a6de8c7
Add test_flatten_predictions_05(). by
2020-06-20 17:13:31 +0200
1b72e41
Add test_[one_hot_]input_layer_parameter_count_01(). by
2020-06-20 15:18:48 +0200
77c52d8
Add test_[dedicom,dist_mult,bilinear,inner_product]_decoder_parameter_count_01(). by
2020-06-20 15:06:08 +0200
6c8fdb7
Add test_decode_layer_parameter_count_[01-03](). by
2020-06-20 15:00:32 +0200
b79487e
Add test_dropout_graph_conv_activation_parameter_count_01(). by
2020-06-20 14:54:11 +0200
108140b
Add test_graph_conv_parameter_count_01(). by
2020-06-20 14:53:26 +0200
094813b
Change test in test_unigram_03(). by
2020-06-19 21:55:10 +0200
03478e2
Add test_parameter_count_01(). by
2020-06-19 21:00:08 +0200
f6006dc
Add test_model_08(). by
2020-06-19 20:14:14 +0200
c023d35
Introduce FlatPredictions, flatten_predictions(), BatchIndices, gather_batch_indices(). by
2020-06-18 18:27:02 +0200
2b388e4
Fix test_model. by
2020-06-18 13:31:18 +0200
9366687
Add TrainLoop. by
2020-06-18 13:00:17 +0200
f11f257
Make Model accept PreparedData rather than Data. by
2020-06-18 09:40:36 +0200
f17f97c
Add test_model_07(). by
2020-06-17 15:33:00 +0200
190a00e
Fix test_input. by
2020-06-17 15:17:08 +0200
414c934
Add test_model_06(). by
2020-06-17 15:13:56 +0200
f6e1024
Add test_model_05(). by
2020-06-17 15:04:19 +0200
aa541e0
Fix test_convolve. by
2020-06-17 12:43:04 +0200
692f221
Add test_model_04(). by
2020-06-17 12:37:28 +0200
cfb3d30
Add test_model_03(). by
2020-06-17 12:28:46 +0200
0578b13
Fix prepare_adj_mat() by specifying shape explicitly. by
2020-06-17 12:14:05 +0200
90d40db
Add test_model_02(). by
2020-06-17 12:03:44 +0200
d5779da
Add test_model_01(). by
2020-06-17 11:30:07 +0200
46bf3b1
Rename test_loss. by
2020-06-16 19:01:07 +0200
346cc74
Add shuffle to PredictionsBatch. by
2020-06-16 18:12:44 +0200
908100c
Add Model. by
2020-06-16 17:42:10 +0200
7235654
Remove loss as no longer necessary. by
2020-06-16 16:57:22 +0200
a9f14d1
Add PredictionsBatch and test_predictions_btch_01(). by
2020-06-16 15:11:25 +0200
4ab4bd6
Add some asserts to test_cross_entropy_loss_01(). by
2020-06-15 18:03:34 +0200
ff4fb96
Add CrossEntropyLoss and test_cross_entropy_loss_01(). by
2020-06-15 17:56:19 +0200
dc8c51a
Ammend test_decode_layer_05. by
2020-06-15 17:12:52 +0200
b92c7a0
Add test_decode_layer_05. by
2020-06-15 17:01:09 +0200
5ab9231
Fix test_decode_layer_04. by
2020-06-15 11:53:24 +0200
fc11f05
Fix test_decode_layer_03. by
2020-06-15 11:42:54 +0200
093d378
Fix test_decode_layer_02. by
2020-06-15 11:38:14 +0200
25e05cf
Use torch.nn.Parameter(List) in decode. by
2020-06-12 21:49:44 +0200
55a5f3a
Use torch.nn.ModuleList in convlayer. by
2020-06-12 21:38:00 +0200
fe6c659
Add module nesting tests. by
2020-06-12 21:09:04 +0200
bf913e4
Add icosagon-reltype-rules. by
2020-06-12 18:55:47 +0200
f3371d9
Add tests for decoders on empty inputs. by
2020-06-12 17:55:19 +0200
8767db6
Remove node_type_row/node_type_column as redundant in add_relation_type(). by
2020-06-12 16:14:03 +0200
e8122c3
Get test_decode_layer_01 to pass. by
2020-06-12 15:16:50 +0200
832f620
Store relation types in a list instead of a dictionary and related fixes. by
2020-06-11 22:48:22 +0200
a72b6b0
Fix regressions in test_input. by
2020-06-11 18:37:32 +0200
7eda6bd
Fix regressions in trainprep. by
2020-06-11 18:28:22 +0200
915c568
Small fix. by
2020-06-11 14:23:16 +0200
6c52bc1
Fix regressions in test_convlayer. by
2020-06-11 14:17:50 +0200
a1b0272
Add icosagon-hierarchy.svg. by
2020-06-09 19:37:45 +0200
b5d2f8f
Start implementing RelationFamily. by
2020-06-09 15:23:43 +0200
4b0f68d
Add symmetry tests for decode. by
2020-06-09 13:34:58 +0200
