What's the point of machine learning?
- Thread starter barkbark
- Start date
Hello barkbark,I've read a few books, and watched youtube videos, but maybe I'm just too retarded to figure it out.
I wish I can help you on this machine learning stuff, but I have no clue on that subject.
But, in my keep-it-simple opinion if I have to read a few books and watch a few videos to show me how to make money trading, and I still do not understand the subject, then I am perfectly fine with giving up and get back to starring at the chart.
In my keep-it-simple opinion, starring at the chart in real time and clicking buy and sell or writing down trading ideas to test quickly back test, its far more efficient then reading a trading book.
I have read several trading books, and watch countless youtube videos on trading, and still poor ASF in trading. But MANNNN, let me tell you something, when I stare at the darn chart and click buy or sell and program some strategies and click back test button, man I feel so rich ASF and alive.
Thank you so MUCH CME for inventing the ES futures market for all to have fun.
%%
OK / i did, good charts.
I like more machine printed + hand printed charts than he used.
That Asian David Chang or whatever his name had some better charts with Dr Christian Kacher; but reading just a few books would be like just a few chapters of Proverbs/ read all 31.
Most traders are not retarded, even if they lose money like a retard.
Another key\ is when you lose an account down to ''absolute nuthin'';
its sure not the 200dma, machine learning fault \ not the cats fault.
Stuff happens, my best friend's cat/ jumped up on my shoulders= claw spiked me abit/LOL.
Funny now but not then= good thing for me i rolled up paper like WSJ or IBD + slapped that cat down good
Good thing for me i had studied mountain lions+ tigers; i could have gotten killed[absolute nuthin ]with them.Am i comparing the market to a tiger??
Not really, its more like female friend, kitten , kitty cat+ sometimes a mountain lion
You are telling your age. I was in the military during Vietnam - when that song came out.
Machine learning is just modelling an outcome to variables. An outcome can be lineair or categoric. Machine learning is just modelling these outcomes from your given variables, and "the machine" returns you the best model. Basically when correlation is high your model will be well fitted to the data. This is always "known data", the past. Models cannot measure causality. The problem you had is that you modelled the past data but the correlation break with future data because they are not causal, or have changed.
Btw ANN's are also build upon lineair and logestic regression functions.
Btw ANN's are also build upon lineair and logestic regression functions.
What do you think of "causal AI" then?
Regression in machine learning is predicting a continuous value as opposed to classification, identifying which category something belongs to.
So you are correct you would use it to try to predict the value of the next step.
You can see all the various flavors of regression on the scikit-learn page
https://scikit-learn.org/stable/supervised_learning.html#supervised-learning
If the system is too good to be true you are probably leaking data if using closing prices. If you are predicting the close of SPY but the model already knows the close of QQQ for that day it is like giving the model a crystal ball that will only work in backtesting. That is an obvious example but there could be much less obvious things like that from the correlation of features with what you are trying to predict.
A random forest regressor can show the feature importance for what contributed to the prediction, that can be interesting in and of itself. A linear model is a good starting point though and then see if something else can do better than the linear model.
So you are correct you would use it to try to predict the value of the next step.
You can see all the various flavors of regression on the scikit-learn page
https://scikit-learn.org/stable/supervised_learning.html#supervised-learning
If the system is too good to be true you are probably leaking data if using closing prices. If you are predicting the close of SPY but the model already knows the close of QQQ for that day it is like giving the model a crystal ball that will only work in backtesting. That is an obvious example but there could be much less obvious things like that from the correlation of features with what you are trying to predict.
A random forest regressor can show the feature importance for what contributed to the prediction, that can be interesting in and of itself. A linear model is a good starting point though and then see if something else can do better than the linear model.
Interesting, will look it up! Was unknown with this field.
With to post i was reffering to the 'standard' ml models, svm, knn etc, fwiw
%%
a] Looks like some good stuff in there....trees;
not that any trees are random. But like you implied, forest = > greater than some trees.
b ]All markets are somewhat alike, but not exactly @ all. So an observer could learn a lot
LR] LR is like a train track, maybe ok for a starting point;
but a real life train track is not straight for long, but straight ======,=== for a while.
Most of the train money money made off[outside ] a train track ; not made by the engineer+ labor though his job is to keep it between the lines/LOL. Good engineers make $77-150,000annulized.....
I've made more off a 200dma + a railroad track top than LR; but not get rich quick like many daytraders try to do.
Most any indicator alone is worth''absolutely nuthin ''; + a train=wreck with wrong size or too much snow or trees on track
You should start with the basics, it's pretty clear you still lack some of the basic understanding of regression analysis and basic ML concepts. Before taking giant steps you need to be solid on the small steps.
I've been mainly playing with the linear regression algorithm in python, but I don't really understand the purpose of it.
I get backtesting. You have a trading strategy with entries and exits, and you see how many times it wins and loses and how much profit you get from it, and so on.
But I can't figure out how to apply linear regression to that. Anytime I put something together it looks really profitable, but I know that the rules I used don't work in real life. On top of that, attempting to ascertain the profitability of the rules, doesn't really seem to work in the way backtesting does.
Some video made it seem like it's supposed to predict the next/future value of the set, given the parameters. In which case I'm just going about it the wrong way.
I've read a few books, and watched youtube videos, but maybe I'm just too retarded to figure it out.