lightwood
lightwood copied to clipboard
mindsdb
Lightwood is Legos for Machine Learning.
This enables easy type inference overriding.
Something like this: ``` pdef = ProblemDefinition.from_dict({'target': target, 'time_aim': 80}) jai = json_ai_from_problem(df, pdef) jai.model['module'] = 'MeanEnsemble' predictor = predictor_from_json_ai(jai) predictor.learn(df) ``` Will currently fail because the autogenerated call to...
As reported by a community user: If a LightGBMArray mixer is trained to forecast, querying conditionally and changing the timestamp may not make a difference in the prediction. We should...
We should add some links to the docs, namely to this github repo *and* to the mindsdb website. We should also plug it into our google analytics.
Potentially very useful for classification tasks. https://github.com/cleanlab/cleanlab
Respectively, `import` should load the predictor+code.
When searching for a lr we should probably avoid anything past 0.1 or so, benchmark and implement if accuracy doesn't decrease (+ time doesn't increase noticeably) or if accuracy increase...
Milestones: - [ ] Flag to design PyTorch-only models - [ ] Export/save said models to ONNX format (#477) - [ ] Implement flow to load a predictor in ONNX...
After #473 is done Try converting the already trained model to the onyx IR and see if that improves performance (if it does think about how this can be integrated...
Lightwood currently fails to install in Windows with Python3.9 due to a lack of precompiled wheels for the sktime package for the specific Win-Py3.9 combination.