mastodon.world is one of the many independent Mastodon servers you can use to participate in the fediverse.
Generic Mastodon server for anyone to use.

Server stats:

8.1K
active users

#tensor

0 posts0 participants0 posts today

tensors arent really real
tensors are just vectors
covectors are just vectors

you can just interpret a vector as a tensor, which basically involves using different bases
what makes a covector a covector is that you use the "original" basis (actually the dual basis) for them
you can just as easily treat covectors as normal vectors with a normal basis (made up of covectors)

#ReleaseWednesday 🚀 — I wonder how many other FLOSS devs are sitting on code for ~8 years prior to first release... In one of these cases (many others readily available in my stash 🙃), triggered by recent major updates to the thi.ng/vectors library, I've refactored (almost 100% rewritten) and applied the same approach to the new/old package:

thi.ng/tensors

This package provides 1D/2D/3D/4D tensors, supporting different storage implementations (currently still all CPU side only) and an extensible set of polymorphic tensor operations (currently ~45 math ops, incl. matrix-matrix/matrix-vector products, reductions, argmin/max, activation functions etc.). The tensor classes themselves also provide several zero-copy slicing, re-ordering, clipping, extraction functions, most of them type-safe.

The original (private) version was heavily reliant on dynamic code generation, which has now been replaced with higher-order functions to provide various dimension-optimized versions of all operations.

This package is NOT specifically aimed at machine learning, even though it could probably used for some tasks in that realm (likely with extra hand holding). There are many other use cases for this kind of data structure...

Also new in other packages in this release cycle (incl. some code examples):

- docs.thi.ng/umbrella/arrays/fu
- docs.thi.ng/umbrella/bidir-ind
- docs.thi.ng/umbrella/transduce

thi.ng/tensors1D/2D/3D/4D tensors with extensible polymorphic operations and customizable storage

Can higher order tensors be represented by a matrix? People think you cannot, but I say you can. A 3x3x3 #tensor can be represented by a 3x9 matrix easily. Here is how.

In fact a 9x9 matrix is symmetric and has much better properties than a 3x3x9 tensor. And we can always represent any tensor of any order using a matrix. Tensor and matrix are nothing more than two different "accounting" systems of keeping a record of a data set containing many variables. #mathematics #physics #AI

Google, could we get the ability to drag and drop icons and widgets to different home screen pages without it saying "no room on this home screen"? Think a little bit like how apple handles it - if you drag an icon or a widget to a filled up home screen, it just pushes the icons and widgets there to the next page over. Thank you.