@365andreas and I have found an issue while running the Baum-Welch algorithm for HMM with TensorFlow 1.15. The tf.reverse() for backward function was called in a wrong way (since it referred to an older version of TensorFlow (see here)) leading to wrong results. We have changed its call.
@365andreas and I have found an issue while running the Baum-Welch algorithm for HMM with TensorFlow 1.15. The tf.reverse() for backward function was called in a wrong way (since it referred to an older version of TensorFlow (see here)) leading to wrong results. We have changed its call.