NEWS

codez 2.0.0 (2026-07-17)

Migrated the neural backend from TensorFlow/Keras to torch.
Added fit_codez() and codez_control() as the v2 workflow API.
Added tidy forecast extraction via codez_forecast_tidy() and S3 helpers for predict(), summary(), autoplot(), plot(), and as.data.frame().
Added rolling naive and average baseline backtests for numeric data.
Added training controls for batch_size, validation_split, and n_sim.
Fixed mixed-type detection, forward-net L2 sampling, finite-solution routing, duplicate plotter() definition, and average benchmark calculation.
Replaced non-core convenience dependencies with local support functions for mapping, dummy encoding, interpolation, smoothing, metrics, distances, moments, entropy, array binding/splitting, timing, and plotting.