pred.py

def predict(params, inputs):
  activations = inputs
  for w, b in params[:-1]:
    outputs = np.dot(activations, w) + b 
    activations = np.tanh(outputs)

  final_w, final_b = params[-1]
  logits = np.dot(activations, final_w) + final_b
  return logits - logsumexp(logits, axis=1, keepdims=True)