I am looking at the Martin Wicke talk. The Estimator API is very reminiscent of SparkML. Nice to see that the tensorflow crew are flexible enough to take good i

I am looking at the Martin Wicke talk. The Estimator API is very reminiscent of SparkML. Nice to see that the tensorflow crew are flexible enough to take good ideas from projects such as SparkML and Keras (now included natively in the TF stack). Other highlights include the hotspot compiler (I was not that impressed so far, but it's early days for them), and embedded visualizations (looked quite cool) for visually inspecting learnt manifolds.