3 ms·
I asked my friend who is both a web developer and does ML if he preferred pytorch to tensorflow and he confirmed what I had read online, which was that pytorch
by poletopole 6y ago
I asked my friend who is both a web developer and does ML if he preferred pytorch to tensorflow and he confirmed what I had read online, which was that pytorch is better for parallel ML and other use cases that TF isn’t. It doesn’t appear that all TF libraries are made equal. An of course, Google does have a long history of abandoning their open source projects. However, my friend did admit that TF excels better at low level matrix operations.
- swyx 6y agoidk if Google has a history of abandoning OSS, I actually can't name one at the tensorflow level that has been abandoned. you may be thinking about their commercial software which they do shutdown quite a lot. I just think it's not good enough a reason to abandon tensorflow bc of some vague unfounded Google fear.
- poletopole 6y agoI stand corrected. Take Polymer for example. I’m very surprised, despite all the negative commentary and controversy surrounding web components, that Google has remained loyal to their Polymer project and even developed it further into LitElement. I think what happens is that Google will bring an OSS project to market and then after a few years the community takes over, so there’s an illusion of “abandonware”. But it’s hard to say whether Google just moved on to something else first, for example from GWT to Angular, and then the community took over or if it’s the other way around. So there’s this implication that Google does abandon their projects or at least stops funding them even when that’s not really the case, but the implication alone almost becomes a self-fulfilling prophecy. In fact, at work I suggested we even do a project with LitElement over React and the reason the team went with React was because they thought Google would abandon the project. ¯\_(ツ)_/¯