* Refine img_conv_op
* Add conv_shift layer
* Add tensor layer, refine selective fc
* Try fix some styles
* simple rnn layers
* Add image layers
* util_layers
* Use collection.Sequence instead of list|tuple
* BugFix for print_layer
* Unitest for poolings.
* Add test_grumemory_layer
* add last_first_seq to unittest
* Add test expand layer, test ntm layers
* add *args for outputs
* Add trans layer
* Add test for hsigmoid
* also add interface to set param attr
* sampling_id, linear_comb
* ctc_layer, crf_layer
* cost_layers
* beam_search

Conflicts:
	python/paddle/trainer_config_helpers/layers.py

* Fix a unittest bug before.
  * dotmul inputs' sizes should be equal
* add equation to recurrent_layer
* unify x,y x1,x2 => a,b
* log fatal when lstm/gru set wrong size