This website requires JavaScript.
b3d646c9e7
add group lasso regularizer impl
dev
René Knaebel
2017-11-30 09:34:45 +0100
f382d06eb5
add initial version of sluice network with alphas, betas, and soft share
René Knaebel
2017-11-27 16:17:19 +0100
349bc92a61
reorder curve storing
René Knaebel
2017-11-13 14:37:41 +0100
d58dbcb101
refactor class weights
René Knaebel
2017-11-10 14:31:32 +0100
461d4cab8f
add deeper domain cnn; refactor hyperband using load_data function
René Knaebel
2017-11-10 12:52:18 +0100
3ce385eca6
update plotting function according the test and beta results
René Knaebel
2017-11-10 12:34:21 +0100
c19d649bc4
add some print lines for better following the process structure
René Knaebel
2017-11-10 11:38:57 +0100
27f4d086eb
add matplotlib agg mode; update beta vis function according to test results
René Knaebel
2017-11-10 11:04:27 +0100
4fc2f0c925
extract weighting function
René Knaebel
2017-11-10 10:18:13 +0100
9ce11e4db4
refactor test function according to the new training procedure
René Knaebel
2017-11-09 13:12:50 +0100
9b8ca8abab
add parameter for hyper band iteration, use hyperband results in new runs
René Knaebel
2017-11-08 11:09:56 +0100
903e81c931
remove identical parameter from data loading function; add runs argument
René Knaebel
2017-11-07 20:47:41 +0100
826357a41f
refactor network models; remove depths
René Knaebel
2017-11-07 20:32:08 +0100
e12bbda8c5
move model creation back into models package
René Knaebel
2017-11-07 20:09:20 +0100
b1f48c1895
add soft parameter sharing network
René Knaebel
2017-11-06 21:51:49 +0100
7b8dfcebbe
add long final implementation
René Knaebel
2017-11-05 22:52:50 +0100
14fef66a55
train multiple models at once
René Knaebel
2017-11-04 17:58:21 +0100
88e3eda595
refactor hyperband; fix domain generation integrate hyperband option in training procedure - start refactoring - remove the index erro in generation and add helper functions
René Knaebel
2017-11-04 12:47:08 +0100
8b17bd0701
add TSNE embedding; server evaluation visualization
René Knaebel
2017-10-19 17:39:37 +0200
a860f0da34
refactor server training into separate file; add additional info to hyperband log
René Knaebel
2017-10-19 17:37:29 +0200
d1da3d6ca3
fix model selection
René Knaebel
2017-10-09 15:10:15 +0200
a686f147f0
add bulk embedding visualization and deep1 network
René Knaebel
2017-10-09 14:19:01 +0200
33063f3081
add sample weight metrics to fit function
René Knaebel
2017-10-08 22:09:09 +0200
e8473048cb
refactor visualization; start plotting server results
René Knaebel
2017-10-08 11:52:10 +0200
0b26c6125c
fix flat 2 model bug caused by changing input layer
René Knaebel
2017-10-06 16:36:58 +0200
5741f8ee0e
fix staggered training
René Knaebel
2017-10-06 10:38:00 +0200
508667d1d0
add server classification model
René Knaebel
2017-10-05 15:26:53 +0200
345afbaef5
add error bar vis, comment unused parameters from parser
René Knaebel
2017-10-05 14:50:59 +0200
b24fa770f9
change hyperband to count minimal val_loss over all losses
René Knaebel
2017-10-05 12:55:46 +0200
371a1dad05
add hyperband savefile config, minor change of parameter name
René Knaebel
2017-10-03 18:58:54 +0200
68254d6629
add load function for hyper band results
René Knaebel
2017-10-02 07:34:04 +0200
f02e0b7f99
fix hyperband wrong variable names
René Knaebel
2017-09-29 23:34:39 +0200
a1e553f8f1
fix rerun wrong path
René Knaebel
2017-09-29 23:03:07 +0200
605447440f
refactor hyperband implementation
