Catboost inference

There are some Catboost Rust crates, but they're based on bindings to the C++ API and handle both training and inference, which makes them highly complicated to build and use. This is a simple library that just handles inference.

To use, save your catboost classifier to JSON, like so (python):

classifier.save_model(
    "my-model",
    format="json",
)

Then use it from Rust like so:

use catboost::Catboost;
use std::path::Path;

let model = CatBoost::load(Path::new("my-model.json")).unwrap();
let test_features: Vec<f32> = vec![0.1276993, 0.9918129, 0.16597846, 0.98612934];
let probability = model.predict(&test_features).unwrap();

Note: this library does not currently support categorical features. (Only float features are supported.) But categorical feature support would probably be pretty simple to add. Leave an issue if that's something you're interested in.