How to build Iris classification network?

[thegoldenmeans0]

Hello, everyone!

I try to make iris classification program.

But, error value can not be converged as below.

I think that structure of last node is wrong in my neural network.In details,I think v11, v12, v13 nodes are wrong.

But I have no idea how to build correctly.

Last node is build by following source code(Whole program is here).

----------------------------------------------------------------------------------------------------------------------------

int y_label_setosa = 0; //: setosa

int y_label_versicolor = 1; //: versicolor

int y_label_virginica = 2; //: virginica

dynet::Expression loss_expr_setosa = dynet::pickneglogsoftmax(y_pred, y_label_setosa); // Output layer (setosa)

dynet::Expression loss_expr_versicolor = dynet::pickneglogsoftmax(y_pred, y_label_versicolor); // Output layer (versicolor)

dynet::Expression loss_expr_virginica = dynet::pickneglogsoftmax(y_pred, y_label_virginica); // Output layer (virginica)

----------------------------------------------------------------------------------------------------------------------------

Additionally, I want to build following network.

Please tell me how to build classification network correctly or give me some hints.

If you need additional information, feel free reply this thread.

Thanks!