The Cat
@thecattrader.bsky.social
I’m a Cat. I collect ticks in the futures market.
Indicator builder & discretionary trader.
9 lives, all speculative
ML-Curious
Nothing here is financial advice. Only speculation.
Indicator builder & discretionary trader.
9 lives, all speculative
ML-Curious
Nothing here is financial advice. Only speculation.
Pinned
The Cat
@thecattrader.bsky.social
· Aug 29
PSA: I don’t DM about ‘signals’ or ‘mentorship.’ If a Cat is promising you riches in Discord, it’s not me.
Just finished the experiment...
And the results are better than I expected...
And the results are better than I expected...
November 19, 2025 at 5:12 AM
Just finished the experiment...
And the results are better than I expected...
And the results are better than I expected...
Late night for me.
Trying to invalidate my work so hard and its just... Air tight...
I found structural issues but nothing that leaks the Agent future information.
With that said; the 5% (issues) is fixed and Im going to go sit, be beside myself for a moment, and think.
Trying to invalidate my work so hard and its just... Air tight...
I found structural issues but nothing that leaks the Agent future information.
With that said; the 5% (issues) is fixed and Im going to go sit, be beside myself for a moment, and think.
November 17, 2025 at 7:15 AM
Late night for me.
Trying to invalidate my work so hard and its just... Air tight...
I found structural issues but nothing that leaks the Agent future information.
With that said; the 5% (issues) is fixed and Im going to go sit, be beside myself for a moment, and think.
Trying to invalidate my work so hard and its just... Air tight...
I found structural issues but nothing that leaks the Agent future information.
With that said; the 5% (issues) is fixed and Im going to go sit, be beside myself for a moment, and think.
Reposted by The Cat
It's a revealing moment...
J.D. Vance welcomes a two-legged robot helping on the job site. But if that helper is a person named Jose, he calls it a threat. Same tasks, same productivity boost.
The inconsistency reveals this isn’t about economics.
J.D. Vance welcomes a two-legged robot helping on the job site. But if that helper is a person named Jose, he calls it a threat. Same tasks, same productivity boost.
The inconsistency reveals this isn’t about economics.
November 16, 2025 at 12:52 PM
It's a revealing moment...
J.D. Vance welcomes a two-legged robot helping on the job site. But if that helper is a person named Jose, he calls it a threat. Same tasks, same productivity boost.
The inconsistency reveals this isn’t about economics.
J.D. Vance welcomes a two-legged robot helping on the job site. But if that helper is a person named Jose, he calls it a threat. Same tasks, same productivity boost.
The inconsistency reveals this isn’t about economics.
Hahaha thanks windows update... there were only 7 combinations left 😑
November 16, 2025 at 3:03 PM
Hahaha thanks windows update... there were only 7 combinations left 😑
We have "tapped the repo sign" (SRF) twice in about two months...
The strain in the plumbing is there.
Not trying to be an alarmist
But it does mean the probability of a violent repricing is MUCH higher than the average person realizes.
#econsky
The strain in the plumbing is there.
Not trying to be an alarmist
But it does mean the probability of a violent repricing is MUCH higher than the average person realizes.
#econsky
November 15, 2025 at 3:38 PM
We have "tapped the repo sign" (SRF) twice in about two months...
The strain in the plumbing is there.
Not trying to be an alarmist
But it does mean the probability of a violent repricing is MUCH higher than the average person realizes.
#econsky
The strain in the plumbing is there.
Not trying to be an alarmist
But it does mean the probability of a violent repricing is MUCH higher than the average person realizes.
#econsky
It's like building a house... but you dont know the dimensions inside and cant walk inside to measure it...
So you change radar settings in order to "feel" how big it is and how big it SHOULD be.
This is what its like to sweep knobs on your ML research project...
So you change radar settings in order to "feel" how big it is and how big it SHOULD be.
This is what its like to sweep knobs on your ML research project...
a dog is sitting at a desk using an apple laptop computer
Alt: a dog is sitting at a desk using an apple laptop computer
media.tenor.com
November 15, 2025 at 5:11 AM
It's like building a house... but you dont know the dimensions inside and cant walk inside to measure it...
So you change radar settings in order to "feel" how big it is and how big it SHOULD be.
This is what its like to sweep knobs on your ML research project...
So you change radar settings in order to "feel" how big it is and how big it SHOULD be.
This is what its like to sweep knobs on your ML research project...
Reposted by The Cat
The last penny has been minted. I joined CNN's Brianna Keilar to dive into what it all means. There's actually a whole lot of economics -- and politics! -- to this story. If the news today feels too heavy, this is a bit lighter. Come and have fun with us!
www.youtube.com/watch?v=332A...
www.youtube.com/watch?v=332A...
The Last Penny Has Been Minted... And Justin Wolfers Analyzes What it All Means
CNN News Central, October 12, 2025.
Why cling to a coin that buys nothing and clogs every pocket in America?
