| 2136d31 | adriap | 29 January 2018, 17:55:18 UTC | Add backwards compatibility with tests for ones and zeros name scopes. PiperOrigin-RevId: 183680464 | 29 January 2018, 17:59:00 UTC |
| 1d2d99e | diegolascasas | 29 January 2018, 11:57:14 UTC | Remove fixed seed dependency in gated_rnn_test.LSTMTest.testRecurrentDropout PiperOrigin-RevId: 183644341 | 29 January 2018, 17:03:51 UTC |
| 132924d | Deepmind | 25 January 2018, 17:25:35 UTC | Refactor out bias construction and application. Refactor out the construction of the bias variable and its application to the output of all Convolution modules. PiperOrigin-RevId: 183249262 | 29 January 2018, 17:03:46 UTC |
| 84bab04 | mareynolds | 25 January 2018, 16:44:55 UTC | Enable custom_getter for TrainableVariable. PiperOrigin-RevId: 183243768 | 29 January 2018, 17:03:40 UTC |
| 438b320 | Deepmind | 25 January 2018, 14:33:20 UTC | Refactor out inputs verification. Remove duplicate code that is performing the same checks on the `inputs` tensor in every Convolution class. PiperOrigin-RevId: 183230711 | 29 January 2018, 17:03:35 UTC |
| 2a8e6e9 | tfgg | 24 January 2018, 13:09:04 UTC | Fix typo. PiperOrigin-RevId: 183071198 | 29 January 2018, 17:03:29 UTC |
| de769df | Deepmind | 22 January 2018, 14:09:01 UTC | 1. Start supporting TensorShapes for the output_shape input for Conv*DTranspose classes. 2. Support the use of tf.Dimension objects instead of integers for dimensions. 3. Add tests to make sure output_shape inference works correctly. PiperOrigin-RevId: 182768071 | 29 January 2018, 17:03:24 UTC |
| 3aead32 | adriap | 22 January 2018, 11:46:16 UTC | Correct typo in warning. PiperOrigin-RevId: 182755077 | 29 January 2018, 17:03:18 UTC |
| 109d606 | Deepmind | 19 January 2018, 12:11:15 UTC | Make masked Conv2D usable with ResourceVariables by avoiding use of *=. PiperOrigin-RevId: 182516172 | 29 January 2018, 17:03:11 UTC |
| 7393cfb | mareynolds | 16 January 2018, 14:13:07 UTC | Explicitly enter the scope of the connected Graph in AbstractModule.get_variables() This fixeds the following inconsistency: ``` with tf.Graph().as_default() as graph: input = tf.constant(np.random.randn(16, 784)) lin = snt.Linear(output_size=256) output = lin(input) print(lin.get_variables()) # prints tuple of 2 variables print(lin.get_variables()) # prints empty tuple, as technically we are in a different Graph. ``` Also added a .graph readonly property to AbstractModule. PiperOrigin-RevId: 182044744 | 29 January 2018, 16:56:19 UTC |
| 7bdd40c | Deepmind | 15 January 2018, 20:38:19 UTC | Provide better examples for snt.reuse_variables. PiperOrigin-RevId: 181984575 | 29 January 2018, 16:56:13 UTC |
| 3943452 | Deepmind | 12 January 2018, 15:51:09 UTC | Push BBB library and Bayesian RNN example to open source sonnet. PiperOrigin-RevId: 181743627 | 29 January 2018, 16:56:08 UTC |
| 315719e | Deepmind | 12 January 2018, 11:35:07 UTC | Select the correct input channel and stride values within Conv{1,2,3}DTranspose.transpose() when the data_format is NC*. Add more tests for Conv*Transpose transpose functionality. PiperOrigin-RevId: 181724912 | 29 January 2018, 16:56:02 UTC |
