4 ms·

Tensorflow is a proof that some people in Google don't understand software engineering: They don't know and can't decide even what interface they want to expose

Tensorflow is a proof that some people in Google don't understand software engineering: They don't know and can't decide even what interface they want to expose/hide. It is a disaster.

gradpanda123 7y ago
It’s not that they don’t know how to be engineers, it’s that politically they all want to own their “thing” in TF. This type of exchange is pretty much what happens all the time: https://github.com/williamFalcon/pytorch-lightning/issues/35 https://github.com/williamFalcon/pytorch-lightning/issues/35 There is no reason why two things can’t be merged together for better UX and functionality, but two humans have egos, want to lead projects, and be in charge, so things remain separate.
- tanilama 7y ago
  I don'think that dispute what I have just said. Software Engineering isn't just about put stuff out there, it also involves careful planning and roadmap around abstractions. As high profile as a project as TF, if they don't have a clear way to tell having many parallel implementations for essentially the same functionality is confusing and hurt developer experience and ultimately harms the project itself, then that only explains the disaster , as a result of mismanagement, but doesn't undo the damage.
  - streetcat1 7y ago
    Not in a land grab. TF is crucial to google dominance in AI.
- blotter_paper 7y ago
  Oh god, those last three comments are so cringey. They both want to merge their projects together, but can't agree on which namespace to use? Somebody needs to go Solomon on these kids and announce that a complete rewrite from scratch will take place unless they can agree on a namespace, then hand over control of the merge to whichever team cedes the point first.
  - p1esk 7y ago
    Oh I bet there will be plenty of people eager to rewrite everything from scratch.
- brown9-2 7y ago
  What does merging two Pytorch related libraries have to do with Tensorflow?
- wfalcon 7y ago
  I totally sympathize with this. As the creator of PyTorch Lightning I focused on the user experience FIRST. Too many times, it seems, engineers forget that NOT everyone using these frameworks are engineers. It really is a fine balance between being opinionated but no more. But as mentioned here, even though Lightning has seen dramatic adoption, there were a few times when the result was judged on the simplicity of it without considering the significant amount of time that went into making the API super simple and infinitely flexible. At least in the PyTorch world, the team's done a phenomenal job at allowing an ecosystem of related projects to contribute to one another and co-exist!
streetcat1 7y ago
They do, but they were under a time constraint to release in order to gain market share. The goal of TensorFlow is to put itself in as many ML deployments as possible, paving the way for auto ml, which is where the real money lies.
reiinakano 7y ago
but they can all invert a binary tree on a whiteboard, so all good

Tensorflow is a proof that some people in Google don't understand software engineering: They don't know and can't decide even what interface they want to expose/hide. It is a disaster.

gradpanda123 7y ago

It’s not that they don’t know how to be engineers, it’s that politically they all want to own their “thing” in TF. This type of exchange is pretty much what happens all the time: https://github.com/williamFalcon/pytorch-lightning/issues/35 https://github.com/williamFalcon/pytorch-lightning/issues/35 There is no reason why two things can’t be merged together for better UX and functionality, but two humans have egos, want to lead projects, and be in charge, so things remain separate.

tanilama 7y ago
I don'think that dispute what I have just said. Software Engineering isn't just about put stuff out there, it also involves careful planning and roadmap around abstractions. As high profile as a project as TF, if they don't have a clear way to tell having many parallel implementations for essentially the same functionality is confusing and hurt developer experience and ultimately harms the project itself, then that only explains the disaster , as a result of mismanagement, but doesn't undo the damage.
- streetcat1 7y ago
  Not in a land grab. TF is crucial to google dominance in AI.
blotter_paper 7y ago
Oh god, those last three comments are so cringey. They both want to merge their projects together, but can't agree on which namespace to use? Somebody needs to go Solomon on these kids and announce that a complete rewrite from scratch will take place unless they can agree on a namespace, then hand over control of the merge to whichever team cedes the point first.
- p1esk 7y ago
  Oh I bet there will be plenty of people eager to rewrite everything from scratch.
brown9-2 7y ago
What does merging two Pytorch related libraries have to do with Tensorflow?
wfalcon 7y ago
I totally sympathize with this. As the creator of PyTorch Lightning I focused on the user experience FIRST. Too many times, it seems, engineers forget that NOT everyone using these frameworks are engineers. It really is a fine balance between being opinionated but no more. But as mentioned here, even though Lightning has seen dramatic adoption, there were a few times when the result was judged on the simplicity of it without considering the significant amount of time that went into making the API super simple and infinitely flexible. At least in the PyTorch world, the team's done a phenomenal job at allowing an ecosystem of related projects to contribute to one another and co-exist!

streetcat1 7y ago

They do, but they were under a time constraint to release in order to gain market share. The goal of TensorFlow is to put itself in as many ML deployments as possible, paving the way for auto ml, which is where the real money lies.

reiinakano 7y ago

but they can all invert a binary tree on a whiteboard, so all good