[net] subdivisions=1 batch = 256 inputs=256 momentum=0.9 decay=0.0 time_steps=128 learning_rate=.002 adam=1 policy=constant power=4 max_batches=1000000 [gru] output = 1024 [gru] output = 1024 [gru] output = 1024 [connected] output=256 activation=linear [softmax] [cost] type=sse