Dask-ML provides scalable machine learning in Python using Dask alongside popular machine learning libraries like Scikit-Learn, XGBoost, and others.
You can try Dask-ML on a small cloud instance by clicking the following button:
Dimensions of Scale¶
People may run into scaling challenges along a couple dimensions, and Dask-ML offers tools for addressing each.
Challenge 1: Scaling Model Size¶
The first kind of scaling challenge comes when from your models growing so large or complex that it affects your workflow (shown along the vertical axis above). Under this scaling challenge tasks like model training, prediction, or evaluation steps will (eventually) complete, they just take too long. You’ve become compute bound.
To address these challenges you’d continue to use the collections you know and
love (like the NumPy
DataFrame, or XGBoost
and use a Dask Cluster to parallelize the workload on many machines. The
parallelization can occur through one of our integrations (like Dask’s
joblib backend to parallelize Scikit-Learn directly) or one of
Dask-ML’s estimators (like our hyper-parameter optimizers).
Challenge 2: Scaling Data Size¶
The second type of scaling challenge people face is when their datasets grow larger than RAM (shown along the horizontal axis above). Under this scaling challenge, even loading the data into NumPy or pandas becomes impossible.
To address these challenges, you’d use Dask’s one of Dask’s high-level
(Dask Array, Dask DataFrame or Dask Bag) combined with one of Dask-ML’s
estimators that are designed to work with Dask collections. For example you
might use Dask Array and one of our preprocessing estimators in
dask_ml.preprocessing, or one of our ensemble methods in
Not Everyone needs Scalable Machine Learning¶
It’s worth emphasizing that not everyone needs scalable machine learning. Tools like sampling can be effective. Always plot your learning curve.
In all cases Dask-ML endeavors to provide a single unified interface around the familiar NumPy, Pandas, and Scikit-Learn APIs. Users familiar with Scikit-Learn should feel at home with Dask-ML.
Partner with other distributed libraries¶
Other machine learning libraries like XGBoost already have distributed solutions that work quite well. Dask-ML makes no attempt to re-implement these systems. Instead, Dask-ML makes it easy to use normal Dask workflows to prepare and set up data, then it deploys XGBoost alongside Dask, and hands the data over.
from dask_ml.xgboost import XGBRegressor est = XGBRegressor(...) est.fit(train, train_labels)
See Dask-ML + XGBoost for more information.