I've found myself putting off retraining because it takes a little time. I want to do it more often, but I wait until I hit a milestone because otherwise it interrupts my flow. I don't have a good sense of how important retraining is or how much difference it makes each time, but my completionist nature makes me uncomfortable when I know it's out of date. If it could happen in the background so I didn't have to think about it, that would reduce my mental load.
I build developer tools and services at Microsoft (currently Codespaces, Live Share, IntelliCode) and maintain some OSS projects (CodeTour, GistPad, CodeSwing, WikiLens)
We’re actually working on support for automatic training, so it’s great to hear this would be valuable to you! Out of curiosity: what would be your ideal experience for training in the background? Periodically running on your local machine? Running as part of a CI/nightly build? Something else?
Also, have you tried sharing a custom model with your team as well? Or are folks individually training their local copies of the project?
I've not used it on any team projects, so I can't speak to that.
I don't know how valuable it is to retrain after small changes to the code. Obviously if small changes don't matter then it can be rarer, and if they do then it should be more frequent. I would like to feel like it's always up to date and Just Workstm without having to worry about that. To that end, I would like to have it updating on my local machine as changes happen, at whatever frequency makes sense.
I don't know what analysis is being done, but could it be iterative as changes happen or does it need to do a full rebuild every time?
My solution that comes in at around 5,000 lines takes about a minute to train. If that number is much larger for bigger solutions, then having the ability to train it on a server through CI or nightlies and pull it down from clients is important. I would like to see both workflows be supported and complementary.
I build developer tools and services at Microsoft (currently Codespaces, Live Share, IntelliCode) and maintain some OSS projects (CodeTour, GistPad, CodeSwing, WikiLens)
I can’t thank you enough for sharing this detail! Would you be open to setting up a quick chat sometime this week? You’re giving some amazing feedback, and I know that our team would love to learn further about your experiences 😁 If so, feel free to ping me (joncart@microsoft.com) and we can go from there 👍
I'd be happy to, but I'm not sure that I have that much great insight. I've been using it casually on a medium sized personal project - a suite of Roslyn analyzers - but I've not thought about it in too much detail, other than what I've posted here. Would it be easier if I just keep posting things here when I think of them?
I build developer tools and services at Microsoft (currently Codespaces, Live Share, IntelliCode) and maintain some OSS projects (CodeTour, GistPad, CodeSwing, WikiLens)
I've found myself putting off retraining because it takes a little time. I want to do it more often, but I wait until I hit a milestone because otherwise it interrupts my flow. I don't have a good sense of how important retraining is or how much difference it makes each time, but my completionist nature makes me uncomfortable when I know it's out of date. If it could happen in the background so I didn't have to think about it, that would reduce my mental load.
We’re actually working on support for automatic training, so it’s great to hear this would be valuable to you! Out of curiosity: what would be your ideal experience for training in the background? Periodically running on your local machine? Running as part of a CI/nightly build? Something else?
Also, have you tried sharing a custom model with your team as well? Or are folks individually training their local copies of the project?
I've not used it on any team projects, so I can't speak to that.
I don't know how valuable it is to retrain after small changes to the code. Obviously if small changes don't matter then it can be rarer, and if they do then it should be more frequent. I would like to feel like it's always up to date and Just Workstm without having to worry about that. To that end, I would like to have it updating on my local machine as changes happen, at whatever frequency makes sense.
I don't know what analysis is being done, but could it be iterative as changes happen or does it need to do a full rebuild every time?
My solution that comes in at around 5,000 lines takes about a minute to train. If that number is much larger for bigger solutions, then having the ability to train it on a server through CI or nightlies and pull it down from clients is important. I would like to see both workflows be supported and complementary.
I can’t thank you enough for sharing this detail! Would you be open to setting up a quick chat sometime this week? You’re giving some amazing feedback, and I know that our team would love to learn further about your experiences 😁 If so, feel free to ping me (joncart@microsoft.com) and we can go from there 👍
I'd be happy to, but I'm not sure that I have that much great insight. I've been using it casually on a medium sized personal project - a suite of Roslyn analyzers - but I've not thought about it in too much detail, other than what I've posted here. Would it be easier if I just keep posting things here when I think of them?
Sure! That would be great
I will do!