Justin Wolfers and CNN's Brianna Keilar unpack the real economics behind ending penny…
www.youtube.com
November 13, 2025 at 11:17 PM
The last penny has been minted. I joined CNN's Brianna Keilar to dive into what it all means. There's actually a whole lot of economics -- and politics! -- to this story. If the news today feels too heavy, this is a bit lighter. Come and have fun with us!
www.youtube.com/watch?v=332A...
www.youtube.com/watch?v=332A...
The deviation between 2 seeds of my newly calibrated instruments is 0.5-1.0.
Which means im on track, but would like to shave that down to 0.5 and lower between seeds.
Will test more seeds tonight to see how close I am to having the components calibrated before looking at Hyperparameters.
Which means im on track, but would like to shave that down to 0.5 and lower between seeds.
Will test more seeds tonight to see how close I am to having the components calibrated before looking at Hyperparameters.
November 13, 2025 at 10:24 PM
The deviation between 2 seeds of my newly calibrated instruments is 0.5-1.0.
Which means im on track, but would like to shave that down to 0.5 and lower between seeds.
Will test more seeds tonight to see how close I am to having the components calibrated before looking at Hyperparameters.
Which means im on track, but would like to shave that down to 0.5 and lower between seeds.
Will test more seeds tonight to see how close I am to having the components calibrated before looking at Hyperparameters.
Data, features, and config files are now fingerprinted and are callable.
This makes it easy to train on a specific hash or test with select features or bundles that I want to experiment on.
This makes it easy to train on a specific hash or test with select features or bundles that I want to experiment on.
November 13, 2025 at 1:02 AM
Data, features, and config files are now fingerprinted and are callable.
This makes it easy to train on a specific hash or test with select features or bundles that I want to experiment on.
This makes it easy to train on a specific hash or test with select features or bundles that I want to experiment on.
Me optimizing my training pipeline
a man in a police uniform is sitting at a desk typing on a laptop computer .
Alt: a man in a police uniform is sitting at a desk typing on a laptop computer .
media.tenor.com
November 12, 2025 at 9:02 PM
Me optimizing my training pipeline
Also worth thinking about hashing data, features and training configs...
November 11, 2025 at 6:05 PM
Also worth thinking about hashing data, features and training configs...
Im to the point where I have to automate a lot of boring things.
There are component weights and smoothing that i need to sweep through values to find sweet spots.
Im just too lazy to do it myself...
Job scheduler it is
There are component weights and smoothing that i need to sweep through values to find sweet spots.
Im just too lazy to do it myself...
Job scheduler it is
November 11, 2025 at 5:34 PM
Im to the point where I have to automate a lot of boring things.
There are component weights and smoothing that i need to sweep through values to find sweet spots.
Im just too lazy to do it myself...
Job scheduler it is
There are component weights and smoothing that i need to sweep through values to find sweet spots.
Im just too lazy to do it myself...
Job scheduler it is
Reposted by The Cat
Facts would like a word.
November 11, 2025 at 2:31 AM
Facts would like a word.
I could look at these all day
November 10, 2025 at 10:34 PM
I could look at these all day
Going to add some noise variation to the transaction cost chained to volatility and volume to simulate overcoming the reward variation from slippage and liquidity.
Im also thinking about adding a small delay from the RL Agents order and execution, but haven't quite figured it out yet.
Im also thinking about adding a small delay from the RL Agents order and execution, but haven't quite figured it out yet.
November 10, 2025 at 9:53 PM
Going to add some noise variation to the transaction cost chained to volatility and volume to simulate overcoming the reward variation from slippage and liquidity.
Im also thinking about adding a small delay from the RL Agents order and execution, but haven't quite figured it out yet.
Im also thinking about adding a small delay from the RL Agents order and execution, but haven't quite figured it out yet.
Going to continue to be risk-off... the risk-event isnt over just yet.
November 10, 2025 at 4:33 AM
Going to continue to be risk-off... the risk-event isnt over just yet.
And squashed the exploit. The RL Agent is learning a policy that actually works on unseen data.
Aka I did it.
Its a REAL start to a real baseline model for Market Stucture Forcasting.
Aka I did it.
Its a REAL start to a real baseline model for Market Stucture Forcasting.
November 9, 2025 at 7:05 PM
And squashed the exploit. The RL Agent is learning a policy that actually works on unseen data.
Aka I did it.
Its a REAL start to a real baseline model for Market Stucture Forcasting.
Aka I did it.
Its a REAL start to a real baseline model for Market Stucture Forcasting.
Well my assumptions are so bad for this environment...
The RL Agent says "i made big number! Exactly how you told me to. Ive made more GDP than the world and now operate in Space Bucks"
And im just like
"Well.. no.. not like that"
Im laughing crying
youtube.com/shorts/mrmqR...
#econsky
The RL Agent says "i made big number! Exactly how you told me to. Ive made more GDP than the world and now operate in Space Bucks"
And im just like
"Well.. no.. not like that"
Im laughing crying
youtube.com/shorts/mrmqR...
#econsky
When your program does as it’s programmed to do
YouTube video by Devslopes
youtube.com
November 9, 2025 at 5:28 PM
Well my assumptions are so bad for this environment...
