[network] Rework the backwarding
authorParichay Kapoor <pk.kapoor@samsung.com>
Thu, 3 Dec 2020 10:14:07 +0000 (19:14 +0900)
committerJijoong Moon <jijoong.moon@samsung.com>
Thu, 10 Dec 2020 10:20:41 +0000 (19:20 +0900)
commitd9ff9a6f4d31b0cd535631e957343be002e2ec8c
tree0d28b073be4d5d9c2c3a7574fc66e64c5583ac20
parent5afcb52dd90ef0c9fb203ab6c9a721822de7b52f
[network] Rework the backwarding

- remove forwarding from backwarding
backwarding should just do backwarding and no more
- moved backwarding back to neuralnetwork so that graph
does not has to care about how to backward etc.
Graph just provides iterators for iterating the graph
in reverse. Graph does not know that layers have backwarding etc.

Also this removes dependency of graph from optimizer.

V2:
Added comment fixes for the corresponding PR

**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped

Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Applications/ReinforcementLearning/DeepQ/jni/main.cpp
nntrainer/graph/network_graph.cpp
nntrainer/graph/network_graph.h
nntrainer/layers/layer_internal.h
nntrainer/models/neuralnet.cpp
nntrainer/models/neuralnet.h
nntrainer/tensor/weight.h
test/unittest/unittest_nntrainer_graph.cpp
test/unittest/unittest_nntrainer_layers.cpp
test/unittest/unittest_nntrainer_models.cpp