Hacker News Reader

Tensorflow Developer Experience issues are due to the fast deep learning algorithm evolution when it was designed.

Now that the invariants are known, better abstractions have been designed, such as Keras.

First Keras version was released the 27 March 2015.

First TensorFlow version was released the 9 November 2015.

Both had to follow the evolution of DL, but Keras was developer-friendly from the beginning.

That's a little unfair no? Supporting TF's API is much easier than building TF for Google (and then releasing it to the public).

The API abstraction for current (2.x) Tensorflow literally is Keras.