Closed HedgehogCode closed 6 years ago
Since TensorFlow 1.5 is out for some time I updated this PR to use TensorFlow 1.5.
The Travis build failed because it the right version of the GNU Standard C++ Library isn't available.
Edit: TensorFlow 1.5 doesn't seem to work on Ubuntu 14.04 because it is built on 16.04 and depends on libraries not available for 14.04. (Travis uses Ubuntu 14.04)
(FYI: TensorFlow 1.6 restores support for Ubuntu 14.04)
Thanks @HedgehogCode! And thanks @asimshankar for the heads up. I merged this PR without the TF1.5 update. And now I'll look at #6, which upgrades to TF1.6, which hopefully avoids the issue.
@HedgehogCode Quick question for you: are you reasonably happy with the current API now? Can we go to 1.0.0
? If we do that, it doesn't mean we'll never break the API—it is just a promise that we will try to deprecate things first when possible, and break things infrequently. We can still add plenty of new things as well and increment the minor version. What do you think? If you agree, I'll bump to 1.0.0-SNAPSHOT
and then release 1.0.0
.
Update to TensorFlow 1.4.
This PR doesn't break our API but the TensorFlow API is different. For example:
Tensor.create(DataType, long[], ByteBuffer)
is nowTensor.create(Class, long[], ByteBuffer)
.I added some functions to the Tensors class which provide typesafe creation of tensors. (And unit tests for them)