| 7865f86 | Deepmind | 11 January 2018, 15:08:07 UTC | Improve an error message. This is supposing `data_format` is a string, but if one is incorrectly passing something else, the error message will fail e.g. with ValueError: Unknown format code 's' for object of type 'int' With this change the error becomes: ValueError: Invalid data_format 256. Allowed formats set(['NCHW', 'NHWC']) PiperOrigin-RevId: 181604070 | 29 January 2018, 16:55:57 UTC |
| 3a21b8d | diegolascasas | 11 January 2018, 11:56:09 UTC | Remove dependency on fixed seed for testZoneout. This means it won't be possible to test for expected state values. PiperOrigin-RevId: 181587515 | 29 January 2018, 16:55:48 UTC |
| 1d2171d | diegolascasas | 10 January 2018, 15:51:51 UTC | Sonnet version update produced on Monday, 8. January 2018 PiperOrigin-RevId: 181464426 | 10 January 2018, 17:19:34 UTC |
| f11b5af | Deepmind | 09 January 2018, 11:04:00 UTC | Add Con[1-3]DTranspose classes to conv_gpu_tests. Expanding testing coverage of the transpose classes. PiperOrigin-RevId: 181293937 | 10 January 2018, 16:53:39 UTC |
| 3c18209 | Deepmind | 05 January 2018, 19:42:45 UTC | Add Causal1DConv testing to conv_gpu_test. PiperOrigin-RevId: 180955048 | 08 January 2018, 12:15:39 UTC |
| a1b1f34 | Deepmind | 05 January 2018, 18:04:43 UTC | Stop using constant initializers in conv_gpu_test.py. Unit tests were using constant initializers to check for correctness in data_format permutation operations. This is problematic; you want the weights to be very random so that they don't inadvertently mask an issue in their functionality. PiperOrigin-RevId: 180941577 | 08 January 2018, 12:15:33 UTC |
| fe91080 | Deepmind | 05 January 2018, 13:24:42 UTC | Rename SUPPORTED_DATA_FORMATS to SUPPORTED_2D_DATA_FORMATS. As we support far more than just 2D conv now, this variable needs to change. PiperOrigin-RevId: 180915317 | 08 January 2018, 12:15:28 UTC |
| 865f4aa | Deepmind | 05 January 2018, 12:41:21 UTC | Refactor CausalConv1D to use _ConvND. Tests for clone functionality in all _ConvND subclasses. CausalConv1D can exist using the new _ConvND superclass. Add support within _ConvND to accommodate padded input. Refactored _ConvND a bit to get smaller methods as _build() was getting unwieldy. With CausalConv1D inheriting from _ConvND, it is now cloneable by default. I noticed that most of the classes lacked testing of their clone functionality, so I added those. Finally, in supporting multiple data_formats, I noticed there aren't enough tests that check to see an exception is thrown when an invalid data_format is used. So I added those. PiperOrigin-RevId: 180912740 | 08 January 2018, 12:15:22 UTC |
| 699f8cb | Deepmind | 05 January 2018, 11:12:33 UTC | Refactor Conv{1,2,3}Transpose classes into one. Add tests for transpose functionality. Rearchitect the Transpose classes into one base class that is then instantiated for each dimension convolution we want. Add tests for Conv{1,3}Transpose; we make sure to test N*C and NC* data formats. PiperOrigin-RevId: 180907942 | 08 January 2018, 12:15:16 UTC |
| 511f11b | Deepmind | 03 January 2018, 20:47:11 UTC | Add a Recurrent Highway Network cell. PiperOrigin-RevId: 180705745 | 08 January 2018, 12:15:06 UTC |
| 5fe263b | mareynolds | 03 January 2018, 16:40:55 UTC | Improve snt.Embed performance in distributed training. Adds a fix to avoid excess computation on parameter servers. PiperOrigin-RevId: 180674688 | 08 January 2018, 12:15:01 UTC |
