CNTK_2_0_Beta_9_Release_Notes

Article
10/18/2018

This is a summary of new features delivered with the Beta 9 release of CNTK V.2.0.

Breaking changes

There is a breaking change in this release:

Image Reader is updated with random area option (useful for Inception-style networks), and color transform option.

New and updated features

Lambda rank and NDCG at 1 are accessible from Python for real this time.
Changes in Learner API: for learners that use momentum (momentum_sgd, nesterov_sgd and adam_sgd) it is now possible to specify if the momentum should be applied in the regular fashion or as a unit-gain filter (default). For more details on the unit-gain momentum, please refer to this section.
No more partial minibatches at the sweep boundary. Now minibatches are allowed to transparently cross the sweep boundary. epoch_size parameter for all hyperparameter schedules now defaults to full data sweep (i.e., by default, hyperparameters change their values on the sweep by sweep basis).

New Examples and Tutorials

Deconvolution layer and image auto encoder example using deconvolution and unpooling (Example 07_Deconvolution in Image - Getting Started).
Basic autoencoder with MNIST data.
LSTM Timeseries with Simulated Data (Part A). (More will come in the next Releases)

Python API

The following updates are introduced to Python API:

Default Python version for binary installation script was changed to 3.5 for both Windows and Linux. As before you can manually select version 2.7, 3.4, or 3.5 during the installation. Please see binary and source setup instructions to find out about how to select Python version.
Docker Hub Runtime Image will also contain Python v. 3.5.
Lambda rank and NDCG at 1 are now accessible from Python.
Preliminary version of the training session API for distributed learning is exposed in Python.

CNTK Evaluation library. NuGet package

A new set of NuGet Packages is provided with this Release.

IMPORTANT! In Visual Studio Manage NuGet Packages Window change the default option Stable Only to Include Prerelease. Otherwise the packages will not be visible. The Package version should be 2.0-beta9.