eugenehp/tabicl-regressor-v2-rs

TabICL regressor v2 — tabicl format

Converted weights for `tabicl-rs`. Mirrors the upstream PyTorch checkpoint Jingang/TabICL-reg (v2 release) in the layout the Rust state-dict loader consumes directly.

See MIT-RLX/tabicl-classifier-v2-rs for the classifier variant and a full description of the conversion pipeline; everything carries over modulo the regression head (quantile decoder instead of softmax).

Usage

use tabicl::{TabICLRegressor, TabICLConfig};

let mut reg = TabICLRegressor::new();
reg.fit(x_train.view(), &y_train)?;
reg.load_checkpoint(TabICLConfig::default(), "weights.safetensors")?;

let preds = reg.predict(x_test.view())?;

License

BSD-3-Clause, matching upstream soda-inria/tabicl.