want to automate your trading? i will show you exactly how to automate your trading from scratch even if you have no tech skills, fancy degree or any idea how to get started. in my algo trade camp i made short concise training videos, give you access to all of my code + invite you to my private quant community. join here before i close it to new entrants: algotradecamp.com/
Yo dude, 777. How have you been dude? I was wondering more about your GE journey. What inspired you to look towards a multipair GE? I've been seeing some GE models like to overfit on market changes. Like if I run one on oct 19 - oct 20 with 1h candles, a good strategy is to only make one short trade and catch the down trend in its entirety. The GE will essentially try to rig the system regardless of penalties enforcing more trades by making a reasonable sell signal and have the buy signal be total garbage that could never occur. So it finds these weird extended positions and makes like 1 trade when I give it around a year's worth of data. Have you come into contact with this and were you able to stop it from returning nonsense (from a live trading point of view)?
yo, 777! great to hear from u! the multipair GE idea came from needing something adaptable across market conditions without overfitting too much-totally get what u mean about GEs rigging the system when there’s only a narrow signal. yeah, i’ve seen that weird behavior where it tries to exploit a single setup for an entire year’s worth of data. adding regularization techniques and penalty adjustments can help, but it’s def an ongoing challenge to keep it from getting “too creative” with bad trades. r u experimenting with any methods to limit those extended positions?
@@moondevonyt 777 Yeah dude! It's crafty af. I tried penalizing, say, taking return I'd multiply it by min(#trades/#desired_trades, 1) for the fitness. I found it's really stubborn and will find its way to the same undesirable solutions as before. One possibility is that training over a year means training over changing market conditions and it's simply unable to find any consistent rules. Reducing to a 50/50 monthly split helped some with the GAs I tried, and the GE will sometimes make up to 5 trades in half a month if you work it that way. The issue I was having with that was seeing a pattern with a series of decent returns around 10.25% for a few months followed by massive failures around -50%. Another thing is it uses the OHLCV as an overfitting tool, since it can simply specify one direction of signals as some nonsense like open > 1000000. Removing might be good. Have you seen more desirable results from your multipair GEs?
yo, sounds like u've put in a ton of work refining this! curious tho-what made u decide to use a 50/50 monthly split over other timeframes? btw, if u wanna dive deeper into algo tuning without the hassle, check out my bootcamp at algotradecamp.com
want to automate your trading? i will show you exactly how to automate your trading from scratch even if you have no tech skills, fancy degree or any idea how to get started. in my algo trade camp i made short concise training videos, give you access to all of my code + invite you to my private quant community. join here before i close it to new entrants: algotradecamp.com/
Yo dude, 777. How have you been dude? I was wondering more about your GE journey. What inspired you to look towards a multipair GE? I've been seeing some GE models like to overfit on market changes. Like if I run one on oct 19 - oct 20 with 1h candles, a good strategy is to only make one short trade and catch the down trend in its entirety. The GE will essentially try to rig the system regardless of penalties enforcing more trades by making a reasonable sell signal and have the buy signal be total garbage that could never occur. So it finds these weird extended positions and makes like 1 trade when I give it around a year's worth of data. Have you come into contact with this and were you able to stop it from returning nonsense (from a live trading point of view)?
yo, 777! great to hear from u! the multipair GE idea came from needing something adaptable across market conditions without overfitting too much-totally get what u mean about GEs rigging the system when there’s only a narrow signal. yeah, i’ve seen that weird behavior where it tries to exploit a single setup for an entire year’s worth of data. adding regularization techniques and penalty adjustments can help, but it’s def an ongoing challenge to keep it from getting “too creative” with bad trades. r u experimenting with any methods to limit those extended positions?
@@moondevonyt 777 Yeah dude! It's crafty af. I tried penalizing, say, taking return I'd multiply it by min(#trades/#desired_trades, 1) for the fitness. I found it's really stubborn and will find its way to the same undesirable solutions as before. One possibility is that training over a year means training over changing market conditions and it's simply unable to find any consistent rules. Reducing to a 50/50 monthly split helped some with the GAs I tried, and the GE will sometimes make up to 5 trades in half a month if you work it that way. The issue I was having with that was seeing a pattern with a series of decent returns around 10.25% for a few months followed by massive failures around -50%. Another thing is it uses the OHLCV as an overfitting tool, since it can simply specify one direction of signals as some nonsense like open > 1000000. Removing might be good. Have you seen more desirable results from your multipair GEs?
yo, sounds like u've put in a ton of work refining this! curious tho-what made u decide to use a 50/50 monthly split over other timeframes?
btw, if u wanna dive deeper into algo tuning without the hassle, check out my bootcamp at algotradecamp.com
777 big bro ❤
777!
Moon Dev are you in Michigan?
not in michigan, but always cool to connect with ppl from all over! are u based there?