Tensorflow 2.0: Google Got More Pythonic

“Just an honorary title, but I get early prototype access and travel costs!”


Veysel Kocaman, Google Developer Expert in ML.

https://www.linkedin.com/in/vkocaman/


“Tensorflow (TF) might have outplayed PyTorch.”


“Pytorch is a lot different than TF. Easy to use, popular, but not as industrially full-fledged.” “Learning curve for TF 1.0 was very high. So many headaches relieved in 2.0.”


“TF is fully integrated with Google’s ecosystem, so it’s easier to integrate into DL pipelines.” “Python is awesome. Very well geared to people who don’t have formal education.”


Veysel tries to explain what ‘Pythonic’ means. It describes a mantra, an ethos, an MO, a mindset – I think it’s spiritual? “Keras is like the python of deep learning.”


“Keras supports any features uploaded by devs, so now TF will support them.”


You might say Keras helps TF become more pythonic. Less high-level maths means less friction for use. The community will grow much faster, now.


We talk about the huge demand for simplicity. “Google autoML is very strong… mostly plug and play.” “Otherwise, you still have to tune parameters to data.” (For now…)


“Google sees what goes on, waits for the best tools to rise, then integrates them into their ecosystem – so users can access all the other services.”


“I should state, my ideas don’t reflect those of Google.”

3 views0 comments

Recent Posts

See All

Machine learning will leave no stone unturned

  • LinkedIn

View Privacy Policy  © Machine Commons 2018