Jensun Ravichandran
8227525c82
Add LambdaLayer
2021-05-31 16:47:20 +02:00
Jensun Ravichandran
e61ae73749
Make components dynamic
2021-05-31 00:31:40 +02:00
Alexander Engelsberger
040d1ee9e8
Add probabilistic losses
...
Based on Soft LVQ paper by Seo and Obermayer
2021-05-28 20:38:50 +02:00
Alexander Engelsberger
7f0da894fa
Add transformation from distances into gaussian distribution
2021-05-28 16:50:04 +02:00
Alexander Engelsberger
62726df278
Add stratified sum as competition
...
For example used in RSLVQ
2021-05-28 16:49:39 +02:00
Alexander Engelsberger
0ba09db6fe
Bump version: 0.4.5 → 0.5.0
2021-05-28 16:17:49 +02:00
Alexander Engelsberger
87334c11e6
Remove Prototypes1D and its tests
2021-05-28 16:17:49 +02:00
Alexander Engelsberger
40ef3aeda2
Remove usage of Prototype1D
...
Update Iris example to new component API
Update Tecator example to new component API
Update LGMLVQ example to new component API
Update GTLVQ to new component API
2021-05-28 16:17:40 +02:00
Christoph
94fe4435a8
Bump version: 0.4.4 → 0.4.5
2021-05-27 09:58:25 +02:00
Alexander Engelsberger
c204bc8e1f
integrate reviews from ChristophRaab:master
2021-05-27 09:43:02 +02:00
Alexander Engelsberger
00615ae837
refactored gtlvq from ChristophRaab:master
2021-05-27 09:40:42 +02:00
Jensun Ravichandran
9f5f0d12dd
[BUGFIX] Parse dictionary distribution appropirately
2021-05-25 20:52:39 +02:00
Jensun Ravichandran
8a291f7bfb
Overload distribution argument in component initializers
...
The component initializers behave differently based on the type of the
`distribution` argument. If it is a Python
[list](https://docs.python.org/3/tutorial/datastructures.html ), it is assumed
that there are as many entries in this list as there are classes, and the number
at each location of this list describes the number of prototypes to be used for
that particular class. So, `[1, 1, 1]` implies that we have three classes with
one prototype per class. If it is a Python
[tuple](https://docs.python.org/3/tutorial/datastructures.html ), it a shorthand
of `(num_classes, prototypes_per_class)` is assumed. If it is a Python
[dictionary](https://docs.python.org/3/tutorial/datastructures.html ), the
key-value pairs describe the class label and the number of prototypes for that
class respectively. So, `{0: 2, 1: 2, 2: 2}` implies that we have three classes
with labels `{1, 2, 3}`, each equipped with two prototypes.
2021-05-25 20:05:29 +02:00
Alexander Engelsberger
21e3e3b82d
Cache pip in CI
2021-05-25 16:43:48 +02:00
Alexander Engelsberger
a6bd6e130a
Add subpackages into prototorch namespace.
2021-05-25 16:40:53 +02:00
Alexander Engelsberger
fcdfa52892
Ignore artiifacts folder
2021-05-25 16:40:34 +02:00
Alexander Engelsberger
73e6fe384e
Use 'num_' in all variable names
2021-05-25 15:57:05 +02:00
Alexander Engelsberger
aff7a385a3
Use dict for distribution
...
This change allows the use of LightningCLI.
2021-05-21 17:10:02 +02:00
Jensun Ravichandran
1e23ba05fa
Add test_components
2021-05-21 16:22:02 +02:00
Alexander Engelsberger
ee30d4da5b
[BUGFIX] Initializers can handle Dataloaders now
2021-05-21 16:00:20 +02:00
Alexander Engelsberger
14508f0600
[DOC] Small improvements
2021-05-21 15:59:44 +02:00
Jensun Ravichandran
e3f8828da4
Accept dataloaders for component initialization
2021-05-21 11:59:57 +02:00
Jensun Ravichandran
30adbf705c
Update dependencies
2021-05-20 11:44:53 +02:00
Jensun Ravichandran
ee42fd68b1
NumpyDataset now has data and targets properties
2021-05-18 19:38:40 +02:00
Jensun Ravichandran
736d9a6349
Rename PositionAwareInitializer to DataAwareInitializer
...
Also, add the aliases `Zeros` and `Ones`.
2021-05-18 19:37:25 +02:00
Alexander Engelsberger
0055e15bc1
[DOC] Fix iris data dimension
2021-05-18 18:57:03 +02:00
Alexander Engelsberger
b2e1df7308
Improve dataset documentation.
2021-05-18 18:54:43 +02:00
Jensun Ravichandran
b935e9caf3
Update _get_dp_dm
2021-05-18 13:09:11 +02:00
Jensun Ravichandran
503ef0e05f
Cleanup components
2021-05-17 16:58:57 +02:00
Jensun Ravichandran
dc6248413c
Apply transformations in component initializers
2021-05-17 16:58:22 +02:00
Jensun Ravichandran
e73b70ceb7
Minor aesthetic change
2021-05-17 16:57:41 +02:00
Jensun Ravichandran
639198e774
Update Iris dataset
2021-05-17 16:57:13 +02:00
Alexander Engelsberger
768d969f89
Device agnostic initialization of components.
2021-05-13 15:21:04 +02:00
Alexander Engelsberger
aec422c277
Remove copy paste error from documentation.
2021-05-13 11:56:38 +02:00
Jensun Ravichandran
6c14170de6
[BUGFIX] Fix typo
2021-05-12 16:31:22 +02:00
Jensun Ravichandran
36a330aa66
Update component initializers
2021-05-12 16:28:55 +02:00
Jensun Ravichandran
acd4ac6a86
Flatten tensors before computing distances
2021-05-12 16:28:34 +02:00
Jensun Ravichandran
abe64cfe8f
Merge pull request #3 from dmoebius-dm/dev
...
Removed wrong parameter.
2021-05-12 16:23:27 +02:00
Danny
caae95d01d
Removed wrong parameter.
2021-05-12 16:00:01 +02:00
Alexander Engelsberger
088429a16a
Bump version: 0.4.3 → 0.4.4
2021-05-11 17:17:49 +02:00
Jensun Ravichandran
b6145223c8
[HOTFIX] Add missing iris.py and fix knnc bug
2021-05-11 17:20:48 +02:00
Alexander Engelsberger
09256956f3
Bump version: 0.4.2 → 0.4.3
2021-05-11 17:04:08 +02:00
Jensun Ravichandran
0ca90fdcee
Merge branch 'dev' of github.com:si-cim/prototorch into dev
2021-05-11 17:07:04 +02:00
Jensun Ravichandran
be21412f8a
Add thin wrapper for the Iris dataset
2021-05-11 17:06:41 +02:00
Jensun Ravichandran
ae6bc47f87
[BUGFIX] Fix knnc
2021-05-11 17:06:27 +02:00
Jensun Ravichandran
7bb93f027a
Support for unequal prototype distributions
2021-05-11 16:11:11 +02:00
Alexander Engelsberger
bc20acd63b
Bump version: 0.4.1 → 0.4.2
2021-05-11 16:08:37 +02:00
Alexander Engelsberger
a864cf5d4d
Bump version: 0.4.0 → 0.4.1
2021-05-11 13:37:54 +02:00
Alexander Engelsberger
2175f524e8
Update bug report issues template.
2021-05-11 13:35:38 +02:00
Alexander Engelsberger
c1c21e92df
Add LVQ 1 and LVQ 2.1 loss functions.
2021-05-11 13:25:10 +02:00
