Discussing the article: "MiniRocket: A Deterministic Time-Series Classifier and What It Finds in Seven Classic Setups"
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: MiniRocket: A Deterministic Time-Series Classifier and What It Finds in Seven Classic Setups.
This article delivers a native MQL5 MiniRocket: 84 fixed convolution kernels yield 9,996 features quickly and deterministically, requiring no training loop and no external runtime. We verify the port against sktime and a float64 reimplementation, then run a reproducible audit of seven classic setups; planted and coin‑flip controls confirm correctness, and a 25%‑flipped control sets the detection threshold.
On this platform, "machine learning for time series" has come to mean a neural network, an ONNX export, and a Python dependency kept alive alongside the terminal. The wider literature has spent years showing that cost is not always necessary: methods built on fixed, untrained convolution kernels match deep models across the standard benchmark archives at a fraction of the compute.
This article ports one of those methods, MiniRocket, to native MQL5. It turns a window of prices into roughly ten thousand features using 84 fixed convolution kernels, then hands them to a linear model. There is no training loop, no gradient descent, no ONNX, and no Python at runtime, and one window takes about two milliseconds to transform.
A classifier solves only half the problem. The second half is the one nobody writes about: when a model finds nothing in a classic setup, you cannot tell whether the setup has no edge or your pipeline is broken. Because of that, null results rarely get published, and the same setups get argued about forever. We solve that directly, with controls built into the harness that prove the measurement works and show how weak an edge it can still detect. By the end you will have a verified transform, a leak-free audit harness, and a reproducible verdict on seven setups that you can re-run on your own symbol.
Author: Muhammad Minhas Qamar