 René Knaebel
|
b3d646c9e7
|
add group lasso regularizer impl
|
2017-11-30 09:34:45 +01:00 |
|
René Knaebel
|
f382d06eb5
|
add initial version of sluice network with alphas, betas, and soft share
|
2017-11-27 16:17:19 +01:00 |
|
René Knaebel
|
349bc92a61
|
reorder curve storing
|
2017-11-13 14:37:41 +01:00 |
|
René Knaebel
|
d58dbcb101
|
refactor class weights
|
2017-11-10 14:31:32 +01:00 |
|
René Knaebel
|
461d4cab8f
|
add deeper domain cnn; refactor hyperband using load_data function
|
2017-11-10 12:52:18 +01:00 |
|
René Knaebel
|
3ce385eca6
|
update plotting function according the test and beta results
|
2017-11-10 12:34:21 +01:00 |
|
René Knaebel
|
c19d649bc4
|
add some print lines for better following the process structure
|
2017-11-10 11:38:57 +01:00 |
|
René Knaebel
|
27f4d086eb
|
add matplotlib agg mode; update beta vis function according to test results
|
2017-11-10 11:04:27 +01:00 |
|
René Knaebel
|
4fc2f0c925
|
extract weighting function
|
2017-11-10 10:18:13 +01:00 |
|
René Knaebel
|
9ce11e4db4
|
refactor test function according to the new training procedure
|
2017-11-09 13:12:50 +01:00 |
|
René Knaebel
|
9b8ca8abab
|
add parameter for hyper band iteration, use hyperband results in new runs
|
2017-11-08 11:09:56 +01:00 |
|
René Knaebel
|
903e81c931
|
remove identical parameter from data loading function; add runs argument
|
2017-11-07 20:47:41 +01:00 |
|
René Knaebel
|
826357a41f
|
refactor network models; remove depths
|
2017-11-07 20:32:08 +01:00 |
|
René Knaebel
|
e12bbda8c5
|
move model creation back into models package
|
2017-11-07 20:09:20 +01:00 |
|
René Knaebel
|
b1f48c1895
|
add soft parameter sharing network
|
2017-11-06 21:51:49 +01:00 |
|
René Knaebel
|
7b8dfcebbe
|
add long final implementation
|
2017-11-05 22:52:50 +01:00 |
|
René Knaebel
|
14fef66a55
|
train multiple models at once
|
2017-11-04 17:58:21 +01:00 |
|
René Knaebel
|
88e3eda595
|
refactor hyperband; fix domain generation
integrate hyperband option in training procedure - start refactoring - remove the index erro in generation and add helper functions
|
2017-11-04 12:47:08 +01:00 |
|
René Knaebel
|
8b17bd0701
|
add TSNE embedding; server evaluation visualization
|
2017-10-19 17:39:37 +02:00 |
|
René Knaebel
|
a860f0da34
|
refactor server training into separate file; add additional info to hyperband log
|
2017-10-19 17:37:29 +02:00 |
|
René Knaebel
|
d1da3d6ca3
|
fix model selection
|
2017-10-09 15:10:15 +02:00 |
|
René Knaebel
|
a686f147f0
|
add bulk embedding visualization and deep1 network
|
2017-10-09 14:19:01 +02:00 |
|
René Knaebel
|
33063f3081
|
add sample weight metrics to fit function
|
2017-10-08 22:09:09 +02:00 |
|
René Knaebel
|
e8473048cb
|
refactor visualization; start plotting server results
|
2017-10-08 11:52:10 +02:00 |
|
René Knaebel
|
0b26c6125c
|
fix flat 2 model bug caused by changing input layer
|
2017-10-06 16:36:58 +02:00 |
|
René Knaebel
|
5741f8ee0e
|
fix staggered training
|
2017-10-06 10:38:00 +02:00 |
|
René Knaebel
|
508667d1d0
|
add server classification model
|
2017-10-05 15:26:53 +02:00 |
|
René Knaebel
|
345afbaef5
|
add error bar vis, comment unused parameters from parser
|
2017-10-05 14:50:59 +02:00 |
|
René Knaebel
|
b24fa770f9
|
change hyperband to count minimal val_loss over all losses
|
2017-10-05 12:55:46 +02:00 |
|
René Knaebel
|
371a1dad05
|
add hyperband savefile config, minor change of parameter name
|
2017-10-03 18:58:54 +02:00 |
|
René Knaebel
|
68254d6629
|
add load function for hyper band results
|
2017-10-02 07:34:04 +02:00 |
|
René Knaebel
|
f02e0b7f99
|
fix hyperband wrong variable names
|
2017-09-29 23:34:39 +02:00 |
|
René Knaebel
|
a1e553f8f1
|
fix rerun wrong path
|
2017-09-29 23:03:07 +02:00 |
|
René Knaebel
|
605447440f
|
refactor hyperband implementation
|
2017-09-29 22:59:57 +02:00 |
|
René Knaebel
|
090c89a127
|
add retrain mode
|
2017-09-28 12:23:22 +02:00 |
|
René Knaebel
|
b157ca6a19
|
add first version of model averaging visualization
|
2017-09-26 19:25:37 +02:00 |
|
René Knaebel
|
49ad506a96
|
remove file name in run script
|
2017-09-22 10:03:01 +02:00 |
|
René Knaebel
|
607d74998c
|
add tsne (does not work with big data)
fix model loading with custom selu function
|
2017-09-22 10:01:12 +02:00 |
|
René Knaebel
|
e2bf2dc90f
|
fix missing parameters, add flat network structure, make larger graphics
|
2017-09-20 14:43:28 +02:00 |
|
René Knaebel
|
fbe6d6a584
|
remove input shape of first conv layer in networks because unnecessary
add selu activation to deeper network designs
|
2017-09-17 17:26:09 +02:00 |
|
René Knaebel
|
6a47b5f245
|
add sample weights function to utils for later usage
|
2017-09-17 10:23:23 +02:00 |
|
René Knaebel
|
f2845e635e
|
fix validation split removal by loading h5data into memory
|
2017-09-17 09:56:18 +02:00 |
|
René Knaebel
|
ec5a1101be
|
remove model selection based on validation loss
|
2017-09-16 15:25:34 +02:00 |
|
René Knaebel
|
b0e0cd904e
|
add start parameter for run script
|
2017-09-15 11:12:42 +02:00 |
|
René Knaebel
|
7f49021a63
|
refactor training - separate staggered training; make differences as small as possible
|
2017-09-12 08:36:23 +02:00 |
|
René Knaebel
|
6ce8fb464f
|
remove reguliarizer on dense too
|
2017-09-11 16:48:49 +02:00 |
|
René Knaebel
|
3a44efa775
|
minor changes for pauls test
|
2017-09-11 12:59:13 +02:00 |
|
René Knaebel
|
6d8d7b19f3
|
fix covariance normalization; add run_model script for multi times training
|
2017-09-11 12:42:44 +02:00 |
|
René Knaebel
|
6121eac448
|
remove regularizer for conv and domain
|
2017-09-10 23:40:14 +02:00 |
|
René Knaebel
|
1cf62423e1
|
add regularization to small networks, fix model name in args, fix visualizations
|
2017-09-10 18:06:40 +02:00 |