What is a good way to test a bidirectional recurrent neural network?

Question

I wrote an implementation of a bidirectional RNN. How can I verify that this implementation is correct by using a small dataset?

Is there any simple test that compares performance of bidirectional RNN to RNN.

Answer 1

I ended up using reber grammar test. BLSTM's can converge in fewer iterations as compared to LSTM on the embedded reber grammar test.