← Research logMachine Learning
KILLED · no improvement
#

Self-attention captures long-range dependencies a recurrent net misses; a Transformer should beat the LSTM (N-038) on directional forecasting.

Attention re-weights value vectors by query–key similarity:

$$ \text{Attn}(Q,K,V) = \mathrm{softmax}\!\Big(\frac{QK^\top}{\sqrt{d_k}}\Big)V $$

Encoder-only Transformer, same feature set and walk-forward protocol as N-038, tuned over depth/heads.

Direction accuracy50.9%
vs LSTM baseline≈ identical
Overfitting on small crypto historysevere
KILLED
A better function approximator on a target with no signal gives a better fit to noise, not better forecasts. No improvement over the LSTM, and worse overfitting given limited history. Killed.
Model capacity helps only when there is structure to capture. On near-random targets, more capacity buys more overfitting, not more edge.

We publish the failures too.

This is one of 100+ documented hypotheses. Browse the full lab notebook, or see the strategies that survived into production.