| 6f9f46e | Deepmind | 02 January 2018, 11:14:13 UTC | Update AbstractModule documentation. PiperOrigin-RevId: 180531448 | 08 January 2018, 12:14:55 UTC |
| ca13720 | Deepmind | 22 December 2017, 13:32:52 UTC | Creation of a ConvND class. More flexible masks. Lots more tests. Creating a ConvND class that is then subclassed for Conv{1,2,3}D. This removes a lot of duplicated boilerplate code. As part of this we, are also adding a `mask` argument to the Conv{1,3}D classes. Finally, we remove rank restrictions on the mask argument. We add testing for new masking functionality. PiperOrigin-RevId: 179918026 | 08 January 2018, 12:14:49 UTC |
| d0f50a6 | Deepmind | 22 December 2017, 12:47:04 UTC | Extend batch_norm_v2 data_format default across dimensions. PiperOrigin-RevId: 179915706 | 08 January 2018, 12:14:44 UTC |
| 89c2091 | Deepmind | 22 December 2017, 11:34:10 UTC | Improve snt.Embed performance in distributed training. Adds a fix to avoid excess computation on parameter servers. PiperOrigin-RevId: 179912285 | 08 January 2018, 12:14:38 UTC |
| 9a547ed | Deepmind | 21 December 2017, 16:56:19 UTC | Implemented snt.BatchNormV2, which differs from snt.BatchNorm in the following ways: * Automatically computes updates to moving statistics by default (i.e. update_ops_collection=None). * Uses moving statistics by default when testing (i.e. test_local_stats=False). * Takes a data_format string (NC/NWC/NCW/NHWC/NCHW/NDHWC/NCDHW) rather than axes; reduces along all non-C axes. * Uses fused batch normalization by default. If the data_format isn't NHWC or NCHW, reshapes the batch internally. * Uses flat variables for the moving statistics, scale, and offset so that they can be shared between different data_formats. PiperOrigin-RevId: 179819339 | 08 January 2018, 12:14:32 UTC |
| b920570 | Deepmind | 20 December 2017, 19:52:49 UTC | tf.float16 support for batch_norm PiperOrigin-RevId: 179714930 | 08 January 2018, 12:14:26 UTC |
| e2ce831 | mareynolds | 19 December 2017, 17:49:09 UTC | Improve error message when partitioners/regularizers/initializers used wrong. PiperOrigin-RevId: 179566444 | 08 January 2018, 12:14:20 UTC |
| 3b9f636 | mareynolds | 19 December 2017, 17:48:00 UTC | When Sequential contains no layers, simply act as identity. Previously, a sequential with no layers would return any input wrapped in a 1 element tuple, which is a strange inconsistency. PiperOrigin-RevId: 179566305 | 08 January 2018, 12:14:14 UTC |
| a7a557e | mareynolds | 19 December 2017, 17:10:36 UTC | Improve error message when partitioners/regularizers/initializers used wrong. PiperOrigin-RevId: 179561994 | 08 January 2018, 12:14:09 UTC |
| f2d6f4b | Deepmind | 19 December 2017, 16:14:37 UTC | Corrected parameter list for testFusedBatchNorm, where `is_training` had been erroneously set to `True` in some cases. PiperOrigin-RevId: 179555601 | 08 January 2018, 12:14:03 UTC |
| 293e86d | mareynolds | 19 December 2017, 13:25:05 UTC | Improve error message when partitioners/regularizers/initializers used wrong. PiperOrigin-RevId: 179541337 | 08 January 2018, 12:13:57 UTC |
| 13c724a | Deepmind | 18 December 2017, 11:56:13 UTC | Support tf.float16 inputs in convolution modules. PiperOrigin-RevId: 179402639 | 18 December 2017, 14:26:26 UTC |
