piskle 0.0.1

Piskle allows you to selectively and efficiently serialize scikit-learn models to save on memory and load times.

Piskle

Piskle allows you to selectively serialize python objects to save on memory and load times.

It has special support for exporting scikit-learn's models in an optimized way, exporting exactly what's needed to make predictions.

via GIPHY

Example:

To use piskle, you first need a model to export. You can use this as an example:

from sklearn import datasets
from sklearn.neural_network import MLPClassifier

data = datasets.load_iris()

model = MLPClassifier().fit(data.data, data.target)

Exporting the model is then as easy as the following:

import piskle

piskle.dump(model, 'model.pskl')

Loading it is even easier:

model = piskle.load('model.pskl')

If you want even faster serialization, you can disable the optimize feature. Note that this feature reduces the size of the exported file even further and improves loading time.

piskle.dump(model, 'model.pskl', optimize=False)

Currently Supported Models

Predictors ( Classifiers, Regressors, ...)

Estimator	Reference
LinearSVC	sklearn.svm
LinearRegression	sklearn.linear_model
LogisticRegression	sklearn.linear_model
Lasso	sklearn.linear_model
Ridge	sklearn.linear_model
Perceptron	sklearn.linear_model
GaussianNB	sklearn.naive_bayes
KNeighborsRegressor	sklearn.neighbors
KNeighborsClassifier	sklearn.neighbors
MLPClassifier	sklearn.neural_network
MLPRegressor	sklearn.neural_network
DecisionTreeClassifier	sklearn.tree
DecisionTreeRegressor	sklearn.tree
KMeans	sklearn.cluster
GaussianMixture	sklearn.mixture

Transformers

Estimator	Reference
PCA	sklearn.decomposition
FastICA	sklearn.decomposition
CountVectorizer	sklearn.feature_extraction.text
TfidfVectorizer	sklearn.feature_extraction.text
SimpleImputer	sklearn.impute
StandardScaler	sklearn.preprocessing
LabelEncoder	sklearn.preprocessing
OneHotEncoder	sklearn.preprocessing

Future Improvements

This is still an early working version of piskle, there are still a few improvements planned:

• More thorough testing
• Version Management: Support for more versions of scikit-learn (earlier versions)
• Support for more Estimators (Feel free to contact us for a specific request)
• Support for "Nested" Estimators (Pipelines, RandomForests, etc...)

Contribute

As this is still a work in progress, while using piskle, you might encounter some bugs. It would be a great help to us, if you could report them in the github repo.

Feel free, to share with us any potential improvements you'd like to see in piskle.