HSIC (Hilbert-Schmidt Independence Criterion) answers a core algo-trading question: do features truly relate to the target, beyond out-of-sample hope. Unlike Pearson correlation, HSIC detects both linear and non-linear dependence using kernel similarity matrices, working with scalars or high-dimensional vectors without explicit joint distribution modeling.
The article implements HSIC in MQL5 with an RBF (Gaussian) kernel, including practical sigma selection via a median-distance heuristic and efficient distance computation in matrix form. Significance is handled two ways: an exact permutation test (accurate but costly) and a gamma approximation (much faster, best for large datasets).
A synthetic trading-style example shows HSIC recovering a non-linear dependency where correlations are near zero, making it useful for feature screening and subgroup testing before...
👉 Read | Docs | @mql5dev
#MQL5 #MT5 #HSIC
The article implements HSIC in MQL5 with an RBF (Gaussian) kernel, including practical sigma selection via a median-distance heuristic and efficient distance computation in matrix form. Significance is handled two ways: an exact permutation test (accurate but costly) and a gamma approximation (much faster, best for large datasets).
A synthetic trading-style example shows HSIC recovering a non-linear dependency where correlations are near zero, making it useful for feature screening and subgroup testing before...
👉 Read | Docs | @mql5dev
#MQL5 #MT5 #HSIC
❤52👍4✍3🤯3👌2👀2⚡1