| 26013e4 | Deepmind | 18 December 2017, 10:52:11 UTC | Conv3DTranspose to initialize biases to zero by default. PiperOrigin-RevId: 179398434 | 18 December 2017, 14:26:21 UTC |
| 85e1b7a | Deepmind | 18 December 2017, 09:53:03 UTC | Adjust test to use .assertDictEqual() rather than .assertItemsEqual() on dicts. Reason: Latter comparison, when used on dicts, only compares dict-keys, ignoring dict-values. PiperOrigin-RevId: 179394091 | 18 December 2017, 14:26:15 UTC |
| beb3fb5 | mareynolds | 15 December 2017, 20:31:48 UTC | Get gated_rnn_test passing again PiperOrigin-RevId: 179226394 | 18 December 2017, 14:26:09 UTC |
| ebd18c8 | Deepmind | 12 December 2017, 14:05:16 UTC | Change "deprecated" to "not supported". PiperOrigin-RevId: 178751493 | 18 December 2017, 14:26:04 UTC |
| a1547dd | Deepmind | 05 December 2017, 17:12:58 UTC | Use a namedtuple for LSTM state, so can safely access the cell and hidden components. PiperOrigin-RevId: 177963672 | 18 December 2017, 14:25:58 UTC |
| bfe2962 | Deepmind | 30 November 2017, 15:45:46 UTC | Slice TensorShape instead of Tensor in basic_rnn.py. This prevents DeepRNN.output_size() from creating a slice operation on a Tensor declared inside the dynamic RNN loop body (output_size() is called outside the loop body, so the resulting slice operation is invalid). This currently works because the slice operation isn't actually run (since it's only used to get its shape), but in the future creating operations with invalid inputs will raise an exception. PiperOrigin-RevId: 177453503 | 18 December 2017, 14:25:52 UTC |
| 1ff6907 | Deepmind | 27 November 2017, 16:25:48 UTC | Adds NCHW support to Conv1D and CausalConv1D. PiperOrigin-RevId: 177019383 | 18 December 2017, 14:25:47 UTC |
| df4b83d | Deepmind | 23 November 2017, 11:28:59 UTC | Add a zoneout wrapper for recurrent neural networks and specialize it for LSTM. PiperOrigin-RevId: 176755442 | 18 December 2017, 14:25:41 UTC |
| 6f95f75 | Deepmind | 21 November 2017, 17:49:47 UTC | Add space at end of deprecation warning. PiperOrigin-RevId: 176521653 | 18 December 2017, 14:25:36 UTC |
| d8503c8 | Deepmind | 21 November 2017, 14:59:43 UTC | Add a recurrent dropout wrapper and specialize it for LSTM. PiperOrigin-RevId: 176502360 | 18 December 2017, 14:25:30 UTC |
| 96d2666 | Deepmind | 21 November 2017, 10:59:52 UTC | Change deprecation update message and nest import PiperOrigin-RevId: 176485081 | 18 December 2017, 14:25:24 UTC |
| 90c37e0 | jwrae | 20 November 2017, 15:52:01 UTC | Sonnet LSTM: add optional projection of hidden recurrent state. PiperOrigin-RevId: 176359570 | 18 December 2017, 14:25:17 UTC |
| 52edfd0 | diegolascasas | 20 November 2017, 15:33:47 UTC | Sonnet version update produced on Monday, 20. November 2017 PiperOrigin-RevId: 176357249 | 22 November 2017, 17:31:58 UTC |
| 6a13bca | Deepmind | 20 November 2017, 11:51:29 UTC | Update batch-norm example in comment. PiperOrigin-RevId: 176338001 | 22 November 2017, 17:31:58 UTC |
| eeccbfb | Deepmind | 16 November 2017, 17:01:04 UTC | Replace sonnet.nest *iterable functions with their equivalent from TF. PiperOrigin-RevId: 175970632 | 22 November 2017, 17:31:58 UTC |
| 6faf485 | tfgg | 14 November 2017, 16:10:07 UTC | Add custom_getter option to snt.Embed. PiperOrigin-RevId: 175680730 | 22 November 2017, 17:31:58 UTC |