The RL Agent says "i made big number! Exactly how you told me to. Ive made more GDP than the world and now operate in Space Bucks"
And im just like
"Well.. no.. not like that"
Im laughing crying
youtube.com/shorts/mrmqR...
#econsky
The RL Agent says "i made big number! Exactly how you told me to. Ive made more GDP than the world and now operate in Space Bucks"
And im just like
"Well.. no.. not like that"
Im laughing crying
youtube.com/shorts/mrmqR...
#econsky
Trying to come up with maths...
The RL Agent figured out how to game the environment...
Which I cant actually be mad about.. I wrote the evnironment, so its kind of on me.
This is kind of what I have 1/2
The RL Agent figured out how to game the environment...
Which I cant actually be mad about.. I wrote the evnironment, so its kind of on me.
This is kind of what I have 1/2
November 9, 2025 at 5:30 AM
Trying to come up with maths...
The RL Agent figured out how to game the environment...
Which I cant actually be mad about.. I wrote the evnironment, so its kind of on me.
This is kind of what I have 1/2
The RL Agent figured out how to game the environment...
Which I cant actually be mad about.. I wrote the evnironment, so its kind of on me.
This is kind of what I have 1/2
Oops all features...
These are all discretionary trader features like RSI, ATR, EMAs etc.
Ive recently added more machine readable features, that actually do better. Will share when ive ran more experiments on them.
#econsky
These are all discretionary trader features like RSI, ATR, EMAs etc.
Ive recently added more machine readable features, that actually do better. Will share when ive ran more experiments on them.
#econsky
November 8, 2025 at 8:39 PM
Oops all features...
These are all discretionary trader features like RSI, ATR, EMAs etc.
Ive recently added more machine readable features, that actually do better. Will share when ive ran more experiments on them.
#econsky
These are all discretionary trader features like RSI, ATR, EMAs etc.
Ive recently added more machine readable features, that actually do better. Will share when ive ran more experiments on them.
#econsky
Okay google, how do i escape refactor hell...
November 8, 2025 at 7:59 PM
Okay google, how do i escape refactor hell...
Testing out fractals with Markov Switching Modeling.
Will probably drop two of these states
I spent most of yesterday just generating different feature heatmaps, most features are just plain useless or the model "Kind Of" does on its own (Like EMAs)
#Econsky
Will probably drop two of these states
I spent most of yesterday just generating different feature heatmaps, most features are just plain useless or the model "Kind Of" does on its own (Like EMAs)
#Econsky
November 5, 2025 at 7:44 PM
Testing out fractals with Markov Switching Modeling.
Will probably drop two of these states
I spent most of yesterday just generating different feature heatmaps, most features are just plain useless or the model "Kind Of" does on its own (Like EMAs)
#Econsky
Will probably drop two of these states
I spent most of yesterday just generating different feature heatmaps, most features are just plain useless or the model "Kind Of" does on its own (Like EMAs)
#Econsky
Thinking about adding more features soon, pruning redundant ones and keeping whats necessary for the model to learn.
November 4, 2025 at 9:09 PM
Thinking about adding more features soon, pruning redundant ones and keeping whats necessary for the model to learn.
#econsky
Mmm numbers... But important numbers.
I'll be able to prune completely cold features (like bIs_Doji)
Its interesting how a Doji means something to discretionary trader, but to the model its like "Oh nah useless"
Probably just adding noise at this point.
Mmm numbers... But important numbers.
I'll be able to prune completely cold features (like bIs_Doji)
Its interesting how a Doji means something to discretionary trader, but to the model its like "Oh nah useless"
Probably just adding noise at this point.
November 2, 2025 at 5:33 PM
#econsky
Mmm numbers... But important numbers.
I'll be able to prune completely cold features (like bIs_Doji)
Its interesting how a Doji means something to discretionary trader, but to the model its like "Oh nah useless"
Probably just adding noise at this point.
Mmm numbers... But important numbers.
I'll be able to prune completely cold features (like bIs_Doji)
Its interesting how a Doji means something to discretionary trader, but to the model its like "Oh nah useless"
Probably just adding noise at this point.
Im going to start openly sharing visual results from the ML research.
Expect to see... a lot of Heatmaps.
Currently coding feature bundles to give the models mathematical understanding of the data it has.
OHLC were only numbers until I gave maths on how to use those numbers for direction etc
Expect to see... a lot of Heatmaps.
Currently coding feature bundles to give the models mathematical understanding of the data it has.
OHLC were only numbers until I gave maths on how to use those numbers for direction etc
November 2, 2025 at 5:24 PM
Im going to start openly sharing visual results from the ML research.
Expect to see... a lot of Heatmaps.
Currently coding feature bundles to give the models mathematical understanding of the data it has.
OHLC were only numbers until I gave maths on how to use those numbers for direction etc
Expect to see... a lot of Heatmaps.
Currently coding feature bundles to give the models mathematical understanding of the data it has.
OHLC were only numbers until I gave maths on how to use those numbers for direction etc