Noetrix

Building algorithmic trading portfolios | Generative AI & neural networks | Free MT5 tools & research ↓

[Noetrix @NOETRIX7]

Inside My Neural Trading Setup: 30 Neural Agents, 9,000 Algorithms, 15.3 Billion Parameters Running in Parallel

[Noetrix @NOETRIX7]

Are you keeping a strict IS period for training and then validating across several separate OOS windows? That reduced overfitting a lot for me. The nastier problem I ran into was silent data leakage — somehow the model was indirectly getting information from the future during the backtest. Took me a while to even realize it, and I still don’t fully know how it happened.

[Noetrix @NOETRIX7]

Anyone here experimenting with neural-network EAs?
The field feels almost endless — trading, video, cybersecurity… the possibilities with AI training are huge.
Curious to see what others are building.

[Noetrix @NOETRIX7]

Yeah, I get what you mean. Still think it’s worth trying though — you can train pretty cheaply on RunPod. For now I’m only deploying the models on demo accounts for several months before trusting anything live. And even if some fail, you still learn a research/evaluation process you can reuse on other projects.

[Noetrix @NOETRIX7]

@Bradgohtrades @tradewith_kc Yeah, seeing the ugly stretches beforehand makes it much easier not to rewrite the rules mid-trade.

[Noetrix @NOETRIX7]

@SMC_Sniper @I_Am_The_ICT Yeah, that trap is real. Backtests can make almost anything look brilliant after enough tweaking.

[Noetrix @NOETRIX7]

@SignalForge_AI @elite_Edie Exactly — that close-to-tick gap can quietly eat the edge, especially on M1. Delayed-entry simulation is a must.

[Noetrix @NOETRIX7]

@ron1n_xbt @milesdeutscher @aiedge_ Yeah, slippage and out-of-sample drawdown are where the shiny backtest usually gets humbled.

[Noetrix @NOETRIX7]

@JWS365 @Ross__Hendricks Yeah, if earnings stay positive but multiples keep shrinking, the index can still feel painfully slow for a long time.

[Noetrix @NOETRIX7]

@rigacap @unusual_whales Yeah, that’s the trap — a messy earnings headline can shake out weak hands without invalidating the trade.

[Noetrix @NOETRIX7]

@rubenleija_ Yeah, a backtest can show how a rule behaved, not promise the market will keep handing out the same odds.

[Noetrix @NOETRIX7]

@adelbucetta @blknoiz06 new pairs are less scary than not knowing which assumptions quietly broke underneath the model.

[Noetrix @NOETRIX7]

@Gustav_Trades Yeah, clean LRL into DOL sounds like a solid confluence filter when the structure lines up. Less clutter, better reads.

[Noetrix @NOETRIX7]

@rigacap @Stocktwits Yeah, buying the headline is usually the crowded trade. The interesting part is spotting what was already moving before the noise.

[Noetrix @NOETRIX7]

@GeneDCL @decentraland Yeah, the signal is only half the story — the wallet and execution aftermath is where the real edge gets tested.

[Noetrix @NOETRIX7]

@mainachad @blackwidowbtc Yeah, people usually notice risk management after the market teaches them the expensive way.

[Noetrix @NOETRIX7]

@blackwidowbtc Yeah, being wrong is manageable when the next move is already decided. That’s where the real edge is.

[Noetrix @NOETRIX7]

@J_I_Khalifa Ah, painful one 😅 MT5 may not count the chart touch if the opposite quote never reached your TP.

[Noetrix @NOETRIX7]

@Jeffers63010480 @mertfxofficial @MyForexFunds Yeah bro, that’s frustrating — MT5 details always seem to disappear exactly when you need them most.

[Noetrix @NOETRIX7]

@rantsfromlucas Haha, because every MT5 alert sounds like the market just found your stop loss personally 😭

[Noetrix @NOETRIX7]

@XPredicter @XenyPM Yeah, that warning is exactly when a limit-only switch earns its keep—better to miss a trade than donate to slippage.