Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Microsoft open-sources Graph Engine

graphengine.io

Post reply on HN
Print view View original

21–30 of 192 posts

Re: Microsoft open-sources Graph Engine

#21

Post by adamnemecek »

mtrn wrote: post #15
adamnemecek wrote: post #4

I've been running into the idea of computational graphs a lot recently. It's at the core of Tensorflow (and NN in general) but it also comes up for example in Apple's AVFoundation where all audio processing happens in a graph of audio units. Does anyone know what's the theoretical foundation of computational graphs? EDIT: I've created a wiki page for computational graphs. https://en.wikipedia.org/wiki/Computational_G…

There is the biological analogue, which has inspired neural networks: > Recall that in our general definition a feed-forward neural network is a computational graph whose nodes are computing units and whose directed edges transmit numerical information from node to node. > Each computing unit is capable of evaluating a single primitive function of its input. In fact the network represents a chain of function composit…

> There is the biological analogue, which has inspired neural network

I'm somewhat aware but it seems like the idea of a computational graph is the most generic computational idea I can think of and I'm surprised it's not more explored.

> Another ancestor would be the Data-Flow paradigm:

Oh yeah, data flow is definitely another one.

Re: Microsoft open-sources Graph Engine

#22

Post by throwawayish »

dman wrote: post #18
dogma1138 wrote: post #7

Earlier quoted context omitted.

I don't see MSFT benefiting from open sourcing windows, so likely never.

I see them benefiting immensely. A big problem that Microsoft has is that once you start going down rabbit holes on Windows the number of people out there who have a clue is vanishingly small. So for systems programming on Windows there is very little coverage online. Opening Windows up will broaden the number of people who understand the platform on a deep level.

Earlier in my life I did a lot of Windows system-level programming, including some minor driver development, and now I'm practically exclusively on the #nix bunch, sometimes doing similar stuff.

I can't say that there is a disparity. If anything, the parts that are documented are usually better documented in the Windows world, and there are very good books about most components. (To be fair I never read a book on anything unix apart from Lion's). There is a large amount of stuff online for either, but I often have the feeling that correct examples are less frequently found for, esp., Linux.

Re: Microsoft open-sources Graph Engine

#24

Post by Kipters »

WayneBro wrote: post #19
bluejekyll wrote: post #5

At this rate, are we going to see Windows open-sourced? MS is on a roll. My bias since 1996 is being eroded with each OSS release they have, and multi-platform targeted support. I started using VSCode regularly as my main Rust IDE, and I feel dirty for liking it. It's seemless across macOS and Linux.

Not yet. Their next strategy is to try and use ARM (again) to shove UWP down your throat. I love Windows but I seriously will not touch UWP until they loosen up the sandbox restrictions so that I can do regular IPC with a win32 desktop app. You can't even send an HTTP request to a little node.js Web server running on your desktop right now. Fuck that. I'm not buying into it.

Care to elaborate on the HTTP part? You just need the "Internet (client)" capability in your manifest, it's even enabled by default when you create a new project

Moreover, win32 desktop apps can use the AppService mechanism exposed by UWP apps (but cannot host an AppService themselves, just connect to UWP-hosted AppServices)

Re: Microsoft open-sources Graph Engine

#25

Post by dogma1138 »

dman wrote: post #18
dogma1138 wrote: post #7

Earlier quoted context omitted.

I don't see MSFT benefiting from open sourcing windows, so likely never.

I see them benefiting immensely. A big problem that Microsoft has is that once you start going down rabbit holes on Windows the number of people out there who have a clue is vanishingly small. So for systems programming on Windows there is very little coverage online. Opening Windows up will broaden the number of people who understand the platform on a deep level.

Windows System Internals are probably the best documented in the world, MSDN alone is a treasure trove of information and there are countless books on the matter.

And unlike some other platforms what is written tends to match reality, and stay relevant for a long time (for better or for worse).

Re: Microsoft open-sources Graph Engine

#26

Post by throwawayish »

adamnemecek wrote: post #21
mtrn wrote: post #15

Earlier quoted context omitted.

