VOID

The counter-factual is untestable. It can only be inferred through loosely correlated statistics.

Personal favorite (go-to) ramen: Shin Black
Most over-rated crowd favorite: Buldak Carbonara

If you have time, add soft boiled egg, fried garlic, spring onion, sesame oil, and gochugaru (coarse and/or fine)... ground beef or pork if you'd like. (or any subset of these)

Good laws are Lindy.

From another perspective, this basically demonstrates that the true minimum wage is zero. Some people will spend their time doing things the other people don't value.

primary source research is expensive. the government takes your tax dollars to do this all the time. you find it abusrd, but this is basically how much grants cost to prove/demonstrate/discover... something that seems inconsequential.

the properly educated, credentialed, qualified 10%... of course.

claim: half the people don't like something.
counter-claim: but 10% love it.

assertion: those 10% prove that the other 50%'s opinions are irrelevant.

typical argument i see on this platform.

ye

*slur* slur *slur*

and tomorrow, you delete, because you were mocking some *slur-ish* person who *slurred* their way into infamy.

mockery is a valid form of critique. but the forever protocol doesn't rate you that way.

imagine impulse-posting, only to almost immediately delete.. but the open protocol allows for your worst takes to be promoted, even after your knowledgebase changed.

if you're not self-promoting, you're self-incriminating.

What does this mean? Is the quoted content blocked because:
1. Original poster blocked the quoter?
2. Original poster blocked me?
3. Original poster disassociated itself from the original context?
4. I don't know. Please describe.

This is the platform of "let them cook!"

But it's also the platform of "I don't like this so this person is a bigot" and their intentions are evil... which kind of negates the "let them cook" approach.

today is thanksgiving in the united states. lots of people are cooking. some are using old hardware to cook their evening takes.

this platform cooks those who go above lukewarm.

This feels very much like the dystopia in Black Mirror's Nose Dive, where you can only interact with and review people relative to your social level. Hyper-Pareto. Explore and exploit. Except, it's a flat game, no sleep, no shelter - with sniper dens; a social Squid Game.

well... @deck.blue solved this. wild! Thank you!

Happy thanksgiving to all!

Screenshot of deck.blue column settings. The first column is the default Home column, where I include my own posts (filter own posts = off), and include replies and reposts. The second column is a modified Home column, where I exclude my own posts, replies and reposts, making it a "post only" column from those I follow.

A tip for @deck.blue users: You can add your Home column multiple times with different settings! 💡 I have the default Home with replies and reposts (which can be hard to follow during active hours), and a "posts only" Home that makes it easier to see and engage with new posts from those I follow 😃

all this needs is the self-loading, user-stream that replicates tweetdeck... and i'd be here 90% vs ...

That's the intentional spelling of the account, of same name and posting style on the other platform. (face cropped for semi-consistency)

next 40 handles? lower. 5950. oh wait, i wrote that 8 hours ago.. lmao.. nm. i'm alive... keeping the seat warm, but i'm not committed to stove-pipes of moralistic hypocrisy.

You may not like it, but this is what peak performance looks like.

Performance != Growth

(yeah, when i say "we," i'm referring to the more skilled part of me, and then just me, the observer.)

So, we trained 2 models on the new target variable, and the results are acceptable. I don't like how the predict_proba score is less than 0.2, but anything above 0.01 is a signal.

Total downtime was 4 hours. The SATA cable failed. fsck found about a dozen files in /tmp that were partially written. No data loss.

Computer crashed. Disk corruption. Had to download a live disk to run fsck to rewrite the inode...

Still hoping for luck, bc right now, the disk is fsck'ed.

Follow-up: Model scores (prediction) on out-of-band data. You can see how it (under) performs. We'll try another target variable variant for the same purpose. Note: My legacy high/low of day models do generalize better with random forest, as opposed to xgboost (used here).

Target variables - version 1 and version 2. I have 3 target labels: (long, short, and no-bias).

After training, there will be noise at important events (e.g. secondary low/high-of-day), even though v2 only targets the official lod/hod.

Mean reversion day was awesome. Models nailed it. 2 model stopouts from smol, and 5 solid signals.

If only I had better resolution data... oh well. I'm scheming a new target variable. Hope to have it ready by Monday.

After a bunch of fussing around with xgboost parameters, trying to trick it into incremental improvements on batches of data, I officially give up. Performance severely degrades with incremental training. Back to strategic down-sampling.

Was at jury duty and basically just bought 4497 in premarket and set a limit to sell at 4519. Everything was fine when I left, so I set a stop at break even... and got stopped out.

4497 was July 12 low around noon, which aligned with June 30 high (practically end of quarter close).

Arbitrage signal got the 10am low. Mean reversion got the AM high. Price action was slow, so I expected market to suck in more bulls by putting price back to lows. AAPL was a disaster.