popsom7 - Release 7.1.2

This release fixes a fatal bug in the sklearn API of the package

Full Changelog: https://github.com/lutzhamel/popsom7/compare/v7.1.1...v7.1.2

- R
Published by lutzhamel about 1 year ago

popsom7 - Release 7.1.1

Fixed a fatal bug in map_build in the Python release.

- R
Published by lutzhamel over 1 year ago

popsom7 - Release 7.1.0

This release reintroduced our fast, vectorized training algorithm for SOM with substantial improvements. In R it is about an order of magnitude faster than the canonical, stochastic C implementation of the training algorithm.

In the Python version worked on the quality of the generated map.

What's Changed

• Vsom algorithm integration by @lutzhamel in https://github.com/lutzhamel/popsom7/pull/25
• Merging 7.1.0 changes into main by @lutzhamel in https://github.com/lutzhamel/popsom7/pull/26

Full Changelog: https://github.com/lutzhamel/popsom7/compare/v7.0.0...v7.1.0

- R
Published by lutzhamel over 1 year ago

popsom7 - Release 7.0.0

This release now includes a Python version R(version 7.0.0) Python(version 7.0.0)

What's Changed

• Popsom7 by @lutzhamel in https://github.com/lutzhamel/popsom7/pull/20
• edits by @lutzhamel in https://github.com/lutzhamel/popsom7/pull/21

Full Changelog: https://github.com/lutzhamel/popsom7/compare/v6.0...v7.0.0

- R
Published by lutzhamel over 1 year ago

popsom7 - Release 6.0

• Renamed the functions in the interface to avoid collisions with S3 functions within the R environment. We know that this is another renaming of the POPSOM interface and we apologize for any inconvenience. We expect that the interface is now stable for the foreseeable future.
• New features:
  • The map.minimal object. This is an object that only contains the trained neurons and nothing else. This is an appropriate model when POPSOM is used as a preprocessing step and no other model information is needed. Note that map.minimal objects cannot be processed by any of the other functions in the POPSOM interface.
  • The map.convergence function provides details about the underlying convergence characteristics.
• Bugfixes
  • Most importantly the artificial limit of a minimum of 50 instances in the training data has been removed.

- R
Published by lutzhamel over 4 years ago

popsom7 - Release 5.2

Reworked the description of the package in order to reflect the capabilities of the package better.

- R
Published by lutzhamel almost 5 years ago

popsom7 - Release 5.1

• Something got rattled with the S3 interface in R 4.x. It no longer works the way it did in release 3.x. Therefore, I took the S3 interface out because I want the package to work with both 3.x and 4.x installations. Furthermore, the advantages of the S3 interface are incremental at best and I don't feel like debugging R internals.

• Implemented a 'summary' function for map objects.

- R
Published by lutzhamel over 5 years ago

popsom7 - Release 5.0

Popsom 5.0 is a complete rarchitecting of the package. It includes the following: - Support for two models: 1. A self-organizing map model 2. A centroid based clustering model - Quality measures available for both models - Streamlined S3 based API - Easy access to the most important map and centroid data structures - Powerful map visualization with centroid identification - Extremely fast training algorithm based on ideas from tensor algebra

- R
Published by lutzhamel over 5 years ago

popsom7 - Release 4.3.1

Fixed code with the deprecated functions.

- R
Published by lutzhamel over 5 years ago

popsom7 - Release 4.3.0

This release introduces a decaying alpha value for the VSOM training algorithm fixing convergence problems at high learning rates. This release also deprecates a number of interface objects in order to get ready for the next release.

- R
Published by lutzhamel almost 6 years ago

popsom7 - Release 4.0.1

fixed Fortran cross-platform issues

- R
Published by almost 10 years ago

popsom7 - Release 4.0 (VSOM)

The biggest change in the 4.0 release of popsom is the inclusion of a vectorized version of the stochastic SOM training algorithm. This new training algorithm runs up to 10 times faster than the batch algorithm and between 50 to 100 times faster than the traditional stochastic training algorithm. Of course the precise numbers depend strongly on the kind of problem you are working on.

The SOM quality reporting functions have been made consistent with our recent publications: see http://homepage.cs.uri.edu/faculty/hamel/pubs/

- R
Published by almost 10 years ago

popsom7 - popsom 3.0.1

This is the current release in R, June 2016

- R
Published by about 10 years ago