Tensorflow is well known library for Deep Learning created by Google. Recently Tensorflow moved from 1.0 to 1.2. In this journey there we many architectural changes in Tensorflow library. I have gone through many Github libraries that cannot be ran because of version differences. I hope all major changes are done and future releases are more functional additions than incompatible refactoring.
Recently Tensorflow also added support for Java, C and Go lang. I can guess some reasons but not sure if Google can confirm so.
Java is a good additional. A lot of Enterprise infrastructure supports Java. It would be very useful to train the model in Python and load the same in Java. Normally Data Scientist and Deep Learning Engineers more familiarity with Python hence training is very production part of Model Development. However running the same model in production is better with Java and could give nicer edge to Tensorflow to its adoption.
Adding Go is on the same line as Java however the support seems more fine tuned to people who would like to run Deep Learning model at a very large scale. It could be a lot of savings in terms of time and infrastructure cost.
For addition of C Language was little surprise to me. However it look a little time to understand why. I am not iOS developer however I did play around with some development for iPad. Hence I believe addition of C Language was to ensure model trained on Python could run on iOS which is second largest mobile market after Android.
I am surprised to see there is no support for R language so far, but I hope it will be added soon enough. Also Microsoft is supporting R language very well so it could be tug of war between these giants to understand and build next platform and which language to choose.
Concluding supporting a lot of languages is not very new. Every big data framework is trying to be language independent atleast access and productionalizing point of view.
Recently Tensorflow also added support for Java, C and Go lang. I can guess some reasons but not sure if Google can confirm so.
Java is a good additional. A lot of Enterprise infrastructure supports Java. It would be very useful to train the model in Python and load the same in Java. Normally Data Scientist and Deep Learning Engineers more familiarity with Python hence training is very production part of Model Development. However running the same model in production is better with Java and could give nicer edge to Tensorflow to its adoption.
Adding Go is on the same line as Java however the support seems more fine tuned to people who would like to run Deep Learning model at a very large scale. It could be a lot of savings in terms of time and infrastructure cost.
For addition of C Language was little surprise to me. However it look a little time to understand why. I am not iOS developer however I did play around with some development for iPad. Hence I believe addition of C Language was to ensure model trained on Python could run on iOS which is second largest mobile market after Android.
I am surprised to see there is no support for R language so far, but I hope it will be added soon enough. Also Microsoft is supporting R language very well so it could be tug of war between these giants to understand and build next platform and which language to choose.
Concluding supporting a lot of languages is not very new. Every big data framework is trying to be language independent atleast access and productionalizing point of view.
No comments:
Post a Comment