Hacker News Reader

Your algorithm will likely not stand up to Walk Forward Analysis/Optimization/Testing: https://en.wikipedia.org/wiki/Walk_forward_optimization

I recommend reading "The Evaluation and Optimization of Trading Strategies" by Robert Pardo.

View on HN

It's really the most fundamental rule for every type of machine learning task -- there should be both a (1) test set and (2) a separate holdout validation set. No matter if the task is classification, sequence prediction, regression, manifold learning, clustering, whatever.

Really surprising how so few amateur traders understand this fundamental principal. Scientists too.

Reply on news.ycombinator.com