TensorBank

TensorBank is a collection of assorted algorithms expressed in Tensors so that they can be quickly executed on batched data in a variety of ML workflows.

Every function is supported in one or more of the following backends:

  • TensorFlow via import tensorbank.tf as tb

  • PyTorch via import tensorbank.torch as tb (currently, 0 functions are supported).

  • NumPy via import tensorbank.numpy as ntb (currently, 0 functions are supported).

Most funcions are designed to work with batched data to better utilize the GPU resources.

Here’s a simple example:

import tensorbank.tf as tb

tb.axis_aligned_boxes.area(
   [[1, 1, 2, 2],
    [-1, -1, 1, 2]])
>>> tf.Tensor([1 6], shape=(2,), dtype=int32)

Indices and tables