| 71f179d | Deepmind | 14 November 2017, 14:39:47 UTC | Expose util.custom_getter_router, which was private. PiperOrigin-RevId: 175672207 | 22 November 2017, 17:31:58 UTC |
| a6cb7f5 | Diego de Las Casas | 09 November 2017, 16:00:49 UTC | Fix :base build rule dependency on :module_pb2. | 09 November 2017, 16:44:36 UTC |
| dcf8c90 | Diego de Las Casas | 09 November 2017, 15:58:31 UTC | Revert initial implementation for eager mode. Reverted due to incompatibility with Tensorflow 1.4.0. This reverts commit e1ef6d503205a7d9790ae45f3c77457992c88e03. | 09 November 2017, 16:43:34 UTC |
| 237c385 | mareynolds | 09 November 2017, 15:07:22 UTC | Internal change. PiperOrigin-RevId: 175153104 | 09 November 2017, 15:08:29 UTC |
| 0a2c31f | mareynolds | 09 November 2017, 14:38:28 UTC | Clarify TensorFlow version PiperOrigin-RevId: 175150554 | 09 November 2017, 15:01:23 UTC |
| b30a0e5 | mareynolds | 09 November 2017, 14:36:17 UTC | Add copyright header to module.proto PiperOrigin-RevId: 175150376 | 09 November 2017, 15:01:17 UTC |
| 7e6193d | mareynolds | 09 November 2017, 14:31:35 UTC | Update Changelog PiperOrigin-RevId: 175150004 | 09 November 2017, 15:01:11 UTC |
| 1e58060 | mareynolds | 09 November 2017, 13:15:09 UTC | Correct link in CONTRIBUTING.md PiperOrigin-RevId: 175144245 | 09 November 2017, 15:01:05 UTC |
| 5f966b5 | diegolascasas | 09 November 2017, 12:38:16 UTC | Add version information in the init. PiperOrigin-RevId: 175141587 | 09 November 2017, 15:00:59 UTC |
| 9d7b9fb | Deepmind | 03 November 2017, 18:01:59 UTC | Allowing __iter__ over 1+dimensional tensors with known shapes. PiperOrigin-RevId: 174484601 | 09 November 2017, 15:00:53 UTC |
| 5c44640 | Deepmind | 03 November 2017, 16:39:42 UTC | Removal of deprecated sonnet/testing/parameterized module due to its migration to Abseil. PiperOrigin-RevId: 174472509 | 09 November 2017, 15:00:47 UTC |
| 0356d00 | diegolascasas | 02 November 2017, 17:30:21 UTC | Internal changes. PiperOrigin-RevId: 174345289 | 09 November 2017, 14:49:27 UTC |
| 354bf3f | Deepmind | 02 November 2017, 13:11:30 UTC | snt.BatchApply now passes scalar-valued non-tensor inputs, such as Boolean flags, without converting them to tensors. PiperOrigin-RevId: 174316427 | 09 November 2017, 14:49:21 UTC |
| e1ef6d5 | Deepmind | 02 November 2017, 11:03:50 UTC | Sonnet initial implementation for eager mode. PiperOrigin-RevId: 174308021 | 09 November 2017, 14:49:13 UTC |
| dc09af1 | Deepmind | 31 October 2017, 19:53:46 UTC | Adds NCHW support to DepthwiseConv2D and SeparableConv2D. Fixes weight initialisation scaling of DepthwiseConv2D and SeparableConv2D. PiperOrigin-RevId: 174077490 | 09 November 2017, 14:17:10 UTC |
| 045c873 | Deepmind | 31 October 2017, 17:57:24 UTC | In BatchApply, when merged batch dimension can be inferred, then do so. PiperOrigin-RevId: 174059838 | 09 November 2017, 14:17:04 UTC |
| 7517d2e | Deepmind | 30 October 2017, 18:23:18 UTC | Migrate to `absl` open-source library. PiperOrigin-RevId: 173921337 | 09 November 2017, 14:16:57 UTC |
| 87c1572 | Deepmind | 30 October 2017, 12:34:53 UTC | Make merge_leading_dims and split_leading_dim part of the public sonnet API. PiperOrigin-RevId: 173879037 | 09 November 2017, 14:09:27 UTC |
