lightwood
lightwood copied to clipboard
mindsdb
Lightwood is Legos for Machine Learning.
~Once #833 is merged,~ We should look into simplifying the family of array-TS encoders. In particular, `TsCatArrayEncoder` and `TsNumArrayEncoder` should be refactored to inherit from `ArrayEncoder` (or the other two...
The timeseries transform step (`data.timeseries_transform`) takes a long time and involves mainly data-frame manipulation, we should figure out if it can be made faster with dask.
We should figure out if the usage of optuna to find hyperparams for the lightgbm boosters can be parallelized with dask. For more info see: https://jrbourbeau.github.io/dask-optuna/
Our internal suite reports a few datasets (diamonds, stack_overflow_survey) that might indicate there is an issue scaling for larger tasks. We should check into this.
### Pending tasks - [x] Integration test - [x] Override learnt weights with user-defined weights - [ ] Benchmark against `BestOf`
This enables e.g. passing holiday indicators belonging to any given forecasted horizon, and the model _should_ leverage this information for improved forecasts.
## Why To offer an alternative method for regression bounds. ## How Introducing a `PLinearWrapper` class for the (already existing) implementation of `PLinear` layers by @torrmal. This enables its usage...
Get `AccStats` to generate every single viable accuracy metrics it can think of for the problem types. For starters let's do everything under sklearn's `metrics` + whatever @paxcema thinks is...
The methods are a bit confusing, especially the `get_{x}` methods, try to take a look at this and see if we can simplify it.
We need to add a static analysis tool that triggers on each PR and provides a report, ideally `flake8` style where we can configure its behaviour *and* have the action...