Why is tf.concat much slower than np.concatenate and the “append” of the native python list? Should we avoid if possible?

Question

I\'m implementing a math library using Tensorflow, which internally calls tf.concat quite often. I was curious about the speed of tf.concat. For te