|
|
d0785e12e2
|
Small changes to make the US weather dataset work properly, example test runs in config file.
|
2021-01-29 10:52:49 +01:00 |
|
|
|
e4c51e2d3d
|
Allow loss function to be defined by the dataset. Add specialized loss function for US weather dataset
|
2021-01-21 21:47:00 +01:00 |
|
|
|
f6a19c4921
|
Allow input shape to be defined by dataset, save loss values as csv after training, implemented basic version of US weather dataset, but it is very slow and has bad results probably due to input encoding issue
|
2021-01-20 21:48:31 +01:00 |
|
|
|
f76374111c
|
Add mssim score calculation, corrupt test data before testing, clip noise to avoid invalid values
|
2021-01-17 16:59:14 +01:00 |
|
|
|
bc95548ae3
|
Move saving of samples to dataset, as the process differs per dataset. Add MNIST dataset. Allow saving labels with the dataset (for use in tabular data in the future)
|
2021-01-14 18:45:26 +01:00 |
|
|
|
fb9ce46bd8
|
Implement types of auto-encoders and corruption, use log scale in loss graphs, lots of helper function hooks in training process to allow implementations
- Encoders: sparse, denoising, contractive and variational
- Noise: gaussian
|
2020-12-29 18:44:44 +01:00 |
|
|
|
62f9b873e9
|
First basic auto-encoder and CIFAR-10 dataset implemented
|
2020-12-28 13:09:18 +01:00 |
|
|
|
51cc5d1d30
|
Initial, very basic framework for running comparison tests
|
2020-11-24 17:19:46 +01:00 |
|
