My first reaction is that an RNN-based code generator trained on a particular dataset (e.g. Plan 9 sources) might be more fruitful. On Mon, Feb 22, 2021 at 3:08 PM William Linkmeyer wrote: > Dearest Nines, > > I’ve been reading these email chains for about three years now. You’ve all > taught me quite a bit. I understand that this will be a bit off-topic as > it’s not directly related to Plan9. So be it. I wanted to start a chain and > chime in on something that I do have experience with. > I’ve been developing a code suggestion platform. The concept is simple > relatively simple. As the programmer is typing, patterns in the given code > block are identified and compared to a database of tagged code snippets. > The code snippet that is most relevant will be presented to the programmer > as a suggestion. > The code snippets all come originally from Stack Overflow, GitHub, and my > own programs. I’m in the process of tagging these snippets for important > features such as purpose, list of operations, any applicable design > patterns, paradigms, etc. The more “essential” features are being tagged as > well — features such as the number of arguments, return type, etc. > It’s an interesting project to me. While I think that it’s going well, I’m > certain that the large-brain folk here have important commentary on it. > > With Peace, > Will > *9fans * / 9fans / see discussions > + participants > + delivery options > Permalink > > ------------------------------------------ 9fans: 9fans Permalink: https://9fans.topicbox.com/groups/9fans/T4f7afadf357a15b1-M37e90a17c9db8faca41e70c4 Delivery options: https://9fans.topicbox.com/groups/9fans/subscription