sklearn-export

This package is based on sklearn porter from https://github.com/nok/sklearn-porter. I choose to build it because sklearn porter saves data in matrix format. However, most popular algebra libraries (e.g., blas and lapack) are used to work with vectors. Then, sklearn-export saves the sklearn model data in Json format (matrices are stored in column major order). Note that, this is a beta version yet, then only some models and functionalities are supported.

New features (0.0.7)

The code was optimized and now it works with sklearn >= 0.24. Some complete examples were added (see Complete Examples section).

Support

Observation: details were extracted from sklearn documentation.

Installation

We recommend to make a instalation using pip:

$ pip install sklearn_export

If you are using jupyter notebooks consider to install sklearn_export through a notebook cell. Then, you can type and execute the following:

import sys
!{sys.executable} -m pip install sklearn_export

Usage

Actually sklearn-export can save Classifiers, Regressions and some Scalers (see Support session).

Saving a Model or Scaler

The basic usage is to save a simple model.

# Basic imports
from sklearn.datasets import load_iris
from sklearn_export import Export
from sklearn.neural_network import MLPRegressor

# Load data and train model
samples = load_iris()
X, y = samples.data, samples.target
clf = MLPRegressor()
clf.fit(X, y)

# Save using sklearn_export
export = Export(clf)
result = export.to_json()

The result is a Json file that can be loaded in any language.

Complete examples

Some complete examples are provided here.

Saving a Model and a Scaler

The sklearn-export can also save more then one class in the same Json. This is usefull to store a Classifier and a Scaler (for example). To be honest, actually is only possible to store a pair Model and Scaler.

# Basic imports
from sklearn.datasets import load_iris
from sklearn_export import Export
from sklearn.preprocessing import StandardScaler
from sklearn.neural_network import MLPRegressor

# Load data
samples = load_iris()
X, y = samples.data, samples.target

# Normalize data
scaler = StandardScaler()
Xz = scaler.fit_transform(X)

# Train model with normalized data
clf = MLPRegressor()
clf.fit(Xz, y)

# Save model and scaler using sklearn_export
export = Export([scaler, clf])
result = export.to_json()

The result is a Json file that contains information about a Model and a Scaler. The file can be loaded in any programing language.

Extra options

The method to_json() also support some other parameters:

Questions

If you have any question please send me a mail charles26f@gmail.com.