author: @dai
Problem Overview
Find the text which is accepted by the given RNN model.
The model contains one RNNCell cell and one Linear out. At first the hidden vector $h \in \mathbb{R}^{520}$ is None and this is equivalent to the zero vector np.zeros(520). After processing the input text, $h$ becomes something and the script shows "Congrats!" only when out(h).sigmoid() > 0.5.
Solution
You notice these weights and biases are unnatural. In fact, I didn't run any training code to build them.