compiler

X-TIME Compiler

The compiler is exposed both as a command line interface and a Python library that binds to its native core written in Rust. It implements the conversion of tree-based machine learning models trained using various Python libraries to the Analog CAM mapping of the X-TIME architecture. Supported libraries are Scikit-Learn, XGBoost, CatBoost and LightGBM.

• Building the CLI
• Building the Python Library
• Usage in Python

Building the CLI

Install Rust:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh  # select "default" (just press enter)
source ~/.cargo/env

Build the binary:

cargo build --release

This will place the xtimec binary inside the target/release folder

To automatically place it in your $PATH, you can alternatively use cargo install:

cargo install --path .

Usage of the command is as follows:

Usage: xtimec [OPTIONS] <INPUT_FILE> <OUTPUT_FILE>

Arguments:
  <INPUT_FILE>   Path to the model JSON dump
  <OUTPUT_FILE>  Path to the resulting NumPy file

Options:
  -t, --type <MODEL_TYPE>  The type of model that is passed [default: treelite] [possible values: treelite, catboost]
  -h, --help               Print help
  -V, --version            Print version

Building the Python Library

Install Rust:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh  # select "default" (just press enter)
source ~/.cargo/env

You should now have the cargo command available.

Create and activate a virtual environment:

python3 -m venv .venv
source .venv/bin/activate

Install Maturin:

pip install maturin

pip install -U pip

Build the project:

maturin build --release

This will place a .whl file that you can pip install inside of the target/wheels folder.

Alternatively, you can use:

maturin develop

to enter a development shell with the library and all its dependencies available.

Note: The resulting Python wheel files include a native library that is linked against the Python version available on the system at build time. This means that the compiler library won't work with e.g. Python 3.10 if it was built against Python 3.9.

Usage in Python

Example:

from sklearn.datasets import make_classification
import xgboost

from xtimec import XTimeModel

# create a random dataset
X, y = make_classification(n_samples=100, n_informative=5, n_classes=2)

# train a model
model = xgboost.XGBClassifier(max_depth=3, n_estimators=5, max_bin=256)
model.fit(X, y)
booster = model.get_booster()

# represent the model encoded for an analog CAM
# alternatively `from_xgboost`, `from_catboost` etc.
xmodel = xtimec.XTimeModel.from_xgboost(booster)

# >>> xmodel.cam
#   np.array([[...]])
# >>> xmodel.leaves
#   np.array([[...]])

See python/xtimec/model.py for the entire Python API.