@trmadimak@TugserOkur Sadece training cutoff'dan sonraki out-of-sample veride değerlendirmek lazım. Bir de naive/random-walk, ARIMA, ETS. TimesFM'in bunları gerçekten geçmesi lazım ki çoğu zaman geçemiyor.
@trmadimak@TugserOkur Risk/sizing katmanı: position sizing, stop, drawdown kontrolü yok. Model belli bir tarihe kadarki veriyle eğitildiği için, o pencere içindeki tarihlerde backtest yaparsanız (yaptım çıldırdım) lookahead bias alırsınız, sonuç sahte güzel çıkıyor.
@kaaaash____ If you are asking what to learn, it is to learn how all moving parts are glued together. Contracts, architecture, type safety, CI/CD, git, modularity, security etc and integrating these to your vibe coding workflow with skills...
@AngusDurrie@md_kasif_uddin Vuhuuu it was 1994 for me with qbasic. I wrote a simple stock & accounting app with it. Then fortran 77 and turbo pascal days.
@neerajjj6785 AI vibe coding winner for now is Go. Because it is observable, less boilerplate, feedback loop is tight and fast. As the context length and memory problem is solved, java .net may be the king
@KaiXCreator The real skill isn’t 'coding' anymore—it’s building the infrastructure where AI can prove itself.
If your environment isn't a fully automated loop of local setup, automated git flows, and rigorous testing (Unit, Integration, Regression), AI is just guessing in the dark.