Commit Graph

45 Commits

Author SHA1 Message Date
Mattéo Delabre 376c7c0bc4
Add illustration 2019-12-18 11:04:58 -05:00
Mattéo Delabre cacf4408e5
Add README 2019-12-18 11:02:03 -05:00
Mattéo Delabre af8862d360
Reflect parameter changes in `experiment.py` 2019-12-18 11:01:45 -05:00
Mattéo Delabre d8a140d793
Simplify arguments for ConstellationNet 2019-12-18 10:26:35 -05:00
Mattéo Delabre a7e9dd2230
Save whole model to avoid definition duplicate 2019-12-18 10:17:59 -05:00
Mattéo Delabre 0769a61fcf
Use best parameters as found by experimentation 2019-12-18 09:41:50 -05:00
Mattéo Delabre 5cb087d971
Add experiment results 2019-12-17 01:47:54 -05:00
Mattéo Delabre fe23fda86d
Shuffle configurations in experiment 2019-12-16 13:32:32 -05:00
Mattéo Delabre 8d64f91661
Add partitioning to experiment 2019-12-16 13:27:43 -05:00
Mattéo Delabre 0dfbed1bb7
Add experiment code 2019-12-16 12:59:39 -05:00
Mattéo Delabre 7efcbd3948
Add product_dict util function 2019-12-16 11:32:25 -05:00
Mattéo Delabre e8e4dcec83
Set learning rate to 10^-3 2019-12-16 10:31:29 -05:00
Mattéo Delabre af59421dc0
Fix variable naming clash 2019-12-16 10:31:16 -05:00
Mattéo Delabre fb2518b321
Show final loss after training 2019-12-16 10:20:01 -05:00
Mattéo Delabre 1d39184036
Use euclidean distance for stop criterion 2019-12-16 09:55:06 -05:00
Mattéo Delabre 4f1514accf
Add noise to plot 2019-12-16 09:22:24 -05:00
Mattéo Delabre 2c4786bdba
Remove reconstruction examples after training 2019-12-16 09:17:15 -05:00
Mattéo Delabre 4bf0c0f363
Add batch normalization and ReLU activation 2019-12-16 02:35:13 -05:00
Mattéo Delabre 8fa6b46ca8
Change training strategy to adaptive learning rate 2019-12-16 02:30:33 -05:00
Mattéo Delabre 3f2c6d18a3
Manually seed training for reproducibility 2019-12-16 01:50:03 -05:00
Mattéo Delabre 4d8d419d31
Only plot noise if noise_samples > 0 2019-12-16 01:49:35 -05:00
Mattéo Delabre 2b93a5f1bc
Fix wrong input dimension for decision region plot 2019-12-15 23:53:14 -05:00
Mattéo Delabre e4457400a6
Add larger set of colors for plot 2019-12-15 23:52:57 -05:00
Mattéo Delabre 722c02ef24
Cleanup channel model 2019-12-15 23:12:00 -05:00
Mattéo Delabre 3638749998
Share plotting code between plot.py and train.py 2019-12-15 20:05:20 -05:00
Mattéo Delabre 34cb3b863b
Simplify channel model to account for normalized power 2019-12-15 19:48:49 -05:00
Mattéo Delabre 59d7adf6bd
Add convergence criterion plot during train 2019-12-15 19:42:50 -05:00
Mattéo Delabre 3c199dfc41
Remove activation function from ConstellationNet 2019-12-15 19:22:35 -05:00
Mattéo Delabre 989a51b72e
Increase batch size halfway through training 2019-12-15 09:42:48 -05:00
Mattéo Delabre 3365603614
Add normalization layer 2019-12-15 09:42:33 -05:00
Mattéo Delabre b97ba61f42
Improve plot legibility 2019-12-15 02:07:52 -05:00
Mattéo Delabre 8a31b22b83
Increase noise stddev 2019-12-15 01:21:11 -05:00
Mattéo Delabre 04c2938b9e
Decrease alpha for noise plot 2019-12-15 01:21:03 -05:00
Mattéo Delabre 9a87b322d4
Plot noise 2019-12-15 01:07:05 -05:00
Mattéo Delabre 6ea0e653c1
Balance training examples 2019-12-15 00:59:10 -05:00
Mattéo Delabre 99c96162c0
Use SELU instead of ReLU 2019-12-15 00:57:31 -05:00
Mattéo Delabre 6d31ab3a13
Use ReLU with normalization 2019-12-15 00:03:20 -05:00
Mattéo Delabre 197a01e993
Fix plot scaling 2019-12-15 00:03:02 -05:00
Mattéo Delabre 8f6363ee21
Plot decision regions from the decoder 2019-12-14 23:31:47 -05:00
Mattéo Delabre 3b40e27070
Add Gaussian channel model 2019-12-14 23:04:35 -05:00
Mattéo Delabre d0af8fc3da
Use Tanh activation to preserve negative values 2019-12-14 23:03:02 -05:00
Mattéo Delabre a01d83f339
Improve plot 2019-12-13 17:10:40 -05:00
Mattéo Delabre 49e63775dd
Save trained models and plot encoding 2019-12-13 15:17:57 -05:00
Mattéo Delabre 2af8354a07
Add initial implementation 2019-12-13 12:11:09 -05:00
Mattéo Delabre 7463e9fe5b
Initial commit 2019-12-11 19:13:40 -05:00