56ce7aa
Add type checks to DecodeLayer. by
2020-06-09 13:07:08 +0200
d06e65b
Add icosagon classes diagram. by
2020-06-08 12:34:46 +0200
15e95bd
Use hints instead of is_autogenerated. by
2020-06-08 10:54:32 +0200
53cd182
Modify decoders to handle one relation at a time. by
2020-06-08 10:44:08 +0200
eda27be
Increase test coverage. by
2020-06-07 18:30:10 +0200
bd894a0
Move back to using single-level dictionary for Data.relation_types. by
2020-06-07 18:10:14 +0200
dc92634
Same test with torch.nn.Sequential. by
2020-06-07 15:55:59 +0200
f51855e
Add first test for declayer. by
2020-06-07 15:54:10 +0200
1868014
Work on declayer. by
2020-06-07 15:39:51 +0200
6d0039b
Add test for init_glorot. by
2020-06-07 15:12:26 +0200
702079c
Add sampling tests. by
2020-06-07 14:55:47 +0200
dd0eb81
Add tests for decode. by
2020-06-07 14:06:56 +0200
a305dc7
Add tests for convlayer. by
2020-06-07 13:40:10 +0200
d4dd1f2
Add input to icosagon. by
2020-06-07 12:51:22 +0200
07a82e0
Work on icosagon. by
2020-06-07 12:05:29 +0200
d503fb5
Bring more elements to icosagon. by
2020-06-07 10:21:14 +0200
706ff0b
New implementation of normalize for one/two node types and sparse/dense - good. by
2020-06-06 19:05:19 +0200
6f80b65
New implementation for one node type normalization seems to work. by
2020-06-06 17:39:35 +0200
05b1ecf
Start rework normalize. by
2020-06-06 17:19:46 +0200
584bf19
Work on icosagon.trainprep. by
2020-06-06 15:09:24 +0200
7ed4bc3
Work on icosagon.trainprep. by
2020-06-06 13:54:31 +0200
3795545
Start icosagon. by
2020-06-06 11:52:44 +0200
456a037
Fix typo. by
2020-06-05 18:07:05 +0200
a272a17
Work on trainprep. by
2020-06-05 17:52:38 +0200
e201608
Add trainprep. by
2020-06-05 15:07:58 +0200
28272e3
Add AdjListRelationType, AdjListData. by
2020-06-05 14:46:32 +0200
3a0c8aa
Add train-val-test-diagram. by
2020-06-05 00:40:45 +0200
fab2104
Correct support for assymetric relations in Data. by
2020-06-03 16:45:03 +0200
a74884d
Add train_val_test_split_adj_mat(). by
2020-06-02 16:28:35 +0200
7ab97e2
Add test_unigram_03(). by
2020-06-02 12:08:34 +0200
f2121be
Add fixed_unigram_candidate_sampler(). by
2020-06-02 11:50:51 +0200
39aa1b4
Implement and test pairwise decoders. by
2020-06-01 14:30:13 +0200
c1689b4
Split decoders into cartesian and pairwise. by
2020-06-01 14:13:11 +0200
b08ae2e
Test for an approach to pairwise (rather than cartesian product) link predition. by
2020-06-01 13:18:11 +0200
4f243fb
Copyright notes. by
2020-05-29 15:28:30 +0200
0f1afa3
Put back layer tests. by
2020-05-29 15:24:46 +0200
c37e4dc
Ok the very first decoding seems to work end-to-end. by
2020-05-29 15:23:58 +0200
1d2fac0
Split layer into submodules. by
2020-05-29 13:14:00 +0200
4ac42e5
Add test_decagon_layer_05(). by
2020-05-29 11:47:33 +0200
52be406
Adjust DecagonLayer to normalize per edge type. by
2020-05-29 11:33:17 +0200
241a2f9
Fix tests. by
2020-05-29 10:56:14 +0200
c74555f
Refactor convolve into 3 separate modules. by
2020-05-29 10:46:57 +0200
02bbfc4
Add test_decagon_layer_04(). by
2020-05-29 10:38:08 +0200
69e8ce5
Add universal convolution units that can handle both sparse and dense inputs. by
2020-05-29 09:51:50 +0200
99dbcde
Prefix dense versions of convolution classes with Dense. by
2020-05-28 11:59:43 +0200
29e81f4
Add test for DecagonLayer. by
2020-05-28 11:24:10 +0200
5d8c2d0
Add tests for OneHotInputLayer. by
2020-05-28 10:41:41 +0200
69dd9a4
Dummy run of DecagonLayer seems to work, that's something. by
2020-05-27 18:50:52 +0200
2188dc7
Basic constructor test for DecagonLayer passes. by
2020-05-27 18:17:15 +0200
60d8a43
Fix the DecagonLayer logic. by
2020-05-27 18:11:32 +0200