Vanilla RNN Sequence

Run a vanilla tanh RNN over an input sequence and return every hidden state. X has one input vector per row, h0 is the initial state, and each step computes $h_t=\tanh (W_x{x}_t+W_h{h}_{t-1}+b)h\_t=\backslash tanh(W\_xx\_t+W\_hh\_\{t-1\}+b)$.

Implement solve(X, h0, Wx, Wh, b).

Returns. Return an array of shape (T, hidden); the initial state is not included.

Examples

Example 1

Input:

solve([[1.0], [2.0]], [0.0], [[1.0]], [[0.5]], [0.0])

Output:

[[0.761594], [0.983041]]

Example 2

Input:

solve([[1.0, 0.0], [0.0, 1.0]], [0.0, 0.0], [[1.0, 0.0], [0.0, 1.0]], [[0.0, 0.0], [0.0, 0.0]], [0.0, 0.0])

Output:

[[0.761594, 0.0], [0.0, 0.761594]]