46 lines
1.9 KiB
Markdown
46 lines
1.9 KiB
Markdown
 | # BOSS
| ||
| |||
Bag-of-SFA-Symbols classifier written from scratch in pure MQL5. Turns price
| |||
windows into words, and bags of words into a regime classifier.
| |||
| |||
Companion code for the MQL5 article: https://www.mql5.com/en/articles/23491
| |||
| |||
## What it does
| |||
| |||
Symbolic Fourier Approximation converts a price window into a short string:
| |||
the window goes through a Fourier transform, the low coefficients are kept, and
| |||
each is binned into a letter. Similar shapes become the same word, and noise
| |||
that does not change the shape drops out.
| |||
| |||
BOSS then counts words rather than comparing series point by point. That is
| |||
what makes it fast, and it is why one classifier is not enough: a single window
| |||
length sees one time scale, so the ensemble spans several.
| |||
| |||
Regimes are labelled without hand-labelling, which matters because hand-labels
| |||
are where this kind of study usually goes wrong.
| |||
| |||
Benchmarked against Dynamic Time Warping on BTCUSD, the ensemble wins on clean
| |||
accuracy and runs roughly twenty times faster. `BOSSvsDTWBenchmark.mq5`
| |||
reproduces that comparison, and `BOSSNoiseSweep.mq5` is the noise robustness
| |||
sweep.
| |||
| |||
## Layout
| |||
| |||
```
| |||
Include/BOSS/SFA.mqh Symbolic Fourier Approximation
| |||
Include/BOSS/FourierTransform.mqh the transform behind SFA
| |||
Include/BOSS/BOSS.mqh bag-of-words model and ensemble
| |||
Include/BOSS/RegimeLabeler.mqh unsupervised regime labels
| |||
Indicators/BOSS/BOSSRegime.mq5 the regime indicator
| |||
Scripts/BOSS/BOSSSelfTest.mq5 correctness checks
| |||
Scripts/BOSS/BOSSvsDTWBenchmark.mq5 accuracy and speed against DTW
| |||
Scripts/BOSS/BOSSNoiseSweep.mq5 robustness under added noise
| |||
```
| |||
| |||
Run `BOSSSelfTest.mq5` first. If it fails, nothing downstream is worth reading.
| |||
| |||
## Disclaimer
| |||
| |||
Educational code. Past behaviour of any model or dataset says nothing about
| |||
future results. Test on your own data and broker conditions before drawing
| |||
conclusions.
|