U tweetove putem weba ili aplikacija drugih proizvođača možete dodati podatke o lokaciji, kao što su grad ili točna lokacija. Povijest lokacija tweetova uvijek možete izbrisati. Saznajte više
The imperative, feels-like-programming nature of PyTorch, in contrast to more declarative or functional frameworks, seems to be a significant factor in the popularity. Research code tends to live in the small scale where side effect laden code isn’t causing intractable problems.
I am an advocate for the benefits of functional programming, but I do tend to think that there is something deeper than just a cultural bias towards imperative programming in people — imperative actions are probably more broadly “natural”.
Declarative frameworks didn’t take off (i think) because the tooling wasn’t polished enough to do rapid iteration cycles (code-debug-code-debug). Theano and TF1 were simple VMs with early debuggers, non-extensible types (only Tensor) and limited docs (tf.scan vs python for loop)
We failed with declarative for ML long ago ... recently gotten one \eps used (Overton/Apple, https://arxiv.org/pdf/1909.05372.pdf …) similar to @w4nderlus7's awesome Ludwig https://uber.github.io/ludwig/ .
IMO declarative helpful when many types of users and model coding not main challenge, c.f. SQL
I think the “imperative” aspect that matters the most is the eagerness - when you multiply two tensors, you don’t get some object symbolically representing the operation, you get the actual result of the operation and you can print it and see its contents without any extra steps.
As someone who really enjoys functional programming (Haskell programmer), I think think pytorch trumps tensorflow's interface mainly through being simpler and less 'enterprisey' (less bullshit in general to get something simple done).
What on Earth does "feels-like-programming" mean? It's programming. It literally is programming.
It you have all the components ready in declarative programming you just compose them it doesn't feel like programming in a normal (imperative) sense.
It seems like functional stuff is pretty valuable as embedded nuggets in a bigger imperative framework. It would be cool if more academic implementations were presented as bite size nuggets.
A big example could be vertex and pixel shaders (“functions”) embedded in complex sw+hw graphics pipelines which are larger stateful frameworks.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
