skjson 0.1.2

Export scikit-learn models to JSON for cross-language inference

skjson

Export scikit-learn models to JSON for cross-language inference.

Related repos

1. Running inference using npm package: skjson-js
2. Check out the demo

Installation

pip install skjson

Quick Start

Check out the demo notebook for an example training a Random Forest on the Iris dataset:

from sklearn.datasets import load_iris
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
import skjson

# Load the Iris dataset and split
X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Train a Random Forest classifier
clf = RandomForestClassifier(random_state=42)
clf.fit(X_train, y_train)

# Export the trained model to JSON
skjson.save(clf, 'demo.json')

# Load and reuse the model
predictor = skjson.load("demo.json")

Convert from Joblib and ONNX

skjson supports converting sklearn models directly from joblib and ONNX format without the need to load the model in Python first.

Convert from Joblib

import skjson

# Convert a joblib-serialized sklearn model to JSON
skjson.joblib_to_json("model.joblib")
# → Creates "model.json" in the same directory

# Or specify a custom output path and feature names
skjson.joblib_to_json(
    "model.joblib",
    output_path="output/model.json",
    feature_names=["sepal_length", "sepal_width", "petal_length", "petal_width"]
)

Convert from ONNX

Works with ONNX models exported from sklearn using skl2onnx:

import skjson

# Convert an ONNX model to JSON
skjson.onnx_to_json("model.onnx")
# → Creates "model.json" in the same directory

# Or specify a custom output path
skjson.onnx_to_json("model.onnx", output_path="output/model.json")

Supported ONNX model types (exported from sklearn via skl2onnx):

ONNX Operator	sklearn Model
LinearRegressor	LinearRegression, Ridge, Lasso, ElasticNet
LinearClassifier	LogisticRegression
TreeEnsembleClassifier	DecisionTreeClassifier, RandomForestClassifier
TreeEnsembleRegressor	DecisionTreeRegressor, RandomForestRegressor
SVMClassifier	SVC, LinearSVC
SVMRegressor	SVR

Supported models for conversion

Linear Models (sklearn.linear_model)

• LinearRegression, Ridge, Lasso, ElasticNet
• LogisticRegression

Tree-based Models (sklearn.tree, sklearn.ensemble)

• DecisionTreeClassifier, DecisionTreeRegressor
• RandomForestClassifier, RandomForestRegressor
• GradientBoostingClassifier, GradientBoostingRegressor

Support Vector Machines (sklearn.svm)

• SVC, SVR, LinearSVC

Neighbors (sklearn.neighbors)

• KNeighborsClassifier, KNeighborsRegressor

Naive Bayes (sklearn.naive_bayes)

• GaussianNB

Preprocessing (sklearn.preprocessing)

• StandardScaler, MinMaxScaler, LabelEncoder

Future Directions

• Support for scikit-learn Pipeline objects and feature unions.
• npm package for inference using our json models
• Support for sklearn.neural_network models

Note

• This project built with Google Gemini and Claude Opus.
• If you wish to contribute, please contact me.