-
## Motivation
Plan for the doc revamp
- [x] A 0-to-1 tutorial or Getting started #861
- [x] A tutorial on building a custom env #911
- [ ] A tutorial on model ensembling #876
- [x] A tutorial…
-
In the upcoming API refactor (PR #509), we should distinguish between attention keys and values in the API, so they can be cached during ensembling independently on the particular attention model impl…
-
@seabbs put together a to do list - adding a copy here for reference/easier access.
Direct copy of text but I have reorganised into chunks.
#### Maintain week-to-week code
- [x] Review current…
-
### ML-Crate Repository (Proposing new issue)
:red_circle: **Project Title**: Hotel Booking Demand Prediction
:red_circle: **Aim**: Predict hotel booking cancellations and analyze factors affect…
-
Can we use algorithms to automatically optimize the merging of the weights and layers of the model along the most efficient path?
-
**Is your feature request related to a problem? Please describe.**
We'd like to use [Pipeline Dependencies](https://sky-uk.github.io/kfp-operator/docs/examples/pipeline_dependencies/) to import artef…
-
Extend the multiarchitecture / multirepeat ensembling with the possibility to train multiple phases of model training on one go.
e.g.
* You want to both pretrain for a large dataset, and then finetun…
-
**A note to investigate further:** We noticed extremely long refit time for AutoSklearn models. For example, if we set the `time_left_for_this_task` to be just 1hr, and so the `per_run_time_limit` is …
-
Add unit tests where we restrict the search space as follows:
- Skip Connections are disabled.
- Snapshot Ensembling is disabled.
- Weight Decay is disabled.
- Dropout is disabled.
Furthermor…
-
### Motivation for the feature
To make model in production more performant, we might want to do ensembling. Eg: Bagging, where multiple models are trained, each with bootstrapped train/valida…