There is the biological analogue, which has inspired neural networks: > Recall that in our general definition a feed-forward neural network is a computational graph whose nodes are computing units and whose directed edges transmit numerical information from node to node. > Each computing unit is capable of evaluating a single primitive function of its input. In fact the network represents a chain of function composit…

> There is the biological analogue, which has inspired neural network I'm somewhat aware but it seems like the idea of a computational graph is the most generic computational idea I can think of and I'm surprised it's not more explored. > Another ancestor would be the Data-Flow paradigm: Oh yeah, data flow is definitely another one.

> I'm somewhat aware but it seems like the idea of a computational graph is the most generic computational idea I can think of and I'm surprised it's not more explored.

ASTs?!

Re: Microsoft open-sources Graph Engine

#27

Post by PatentTroll »

WayneBro wrote: post #19
bluejekyll wrote: post #5

At this rate, are we going to see Windows open-sourced? MS is on a roll. My bias since 1996 is being eroded with each OSS release they have, and multi-platform targeted support. I started using VSCode regularly as my main Rust IDE, and I feel dirty for liking it. It's seemless across macOS and Linux.

Not yet. Their next strategy is to try and use ARM (again) to shove UWP down your throat. I love Windows but I seriously will not touch UWP until they loosen up the sandbox restrictions so that I can do regular IPC with a win32 desktop app. You can't even send an HTTP request to a little node.js Web server running on your desktop right now. Fuck that. I'm not buying into it.

I don't get the outrage over the rumored UWP only windows. There is no indication that MS would drop win32. This is a separate product, likely for a particular market. It's more of an iPad competitor than a Windows desktop replacement. So don't use it if it offends you, that's OK. I don't think this is all that forceful really.

Re: Microsoft open-sources Graph Engine

#28

Post by simias »

throwawayish wrote: post #22
dman wrote: post #18

Earlier quoted context omitted.

I see them benefiting immensely. A big problem that Microsoft has is that once you start going down rabbit holes on Windows the number of people out there who have a clue is vanishingly small. So for systems programming on Windows there is very little coverage online. Opening Windows up will broaden the number of people who understand the platform on a deep level.

Earlier in my life I did a lot of Windows system-level programming, including some minor driver development, and now I'm practically exclusively on the #nix bunch, sometimes doing similar stuff. I can't say that there is a disparity. If anything, the parts that are documented are usually better documented in the Windows world, and there are very good books about most components. (To be fair I never read a book on any…

I've never done kernel work on Windows but on Linux at least you can dig into the source code to figure out how stuff works when the documentation is lacking. I suppose that's what the parent was talking about.

Re: Microsoft open-sources Graph Engine

#29

Post by adamnemecek »

throwawayish wrote: post #26
adamnemecek wrote: post #21

Earlier quoted context omitted.

> There is the biological analogue, which has inspired neural network I'm somewhat aware but it seems like the idea of a computational graph is the most generic computational idea I can think of and I'm surprised it's not more explored. > Another ancestor would be the Data-Flow paradigm: Oh yeah, data flow is definitely another one.

> I'm somewhat aware but it seems like the idea of a computational graph is the most generic computational idea I can think of and I'm surprised it's not more explored. ASTs?!

Yeah, I think that ASTs are only a special case of computational graphs.

Re: Microsoft open-sources Graph Engine

#30

Post by Jasper_ »

adamnemecek wrote: post #21
mtrn wrote: post #15

Earlier quoted context omitted.

There is the biological analogue, which has inspired neural networks: > Recall that in our general definition a feed-forward neural network is a computational graph whose nodes are computing units and whose directed edges transmit numerical information from node to node. > Each computing unit is capable of evaluating a single primitive function of its input. In fact the network represents a chain of function composit…

> There is the biological analogue, which has inspired neural network I'm somewhat aware but it seems like the idea of a computational graph is the most generic computational idea I can think of and I'm surprised it's not more explored. > Another ancestor would be the Data-Flow paradigm: Oh yeah, data flow is definitely another one.

Functional programming can be represented as a graph, where calls are directed edges. Flow control is often analyzed in graph form in real compilers:

https://en.wikipedia.org/wiki/Control_flow_graph

Post reply on HN