René Knaebel
2017-09-29 22:59:57 +0200
090c89a127
add retrain mode
René Knaebel
2017-09-28 12:23:22 +0200
b157ca6a19
add first version of model averaging visualization
René Knaebel
2017-09-26 19:25:37 +0200
49ad506a96
remove file name in run script
René Knaebel
2017-09-22 10:03:01 +0200
607d74998c
add tsne (does not work with big data) fix model loading with custom selu function
René Knaebel
2017-09-22 10:01:12 +0200
e2bf2dc90f
fix missing parameters, add flat network structure, make larger graphics
René Knaebel
2017-09-20 14:43:28 +0200
fbe6d6a584
remove input shape of first conv layer in networks because unnecessary add selu activation to deeper network designs
René Knaebel
2017-09-17 17:26:09 +0200
6a47b5f245
add sample weights function to utils for later usage
René Knaebel
2017-09-17 10:23:23 +0200
f2845e635e
fix validation split removal by loading h5data into memory
René Knaebel
2017-09-17 09:56:18 +0200
ec5a1101be
remove model selection based on validation loss
René Knaebel
2017-09-16 15:25:34 +0200
b0e0cd904e
add start parameter for run script
René Knaebel
2017-09-15 11:12:42 +0200
7f49021a63
refactor training - separate staggered training; make differences as small as possible
René Knaebel
2017-09-12 08:36:23 +0200
6ce8fb464f
remove reguliarizer on dense too
René Knaebel
2017-09-11 16:48:49 +0200
3a44efa775
minor changes for pauls test
René Knaebel
2017-09-11 12:59:13 +0200
6d8d7b19f3
fix covariance normalization; add run_model script for multi times training
René Knaebel
2017-09-11 12:42:44 +0200
6121eac448
remove regularizer for conv and domain
René Knaebel
2017-09-10 23:40:14 +0200
1cf62423e1
add regularization to small networks, fix model name in args, fix visualizations
René Knaebel
2017-09-10 18:06:40 +0200
6fef2b8b84
refactor all visualization for pauls changes - evaluate on max windows per users
René Knaebel
2017-09-08 22:59:55 +0200
9a51b6ea34
refactor test function working on full unfiltered data
René Knaebel
2017-09-08 19:10:23 +0200
edc75f4f44
refactor dataset creation, split up functions
René Knaebel
2017-09-08 17:11:13 +0200
528829bb33
fix file name in run script
René Knaebel
2017-09-08 13:57:18 +0200
70d00efb01
refactor using joblib for test results, make h5py store/load more flexible
René Knaebel
2017-09-08 13:55:13 +0200
1ab0108c78
add window to file names for visualization
René Knaebel
2017-09-07 17:38:21 +0200
595c2ea894
change argument interface - add more properties for network specification - change names for consistency
René Knaebel
2017-09-07 15:53:58 +0200
71f218888d
set server to be not trainable too; refactor visualization script
René Knaebel
2017-09-07 15:31:04 +0200
5bd8e41711
add staggered model training for intermediate sever prediction; refactor model return values
René Knaebel
2017-09-07 14:24:55 +0200
2080444fb7
add parser argument for naming in multi model modes, minor fixes, re-set fix vals for network - need to make them flexible
René Knaebel
2017-09-05 17:40:57 +0200
ed4f478bad
add parser argument for naming in multi model modes, minor fixes,
René Knaebel
2017-09-05 12:40:37 +0200
1da31cc97c
visualize per user stats
René Knaebel
2017-09-04 13:37:26 +0200
3f6779fa3d
load names with data for per-user evaluation
René Knaebel
2017-09-02 16:02:48 +0200
0db8427457
update test script with bash loop
René Knaebel
2017-09-02 12:58:48 +0200
dc9180da10
refactor visualization, change arguments for model type and its depth
