Mars leverages parallel and distributed technology to accelerate numpy, pandas, scikit-learn and Python functions.
There are four main APIs in Mars:
Mars tensor, which mimics numpy API and provide ability to process large tensors/ndarrays.
Mars DataFrame, which mimics pandas API and be able to process large DataFrames.
Mars learn, which mimics scikit-learn API and scales machine learning algorithms.
Mars Remote, which provide the ability to execute Python functions in parallel.
Mars is lazy evaluated by default,
.execute() is required to perform computation,
however, eager mode is supported as well,
if eager mode is on, execution will be triggered
every time when each tensor, DataFrame, and so forth is created.
Mars can leverage NVIDIA GPU to accelerate computation.