Swift for TensorFlow - Tanmay Bakshi - Advanced Spark and TensorFlow Meetup - June 17, 2019
2 likes1,246 views
The document discusses Tanmay Bakshi, a TED speaker and author, who focuses on using Swift for TensorFlow, highlighting its benefits such as performance, flexibility, and a gentle learning curve. It emphasizes that TensorFlow is not inherently Pythonic, but efforts are being made to improve this aspect. The text also contains examples of code and various functionalities related to Swift and TensorFlow optimization.
Swift for TensorFlow - Tanmay Bakshi - Advanced Spark and TensorFlow Meetup - June 17, 2019
- 1. Tanmay Bakshi TED Speaker, Author, Algorithm-ist Swift for TensorFlow
- 4. First-class compiler support for Machine Learning
- 6. 1 1 + 𝑒$% 𝑤1 = 1 𝑤3 = 𝑤1 ⋅ −𝑤2 𝑤3, 𝑤5 = 𝑤3 𝑤7 = 𝑤5 ⋅ 𝑤5 ⋅ ln 𝑤6 𝑤8 = −𝑤7
- 10. Julia Other
- 11. Why change?
- 12. Performance & optimizations Flexibility & Control Rapid Prototyping Interoperability & reverse-compatibility Gentle learning curve
- 13. TensorFlow is fundamentally not Pythonic. But we’re trying!
- 14. import as 10 8 3 3 10 def return for in 1 10000 with as -0.1 print
- 15. import var ref Float 10 8 var guess Float 3 3 var distance Float 10 func error Tensor Float Tensor Float Tensor Float Tensor Float let return for _ in 1 10000 let Tensor Float in return -0.1 Optimized for you.
- 17. It’s all in your hands.
- 18. Write code naturally, Swift for TensorFlow takes care of the rest.
- 19. import extension Tensor where Scalar TensorFlowFloatingPoint func squaredError Tensor Scalar Tensor Scalar return self
- 20. 1.05x Circle-edge finding 0.95x MLPNN Training Without Compiler Optimizations!
- 23. import let np "numpy" let p1 2 -2 let p2 10 8 func norm PythonObject PythonObject return 2 let ex
- 24. Code time!