You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Is it possible update model without retraining? #257
1, I want to update stored model after read it. I don't want to retrain whole data. How can I do this?
2. Can I load data from c# collection or just string for generate model, not file or stream?
Hi~ @TomFinley
Yes. It could be one of the scenarios. I want to update model with newly inserted records in database, not whole records every time when I try to retrain sometimes. It could be real time or not. So, I want to know how can I do this.
And I convert records from database into text file for training. I think this is inefficient. So, I hope I had another options like training from String, Collections. Maybe I could be mistaken about that.
Regarding the first question, no, sorry. The underlying code supports it for a few learners, but this sort of graph composition layer does not. So we're somewhat out of luck at the moment.
Regarding the second question, about working over stuff in memory vs. materializing to disk, do you think #106 helps? Or is that insufficient?
Activity
TomFinley commentedon May 31, 2018
Hi @ecofee just clarifying, are you talking about online learning? Or some other scenario?
ecofee commentedon May 31, 2018
Hi~ @TomFinley
Yes. It could be one of the scenarios. I want to update model with newly inserted records in database, not whole records every time when I try to retrain sometimes. It could be real time or not. So, I want to know how can I do this.
And I convert records from database into text file for training. I think this is inefficient. So, I hope I had another options like training from String, Collections. Maybe I could be mistaken about that.
TomFinley commentedon Jun 1, 2018
Regarding the first question, no, sorry. The underlying code supports it for a few learners, but this sort of graph composition layer does not. So we're somewhat out of luck at the moment.
Regarding the second question, about working over stuff in memory vs. materializing to disk, do you think #106 helps? Or is that insufficient?
ecofee commentedon Jun 1, 2018
OK. I understand. Thanks. @TomFinley
abgoswam commentedon Oct 18, 2018
DRI RESPONSE: Issue seems to have been answered. I plan to close the issue withing the next few days.
Averaged Perceptron pipeline serialization fix (dotnet#257)
code migration from machinelearning-automl to machinelearning repo fe…
Averaged Perceptron pipeline serialization fix (dotnet#257)
Added onnx export functionality for LpNormNormalizingTransformer (#4161)
Added Onnx export functionality to PCATransformer (#4188)