creme 0.0.1

Incremental machine learning in Python

creme is a library for incremental learning. Incremental learning is a machine learning regime where the observations are made available one by one. It is also known as online learning, iterative learning, or sequential learning. This is in contrast to batch learning where all the data is processed at once. Incremental learning is desirable when the data is too big to fit in memory, or simply when it isn't available all at once. creme's API is heavily inspired from that of scikit-learn, enough so that users who are familiar with scikit-learn should feel right at home.

Useful links

• Documentation
• Example notebooks
• Issue tracker
• Releases

Installation

pip install creme

Quick example

In the following snippet we'll be fitting an online logistic regression. The weights of the model will be optimized with the AdaGrad algorithm. We'll scale the data so that each variable has a mean of 0 and a standard deviation of 1. The standard scaling and the logistic regression are combined using a pipeline. We'll be using the stream.iter_sklearn_dataset function for streaming over the Wisconsin breast cancer dataset. We'll measure the ROC AUC using progressive validation.

>>> from creme import linear_model
>>> from creme import model_selection
>>> from creme import optim
>>> from creme import pipeline
>>> from creme import preprocessing
>>> from creme import stream
>>> from sklearn import datasets
>>> from sklearn import metrics

>>> X_y = stream.iter_sklearn_dataset(
...     load_dataset=datasets.load_breast_cancer,
...     shuffle=True,
...     random_state=42
... )
>>> optimizer = optim.AdaGrad()
>>> model = pipeline.Pipeline([
...     ('scale', preprocessing.StandardScaler()),
...     ('learn', linear_model.LogisticRegression(optimizer))
... ])
>>> metric = metrics.roc_auc_score

>>> model_selection.online_score(X_y, model, metric)
0.992977...

Comparison with other solutions

• scikit-learn: Some of it's estimators have a partial_fit method which allows them to update themselves with new observations. However, online learning isn't a first class citizen, which can make it a bit awkward to put a streaming pipeline in place. You should definitely use scikit-learn if your data fits in memory and that you can afford retraining your model from scratch when you have new data to train on.
• Vowpal Wabbit: VW is probably the fastest out-of-core learning system available. At it's core it implements a state-of-the-art adaptive gradient descent algorithm with many tricks. It also has some mechanisms for doing active learning and using bandits. However it isn't a "true" online learning system as it assumes the data is available in a file and can looped over multiple times. Also it is somewhat difficult to grok for newcomers.
• Spark Streaming: This is an extension of Apache Spark which caters to big data practitioners. It provides a lot of practical tools for manipulating streaming data in it's true sense. It also has some compatibility with the MLlib for implementing online learning algorithms, such as streaming linear regression and streaming k-means. However it is a somewhat overwhelming solution which might be a bit overkill for certain use cases.
• TensorFlow: Deep learning systems are in some sense online learning systems. Indeed it is possible to put in place a DL pipeline for learning from incoming observations. Because frameworks such as Keras and PyTorch are popular and well-backed, there is no real point in implementing neural networks in creme. For a lot of problems neural networks might not be the right tool, and you might want to use a simple logistic regression or a decision tree (for which online algorithms exist).

Feel free to open an issue if you feel like other solutions are worth mentioning.

Development

creme is very young so there is a lot to do. The broad goals for the near future are to:

• implement simple but useful algorithms
• identify bottlenecks and use Cython when possible
• write good documentation and write example notebooks
• make life easier for those who want to put a streaming pipeline in production

License

See the license file.