| 869b558 | Deepmind | 30 October 2017, 12:14:54 UTC | Make mask argument to Conv2D more flexible The mask argument to Conv2D needed to be something that is convertible to a numpy array. The numpy mask was then automatically converted to a TensorFlow constant when applied. Because the constants are stored in the graph, this increases the size of graph quite considerably (by the size of the Conv2D filters). If we allow the user to pass in a TensorFlow Tensor instead, it is easily possible to use a PyFunc wrapper around a numpy array, or to cache constants in between calls to Conv2D. This extends Conv2D to be able to handle any mask input that is convertible to a float32 or float64 Tensor. PiperOrigin-RevId: 173877530 | 09 November 2017, 14:09:21 UTC |
| 4245e51 | Deepmind | 30 October 2017, 11:59:02 UTC | Internal change. PiperOrigin-RevId: 173876219 | 09 November 2017, 14:09:15 UTC |
| 20a32c7 | fviola | 30 October 2017, 10:21:53 UTC | Cast input shape to tuple to allow for robust concatenation of shape information PiperOrigin-RevId: 173870205 | 09 November 2017, 14:09:04 UTC |
| 469a933 | tfgg | 12 October 2017, 18:50:49 UTC | Update docstring for snt.custom_getters. PiperOrigin-RevId: 171990392 | 16 October 2017, 14:11:33 UTC |
| d80a554 | Deepmind | 10 October 2017, 14:39:29 UTC | Fixed typo in doc string. PiperOrigin-RevId: 171676263 | 16 October 2017, 14:11:27 UTC |
| 316e453 | Deepmind | 09 October 2017, 21:23:31 UTC | Make `base_info._is_iterable` safer by only returning True for supported types: list, tuple and dict. PiperOrigin-RevId: 171586466 | 16 October 2017, 14:11:22 UTC |
| da32486 | Deepmind | 09 October 2017, 16:20:51 UTC | Make sure module `is_connected` if it's connected to graph using a `@reuse_variables` method other than `_build`. PiperOrigin-RevId: 171543901 | 16 October 2017, 14:11:16 UTC |
| b5bc2e5 | diegolascasas | 09 October 2017, 14:17:55 UTC | Sonnet version update produced on Monday, 09. October 2017 PiperOrigin-RevId: 171531910 | 16 October 2017, 14:11:10 UTC |
| 3999868 | Deepmind | 09 October 2017, 11:13:01 UTC | Avoid recursion when serializing ModuleInfo. PiperOrigin-RevId: 171519454 | 16 October 2017, 14:11:05 UTC |
| 365ddaa | Deepmind | 06 October 2017, 17:07:45 UTC | scale_gradient now handles all float dtypes. PiperOrigin-RevId: 171306048 | 16 October 2017, 14:11:00 UTC |
| 83603ce | Deepmind | 06 October 2017, 15:00:00 UTC | Defun in clip_gradient takes tensor min/max clip value args in forward pass so that they can be used in backward pass. PiperOrigin-RevId: 171292623 | 16 October 2017, 14:10:54 UTC |
| 74ef80e | Deepmind | 05 October 2017, 16:28:47 UTC | Added `class_name` to ModuleInfo. PiperOrigin-RevId: 171163066 | 16 October 2017, 14:10:48 UTC |
| 84f4c4e | Deepmind | 05 October 2017, 11:16:41 UTC | Decorating the _build function with memoize is breaking the connected_subgraph code. This CL fixes this issue. Note sure if this use-case is valid in the first place but let's have this discussion later! PiperOrigin-RevId: 171134926 | 16 October 2017, 14:10:41 UTC |
| a416704 | Deepmind | 04 October 2017, 10:46:41 UTC | Added Sonnet `ModuleInfo` to the "sonnet" graph collection. This allows to keep track of which modules generated which connected sub-graphs. This information is serialised and available when loading a meta-graph-def. This can be used, for instance, to visualise the TensorFlow graph from a Sonnet perspective. PiperOrigin-RevId: 170989708 | 16 October 2017, 13:57:39 UTC |
