Quick grid tuning for tabular pipeline
tune_ml_backtest(
features_list,
labels,
prices,
fit_fn,
predict_fn,
schedule = list(is = 104L, oos = 4L, step = 4L),
grid = list(top_k = c(10L, 15L), temperature = c(8, 12), method = c("softmax", "rank"),
transform = c("zscore")),
group = "pooled",
selection_defaults = list(top_k = 15L, max_per_group = NULL),
weighting_defaults = list(method = "softmax", temperature = 12, floor = 0),
caps = list(max_per_symbol = 0.08),
group_map = NULL,
cost_bps = 0,
freq = 52
)data.table with metrics per grid row.
List of feature panels.
Label panel.
Price panel used for backtests (Date + symbols).
Model fit and predict functions.
List with elements is, oos, step.
list of vectors: top_k, temperature, method, transform.
Grouping mode for roll_fit_predict ('pooled'/'per_symbol'/'per_group').
Default selection settings (e.g., top_k).
Default weighting settings (e.g., method, temperature).
Exposure caps (e.g., max_per_symbol/max_per_group).
Optional Symbol->Group mapping.
optional one-way cost in basis points for net performance.
re-annualization frequency (e.g., 52).