[dart-misc] TensorFlow and Dart

[Robert Åkerblom-Andersson]

Hi,

Google just released a new library/project called TensorFlow
<http://tensorflow.org/>, "*an Open Source Software Library for Machine
Intelligence*". I looked into it some this morning with the eyes of a Dart
developer, thinking if it would be possible to use in Dart in the future.

My finding are these:

- Today only C++ and Python is supported
- Python support is provided via a library called "swig
<http://www.swig.org/>"

The homepage of TensorFlow also says this:

*"we’re hoping to entice you to contribute SWIG interfaces to your favorite
language -- be it Go, Java, Lua, Javascript, or R."*


So, it seems to get Dart programs to use SWIG, one would first have to add
support for Dart in SWIG (I have created an issue for that here
<https://github.com/swig/swig/issues/557>), and then add Dart SWIG
interfaces to the TensorFlow project (I have not created an issue for
that). I just thought I'd share my findings here, I don't have the time to
work on this right now, but if someone on the list find Dart support with
TensorFlow interesting this might be a good start.

I also played with the idea of using Dart's own native extensions instead
of SWIG, not sure if that would be a better approach, if someone with more
knowledge of Dart's Native Extensions have time to share his/her thoughts
it would be interesting to hear potential pros and cons compared to SWIG.
My initial thought is that you would save some work using SWIG since it's
already used by Python, and encouraged by the project for language
bindings, at the same time I'm not sure how much work is involved with the
creation of TensorFlow specific SWIG interface, and/or if Dart's native
extensions simply have other benefits (or maybe Dart's native extensions
should be used together with SWIG even, I don't know...).

Best Regards, Robert

[Anders Holmgren]

Yeah I'd love a dart interface into tensor flow. Given that the tensor flow doco talks about using on mobile then dart would be an even better for than Python I would have thought.

Mind you Python seems very popular in the machine learning world for some reason

Hi,
Google just released a new library/project called TensorFlow, "an Open Source Software Library for Machine Intelligence". I looked into it some this morning with the eyes of a Dart developer, thinking if it would be possible to use in Dart in the future. 
- Today only C++ and Python is supported
- Python support is provided via a library called "swig"
The homepage of TensorFlow also says this: 
"we’re hoping to entice you to contribute SWIG interfaces to your favorite language -- be it Go, Java, Lua, Javascript, or R."
So, it seems to get Dart programs to use SWIG, one would first have to add support for Dart in SWIG (I have created an issue for that here), and then add Dart SWIG interfaces to the TensorFlow project (I have not created an issue for that). I just thought I'd share my findings here, I don't have the time to work on this right now, but if someone on the list find Dart support with TensorFlow interesting this might be a good start. 
I also played with the idea of using Dart's own native extensions instead of SWIG, not sure if that would be a better approach, if someone with more knowledge of Dart's Native Extensions have time to share his/her thoughts it would be interesting to hear potential pros and cons compared to SWIG. My initial thought is that you would save some work using SWIG since it's already used by Python, and encouraged by the project for language bindings, at the same time I'm not sure how much work is involved with the creation of TensorFlow specific SWIG interface, and/or if Dart's native extensions simply have other benefits (or maybe Dart's native extensions should be used together with SWIG even, I don't know...).
Best Regards, Robert