Dude. Tensorflow version changes are so fucking bad. It's even worse with keras because they create an echo chamber for shit. I'm trynna reset a fuckin model here, yet everything throws 99 more errors to the pile. Like, wtf?

***** For stackoverflow enthusiasts: found a solution, don't need your groundbreaking shit either.

Add twist to the ongoing drama: they're sending me thesis to reviewers and they had to announce it to me. Kill me now!

There's always Pytorch I guess...

@AlmondSauce ew

tensorblow

Fuck Tensorflow i quit using it after so many issues. And have you read the source code? Idk why google goes hard with their interviews when the devs write like they don’t give a fuck or don't know how to program. The code examples are worse; global variables all over - which is a BIG flag of spaghetti and buggy logic. Meanwhile, PyTorch’s source code is clean, elegant, and straightforward.

@amoux I have to disagree. I haven't specifically looked at their code, but I know for every trick and tweak, tf gives you some option. That is the whole point. And also why using keras on top of it. It takes care of some of those choices.

I will learn about pytorch eventually I just haven't found the time or reason yet.

@NoMad Yeah, I mean, if you know the tricks, then that's awesome. I want to like tensorflow but I just had many continuous issues (including the issue you mentioned) that I gave up. On the other hand, Keras is elegant and simple - I used it a lot when I first started learning NNs.

The problem with highly mathematical code is, that it tends to get written by mathematicians. They normally aren't good at writing maintainable code - but they are good at math and like to solve math problems...

@Oktokolo Yeah, you also see it with data-scientist a lot. Unused variables here and there, passing objects back and forward like if they were playing basketball.