* it is not necessary, because user could print sequence info
  by ValuePrinter.

* Simplify Evaluator::create logic. Always use ClassRegister to
  create Evaluator.

The chunks of types in excluded_chunk_types will not be counted in
ChunkEvaluator. This is useful for tasks such as SRL, in which chunks of
type V (verb) will not be taken into account in evaluation.

reference or raw matrix (Matrix & or Matrix *)
contextProjectionForward
contextProjectionBackward
contextProjectionBackwardData
contextProjectionBackwardWeight
classificationError
The mul functions would be updated later.

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Fix sparse training for trainer_count=1

For trainer_count=1, the gradient machine is NeuralNetwork, which does not create parameter buf for PARAMETER_GRADIENT for sparse update in Parameter::enableType. But gradient parameter buf is still used in SgdThreadUpdater.

* Minor update to comment

* try to fix bug for ctc_error_evaluator

Change-Id: Id66da7b7f5bf9ec80cc19b347e4fb822a5a6f197

* by default, synchronize default_stream after resizeAndCopyFrom
* add sync in some places after resizeAndCopyFrom using other streams

ISSUE=4592951

git-svn-id: https://svn.baidu.com/idl/trunk/paddle@1447 1ad973e4-5ce8-4261-8a94-b56d1f490c56

Also add some comments for cost layers.
ISSUE=4580653

git-svn-id: https://svn.baidu.com/idl/trunk/paddle@1410 1ad973e4-5ce8-4261-8a94-b56d1f490c56

ISSUE=4586495

git-svn-id: https://svn.baidu.com/idl/trunk/paddle@1408 1ad973e4-5ce8-4261-8a94-b56d1f490c56