-
Notifications
You must be signed in to change notification settings - Fork 4.3k
CNTK_2_0_Beta_7_Release_Notes
This page has migrated to our new site. Please update any bookmarks.
This is a summary of new features delivered with the Beta 7 release of CNTK V.2.0.
In the past, the Python API was very permissive in what it accepted as data input types. This caused a lot of confusion (e.g. non-sequence data being wrongly then interpreted as sequences). Therefore we have made the Python API stricter. In particular:
- Sequences now have to be provided as single NumPy or Scipy sparse arrays. I.e., it is not any more allowed to provide a sequence as a list of NumPy arrays or Scipy sparse arrays.
- The helper function
cntk_devicehas been hidden in the testing modules. For device selection use the modulecntk.device - Python default initialization is set to normal (
std=1) and uniform ([-1,1]). -
gaussianis renamed tonormal.
We have prepared the following Examples and Tutorials:
We have made some changes and improvements in CNTK Backend:
- Changed data distribution in multi-GPU training for the composite reader in the frame mode
- Improved performance in the composite reader to use all available CPU threads during deserialization/transformation
- Random number generator and distributions were updated at many places to be the same across supported platforms (Linux and Windows)
- Backward compatibility note: training (in particular initialization) may not be exactly reproducible with respect to the previous versions
- Implemented an optimization to elide the initial zeroing and subsequent accumulation into the gradients for nodes with just one parent/ancestor node
A new Nuget package with the latest eval DLL (managed and native) is available.
IMPORTANT! In Visual Studio Manage Nuget Packages Window change the default option Stable Only to Include Prerelease. Otherwise the package will not be visible. The Package version should be 2.0-beta7.