- 24 6月, 2016 31 次提交
-
-
由 A. Unique TensorFlower 提交于
Change: 125783336
-
由 A. Unique TensorFlower 提交于
Change: 125783060
-
由 A. Unique TensorFlower 提交于
Change: 125780396
-
由 A. Unique TensorFlower 提交于
Fix a flaky test by computing half-precision gamma samples with float precision intermediate calculations. Change: 125776264
-
由 Pete Warden 提交于
Change: 125761757
-
由 A. Unique TensorFlower 提交于
Change: 125752088
-
由 A. Unique TensorFlower 提交于
Change: 125738602
-
由 A. Unique TensorFlower 提交于
Change: 125737177
-
由 A. Unique TensorFlower 提交于
Change: 125735275
-
由 Benoit Steiner 提交于
Also updated the weight initialization to ensure that the initial weights are likely to be greater than the smaller possible value representable using a 16 bit float (~0.00006). Change: 125733195
-
由 A. Unique TensorFlower 提交于
Change: 125729946
-
由 A. Unique TensorFlower 提交于
Change: 125729599
-
由 A. Unique TensorFlower 提交于
Change: 125729413
-
由 Eugene Brevdo 提交于
One TODO for DirichletMultinomial. Change: 125729016
-
由 A. Unique TensorFlower 提交于
RE the bugfix, I noticed that my tf.slim loss metric is worse than my training loss, and I'm guessing it's because my model is evaluated concurrent with its variables being loaded from a checkpoint. I'm running now to see if this fixes things. Feel free to ignore this CL until then. Change: 125728083
-
由 A. Unique TensorFlower 提交于
Change: 125724750
-
由 A. Unique TensorFlower 提交于
Change: 125722665
-
由 A. Unique TensorFlower 提交于
Change: 125720510
-
由 Mustafa Ispir 提交于
Added max_steps into training. Distributed training will use max_steps instead of incremental steps. Change: 125719014
-
由 Dan Mané 提交于
Change: 125714059
-
由 A. Unique TensorFlower 提交于
Change: 125713863
-
由 Wei Ho 提交于
Change: 125712724
-
由 A. Unique TensorFlower 提交于
Change: 125704382
-
由 Martin Wicke 提交于
Add more unit tests for BaseEstimator/Estimator (and one for tensor_signature). Fix bug if no targets used in training. Factor out some repeated code for better test coverage. Change: 125701861
-
由 A. Unique TensorFlower 提交于
Change: 125701630
-
由 Benoit Steiner 提交于
Upgraded Eigen to the latest version that provides new scan operations. This will enable the implementation of the cumsum operation in TensorFlow Change: 125697517
-
由 A. Unique TensorFlower 提交于
Write code to build and register entrywise unary transforms. Create transforms for unary TF arithmetic ops. Change: 125691817
-
由 Alexey Surkov 提交于
In some cases TensorFlow reads the data via RandomAccessFile in really small chunks, which doesn't work very efficiently with HTTP requests. Adding a read-ahead cache significantly boosts the performance. Change: 125691397
-
由 A. Unique TensorFlower 提交于
non-deterministic. Change: 125689946
-
由 A. Unique TensorFlower 提交于
Change: 125684645
-
由 Eugene Brevdo 提交于
Change: 125684232
-
- 23 6月, 2016 9 次提交
-
-
由 Pete Warden 提交于
Change: 125669565
-
由 Pete Warden 提交于
Change: 125660054
-
由 A. Unique TensorFlower 提交于
Change: 125644391
-
由 A. Unique TensorFlower 提交于
Change: 125639609
-
由 A. Unique TensorFlower 提交于
Add tests to assert all update_op tensors returned from metric_ops produce the proper metric value, and fix streaming_auc to do so. Change: 125639012
-
由 A. Unique TensorFlower 提交于
Change: 125637583
-
由 A. Unique TensorFlower 提交于
Change: 125636045
-
由 A. Unique TensorFlower 提交于
Change: 125624938
-
由 A. Unique TensorFlower 提交于
Change: 125624711
-