What is Shapelets-Compute?¶
Within Shapelets software stack, Shapelets-Compute is a collection of algorithms and computational facilities to work efficiently with time series represented as dense arrays.
The server side version of Shapelets, which deals with complexities like storage, indexing, ingestion, etc…, exposes a super set of the API exposed by Shapelets-Compute, but with a very different aim: In Shapelets, algorithms are transformed into a DAG (directed acyclic graph), describing a distributed computation; the leaf nodes of such graphs are operations physically implemented by worker nodes, who uses Shapelets-Compute to execute operations and instructions over sequences of data.
Therefore, Shapelets-Compute offers you the same computational semantics as Shapelets server side, with the benefit of running it as open source library, directly in your machine without complex deployments nor the need to acquire server resources.
Shapelets-Compute should help you during your investigation phase, where versatility and computational efficiency allows to try new analytical pipelines as fast as possible.
When defining an API suitable to data scientist and data engineers, we wanted to reuse some of the existing semantics of libraries like NumPy, to reduce as much as possible the learning curve; at the same time, we set ourselves an objective to be as efficient as possible in terms of execution time, maximizing the facilities offered by from multicore hosts, OpenCL and CUDA devices.
Therefore, whilst examining Shapelets-Compute, you will see that:
ArrayFire is used internally as the underlying driver for most of our algorithms; this allows you to choose among different backends and devices to accelerate your computations without having to change a single like of code.
Shapelets-Compute goes at great length to offer a similar API to NumPy, creating a seamless experience when coding new algorithms to run in either Shapelets-Compute or Shapelets server side.
ShapeletsArrays are our internal abstraction to work with dense array and tensors. The compatibility with the rest the Python infrastructure and libraries is achieved by implementing Buffer Protocol on top of ArrayFire abstractions.
Follow the installation guide
Play with some examples and tutorials
Read about array operations.