| f5633ca | arahuja | 28 September 2017, 23:25:42 UTC | Allow ConvNet to use NCHW data_format PiperOrigin-RevId: 170415564 | 16 October 2017, 13:49:36 UTC |
| 9743f76 | arahuja | 27 September 2017, 12:02:27 UTC | Fix stride property on Conv2D for NCHW inputs. PiperOrigin-RevId: 170182706 | 16 October 2017, 13:43:54 UTC |
| b323508 | mareynolds | 27 September 2017, 09:00:49 UTC | Add docfix to open source doc. PiperOrigin-RevId: 170170288 | 16 October 2017, 12:28:39 UTC |
| 5d31e36 | tfgg | 25 September 2017, 18:18:57 UTC | Clean up and improve example text for snt.custom_getters.Context. Add `verbose` logging mode for easier debugging. PiperOrigin-RevId: 169934019 | 16 October 2017, 11:42:31 UTC |
| 7da90b8 | diegolascasas | 25 September 2017, 13:53:45 UTC | Remove deprecated functions in shakespeare example. PiperOrigin-RevId: 169901199 | 25 September 2017, 15:06:05 UTC |
| 57e91bc | mareynolds | 25 September 2017, 13:17:14 UTC | Update changelog for 1.13 PiperOrigin-RevId: 169898474 | 25 September 2017, 13:53:55 UTC |
| 236de5a | diegolascasas | 25 September 2017, 13:16:49 UTC | Sonnet version update produced on Monday, 25. September 2017 PiperOrigin-RevId: 169898437 | 25 September 2017, 13:53:50 UTC |
| 21f552d | mareynolds | 25 September 2017, 13:13:31 UTC | Clarify readme example. PiperOrigin-RevId: 169898200 | 25 September 2017, 13:53:44 UTC |
| a27371f | Deepmind | 18 September 2017, 17:20:16 UTC | Finalise the deprecation of batch norm support in snt.LSTM. snt.LSTM and snt.BatchNormLSTM are now separate implementations. PiperOrigin-RevId: 169107225 | 25 September 2017, 13:53:34 UTC |
| 2a00d7c | mareynolds | 18 September 2017, 13:37:20 UTC | Internal change PiperOrigin-RevId: 169081550 | 18 September 2017, 14:01:18 UTC |
| 0d57ead | mareynolds | 18 September 2017, 13:18:06 UTC | Update changelog for 1.12 PiperOrigin-RevId: 169080040 | 18 September 2017, 14:01:12 UTC |
| 7957296 | diegolascasas | 18 September 2017, 13:11:42 UTC | Sonnet version update produced on Monday, 18. September 2017 PiperOrigin-RevId: 169079587 | 18 September 2017, 14:01:07 UTC |
| 4f0321b | mareynolds | 15 September 2017, 17:21:38 UTC | Provide warning in Sequential.get_variables() PiperOrigin-RevId: 168851699 | 18 September 2017, 14:01:01 UTC |
| 8c61488 | tfgg | 15 September 2017, 13:17:27 UTC | Create snt.custom_getters sub-package. PiperOrigin-RevId: 168826905 | 18 September 2017, 14:00:56 UTC |
| 6aca22e | Deepmind | 08 September 2017, 16:21:12 UTC | Added dilated convolution option to ConvLSTM. Dilated convolutions are useful when you want a larger receptive field in fewer number of layers. Since usually only a few ConvLSTM layers are ever stacked together, adding the option for dilated convolutions in the LSTM seems especially appropriate. Due to the underlying conv operation already accepting a dilation rate keyword argument, only minor changes are required for this functionality. Added test for instantiating, connecting, and running a ConvLSTM with the dilated convolution option. PiperOrigin-RevId: 168005030 | 18 September 2017, 14:00:50 UTC |
| eb8feb0 | Deepmind | 08 September 2017, 11:11:14 UTC | conv module: use `is_compatible_with` for type checks to support tf.float32_ref type inputs PiperOrigin-RevId: 167981266 | 18 September 2017, 14:00:44 UTC |
