Open ladykillerjason opened 6 years ago
This does not seem correct to me. Don't you also need do = np.tanh(self.state.s) * top_diff_h ?
do = np.tanh(self.state.s) * top_diff_h
This does not seem correct to me. Don't you also need
do = np.tanh(self.state.s) * top_diff_h
?