Manipulating tensors is not an easy task as it requires a lot of prerequisites, such as keeping track of multiple dimensions, Dtype compatibility, mathematical correctness, and tensor shape. The real challenge is identifying the right TensorFlow operations from the hundreds of options available.
TensorFlow Coder(TF-Coder) makes the tensor manipulation possible without coding and by using examples. TF-Coder helps you write the TensorFlow code. The process is providing input-output examples of the required transformation. Then it finds the TensorFlow that does the transformation. It gives the real TensorFlow code as output.
TF-Coder has many uses some of which are
- Find the right functions to use
- Combine functions in a smart way
- Write correct code with less debugging
At present some of TF-Coders drawbacks are
- Within a minute it can find solutions involving 3 or 4 operations, but when there are more operations then the time taken is long
- Does not support string tensors or complex tensors
- The answer entirely depends on the input-output example provided to the system
Paper: https://arxiv.org/abs/2003.09040
GitHub: https://github.com/google-research/tensorflow-coder
Colab notebook: https://colab.research.google.com/github/google-research/tensorflow-coder/blob/master/TF-Coder_Colab.ipynb
Demo:
AI Marketing Strategy Intern: Young inspired management student with a solid background in engineering and technical know-how gathered through work experience with Robert Bosch engineering and business solutions. As a continuous learner, she works towards keeping up-to date with cutting-edge technologies while honing her management and strategy skills to develop a holistic understanding of the modern tech driven industries and drive developments towards better solutions.