René Knaebel
2017-09-01 10:42:26 +0200
933eaae04a
change exception type in get_flow_per_user function and replace index to new range index
René Knaebel
2017-08-31 13:49:33 +0200
dceaf47211
fix typo
René Knaebel
2017-08-31 11:25:49 +0200
954dfcf9f9
fix run scripts
René Knaebel
2017-08-05 09:40:40 +0200
5a02f582cd
change model - add dense before server output in new model add some new run scripts
René Knaebel
2017-08-05 09:33:07 +0200
6e7dc1297c
fix lazy domain loading and generation process
René Knaebel
2017-08-03 12:27:17 +0200
7f1d13658f
store domain embeddings while test main
René Knaebel
2017-08-03 09:08:24 +0200
452f9e0456
fix array() to zeros(0)
René Knaebel
2017-08-03 08:00:20 +0200
787f43b328
fix test predictions depending on model output specification
René Knaebel
2017-08-03 07:51:58 +0200
8ac195ba6f
add some static configs to run script
René Knaebel
2017-08-02 15:13:09 +0200
1e781d5491
add argument for model outputs BUG: need to check --new_model --model_output server
René Knaebel
2017-08-02 12:58:09 +0200
f4da147688
refactor cmd argument to have single value for mode
René Knaebel
2017-07-30 15:49:37 +0200
e24f596f40
add argument for using the new model architecture
René Knaebel
2017-07-30 14:07:39 +0200
ebaeb6b96e
move vocab_size into implementation (not user dependent)
René Knaebel
2017-07-30 13:47:11 +0200
d97785f646
replace softmax by sigmoid in final layer, also adjust dataset for that
René Knaebel
2017-07-30 12:50:26 +0200
b0da2de0ea
move utils functions to new file
René Knaebel
2017-07-29 19:47:02 +0200
820a5d1a4d
add new network architecture - server label moves to the middle
René Knaebel
2017-07-29 19:42:36 +0200
8cd1023165
replace pca reduction by sklearn's truncated svd
René Knaebel
2017-07-29 19:41:14 +0200
2593131e9e
add embedding visualization and domain encoding generator
René Knaebel
2017-07-29 10:43:59 +0200
c1535b941b
remove max pooling from models for better infromation flow
René Knaebel
2017-07-28 17:25:08 +0200
18b60e1754
add extended test mode for embeddings
René Knaebel
2017-07-17 19:30:56 +0200
79fc441fe1
wip
René Knaebel
2017-07-17 08:44:58 +0200
d33c9f44ec
fix chunks per user function bug caused by numpy version of array_split
René Knaebel
2017-07-16 18:49:14 +0200
844494eca9
add multi-threading for pre-processing
René Knaebel
2017-07-16 09:42:52 +0200
336be37032
fix network props, add PCA to visualize main
René Knaebel
2017-07-14 21:01:08 +0200
6b787792db
add custom class weights based on sklearn balance
René Knaebel
2017-07-14 15:57:52 +0200
b35f23e518
add visualization for training curves, pr, roc
René Knaebel
2017-07-14 14:58:17 +0200
d0418b9efa
fix hyperband adding h5data and remove param sample
René Knaebel
2017-07-12 15:18:45 +0200
2afaccc84b
refactor argparser into separate file, add logger
René Knaebel
2017-07-12 10:25:55 +0200
9f0bae33d5
refactor dataset generation, add callbacks
René Knaebel
2017-07-11 21:06:58 +0200
a196daa895
add simple flow feature extraction function
René Knaebel
2017-07-11 13:46:25 +0200
522854ee0d
add h5 support for pauls best config main
René Knaebel
2017-07-11 11:12:03 +0200
41b38de1ab
add feature: generate and use h5 data
René Knaebel
2017-07-09 23:58:08 +0200
fdc03c9922
add h5py example
René Knaebel
2017-07-08 17:46:07 +0200
4a9f94a029
add output for main_test
René Knaebel
2017-07-08 15:04:58 +0200
21b9d7be73
merged
René Knaebel
2017-07-08 13:04:07 +0200