tenops 0.2.0

Python package for creating a common interface for using numpy, pytorch, etc.

Ten(sor) Op(eration)s

The basic idea behind this repository is to allow for seamlessly switching between libraries like numpy, pytorch, and tensorflow by checking the input types and returning the proper function for use in some operation.

Installation

Installation can be done over pip

pip install tenops

Note that for the library to work you'll need to have at least one of the following libraries installed:

• numpy
• torch
• tensorflow

Usage

Functions work like regular functions, one just needs to pass in the object and the function will identify the corresponding module and use the appropriate function:

>>> import torch, numpy, tensorflow as tf
>>> from tenops.special import exp
>>> exp(numpy.array([0]))
array([1.])
>>> exp(torch.tensor([0]))
tensor([1.])
>>> exp(tf.constant([0.]))
<tf.Tensor: shape=(1,), dtype=float32, numpy=array([1.], dtype=float32)>

One can also just specify the module using the default parameter rather than typecasting directly (note that this requires that the specified library is installed):

>>> from tenops.special import exp
>>> exp([0], default="numpy")
array([1.])
>>> exp([0], default="torch")
tensor([1.])
>>> exp([0.], default="tensorflow")
<tf.Tensor: shape=(1,), dtype=float32, numpy=array([1.], dtype=float32)>

Development

Poetry is used to manage the build process.