search

eglassman / prose

Microsoft Program Synthesis using Examples SDK is a framework of technologies for the automatic generation of programs from input-output examples. This repo includes samples and sample data for the Microsoft Program Synthesis using Example SDK.
https://microsoft.github.io/prose/
Other
2 stars 0 forks source link

Understand JSON Transformation to experiment with learning based off of parse trees #14

Open Steven-Hewitt opened 7 years ago

Steven-Hewitt commented 7 years ago

I've looked into JSON Extraction already, and it seems far too simple to do what we want; the most advanced a program can get is something like a SQL table join where each input region is treated like a set of rows, and I'm failing to find any indication that it's capable of more advanced things (even JsonRegions, while it seemed to be similar to what we wanted, it doesn't seem as fully-featured as we'd need and, in addition, the given samples don't use it at all so it'll take a while to figure out). JSON Transformation, on the other hand, seems capable of manipulating the data into something more workable (and determining which data to manipulate, which is why I'm very much more hopeful about our prospects with that), with relatively less effort. Either way, the raw text seems like a better prospect for us to use for as long as we possibly can.

eglassman commented 7 years ago

Thanks for looking into this! Gustavo is also going to try to change his Python-processing DSL to handle JS. Hopefully by Friday. If that fails, I guess we're stuck with text processing on raw or parsed code. I'm creating labels by hand for the undergrads at the moment.

On Wed, Feb 15, 2017, 4:35 AM Steven Hewitt notifications@github.com wrote:

I've looked into JSON Extraction already, and it seems far too simple to do what we want; the most advanced a program can get is something like a SQL table join where each input object is treated like a set of rows, and I'm failing to find any indication that it's capable of more advanced things. JSON Transformation, on the other hand, seems capable of manipulating the data into something more workable (and determining which data to manipulate, which is why I'm very much more hopeful about our prospects with that). Either way, the raw text seems like it's going to be much better for what we're trying to do.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/eglassman/prose/issues/14, or mute the thread https://github.com/notifications/unsubscribe-auth/ACNnQjG5-aJ0-XPBwLC69Z6zU7Na4EzYks5rcvEdgaJpZM4MBqbW .