "Quantum computing for the very curious"!
You will not be disappointed, I promise! 😃
1
3
I started a YouTube channel! The first video is an introduction to p-adic numbers, without relying on modular arithmetic. Animated with #SoME2
11
42
127
Very much looking forward to the new Nolan movie about J. Robert Oppenheimer! Look at that cast!! 😱
imdb.com/title/tt153987
1
4
May you become so legendary that even your False Starts are officially published!
Bonus articles not mentioned in the Notebook, but amazing articles to better understand Fourier transforms:
- Understanding FFT:
jakevdp.github.io/blog/2013/08/2
- Fourier by example:
ritchievink.com/blog/2017/04/2
- Quantum Python:
jakevdp.github.io/blog/2012/09/0
Show this thread
Question remains: Why aren't all modern CNNs implemented using Fast Fourier Transforms? There is loss in interpretability, but for many applications the speedup looks worth it. It seems there haven't been many innovations in Fourier CNNs since 2017:
link.springer.com/chapter/10.100 (2/2)
1
1
Show this thread
This is Spike. He is officially the oldest living dog in the world, at a spry 23 years and 52 days young. That means he was born in 1999. Claims the secret to his longevity is frequent, uninterrupted naps. 14/10 #SeniorPupSaturday
read image description
ALT
read image description
ALT
read image description
ALT
read image description
ALT
247
4,464
48.8K
Everyone: I am afraid, AGI is coming, robots taking our jobs, creating art, etc, etc.
AGI:
1
4
26
Show this thread
This amazing Youtube channel is becoming even more awesome!
Go check it out if you like AI and Games, but somehow are not subscribed to this channel yet! 😎
Quote Tweet
Today is a new beginning.
2023 is a big year of change for 'AI and Games' on YouTube. To explain what's coming - and how you can help support it - check out the full video linked to here.
LINK: youtu.be/QcGF6zP8SsY
Let's write a 
on all the big changes.
1/20
on all the big changes.
1/20Show this thread
3
Demystifying Fourier analysis dsego.github.io/demystifying-f
8
100
498
I just learned that the first formula for calculating the n-th decimal digit of pi (without calculating all the preceding digits) has been found by Simon Plouffe in 2022 😲
495
7,533
62.9K
Show this thread
"Not seeing a tsunami or an economic event coming is excusable; building something fragile to them is not." - Nassim Nicholas Taleb in Antifragile
17
131
The Linux kernel, this is really cute!
12
497
1,932
Recorded another podcast episode about applied AI.
Love talking about this stuff.
If you have a show, and want me on, talking about machine learning, large AI models like ChatGPT, DALLE, etc -- or live questions from user questions -- would love to do more of these.
3
2
13
Show this thread
Amazing talk by on diving into and uncovering the mysteries behind (legacy) code!
So cool that records and uploads all talks! Unfortunately missed this talk live when I was there last December.
2
5
Excited to be working on NumerBlox again! 🔨 Some big compatability improvements on the roadmap!
Curious to hear from participants on what they are struggling with when it comes to (daily) model inference & deployment. Open to consider integrating it in NumerBlox!
Quote Tweet
We just released numerblox v0.5!
Biggest update is support for easily downloading the new @numerai v.4.1 (Sunshine) dataset. Check out the code snippet below to get started.
Repo:
github.com/crowdcent/nume
Docs:
crowdcent.github.io/numerblox
pip install numerblox
(1/3)
Show this thread
3
Me: Let's write programs where subtraction can be done with a "+" operator? 😎🤠
Common Lisp: Sure bro, why not? 😎
1
Lexica! 💕 Looking forward to see how this website will evolve in 2023!
Quote Tweet
The images on Lexica's home feed are now *significantly* better.
It's absolutely insane how far Stable Diffusion has come in 4 months. Here's to an even more hype 2023! 
lexica.art
lexica.art
0:13
49.9K views
3
Surprisingly little AI progress in 2023 so far. What’s going on??
134
372
5,216
Happy new year everyone and all the best for the new year! Let's go 2023!! 🔥🔥
2
7
Below also shows all combos for separate situations.
- Max for 1 dart is 50 (bullseye)
- Max for 2 darts is 110 (Triple 20 and bullseye)
- Max for 3 darts is 170 (2x Triple 20 and bullseye)
2
Show this thread
What do you call this distribution?! Beautiful! 👀😁😎
In a game of darts you have to either finish by throwing in a double section or in the bullseye. For low values you can use 1 or 2 throws. With <171 points to go you can finish a round (i.e. leg), which is called a checkout.
2
1
3
Show this thread
"creativity is just connecting things. when you ask creative people how they did something, they feel a little guilty because they didn't really do it, they just saw something. it seemed obvious to them after a while.”
11
17
164
Show this thread
"In my whole life, I have known no wise people who didn't read all the time -- none, zero. You'd be amazed at how much Warren reads--and at how much I read. My children laugh at me. They think I'm a book with a couple of legs sticking out"
--- Munger
1
37
252
Last line for the plotly plot 😅:
Show this thread
Also added copyable code blocks for all structures in the documentation so you can easily start to play around with this.
pip install -U blackscholes
Learn more about Iron Butterfly option structures:
carlolepelaars.github.io/blackscholes/6. (2/2)
1
Show this thread
Added 4 more option structures in blackscholes v0.1.2:
- (Bull and Bear) Spreads
- (Long and Short) Iron Butterfly
pip install -U blackscholes
Repository:
github.com/CarloLepelaars
Docs:
carlolepelaars.github.io/blackscholes
Below a visualization of a long iron butterfly near maturity. (1/2)
1
4
Show this thread
In what situations is NumPy faster than vanilla Python?
Short answer: if you want to perform the same operation on each item in a sequence, it's likely NumPy will be faster
Let's look at an example in which we compare using lists with append, list comprehensions, and NumPy
/1
read image description
ALT
8
43
242
Show this thread
The length of the Christmas tree lights to wrap around the tree perfectly
66
1,222
8,781
Whoops! I discovered a bug where a long iron condor was computed as a short iron condor and vice versa. This has been fixed in the most recent version (v0.1.1).
1
Show this thread
Example use case: evaluating the payoff of a long iron condor near maturity.
The highest payoff occurs from a large price move in either direction. Both losses and profits are bounded (assuming Black-Scholes pricing, position stays balances and no options are exercised, etc.).
Show this thread
pip install blackscholes
Repository:
github.com/CarloLepelaars
Learn more about option structures:
carlolepelaars.github.io/blackscholes/6 (3/3)
Show this thread
Would love to talk with quant/derivative traders about what functionality would be most helpful.
I'm thinking of integrating some visualizations to see the payoff of these option structures in various scenarios. Please let me know if are open to talk and/or contribute! (2/3)
1
1
Show this thread
blackscholes v0.1.0 is out! In this version I introduce 4 well-known option structures.
1. Straddle
2. Strangle
3. Butterfly
4. Iron Condor
Below an example of how you would retrieve attributes for a "Long Iron Condor", which consists of 2 call options and 2 put options. (1/3)
4
1
11
Show this thread
Practice makes perfect, I guess. Still needs a lot of practice!
| Lynyrd Skynyrd - Free Bird -> expert (original speed)
2