ml-juelich issueshttps://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues2020-10-05T15:53:10+02:00https://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/13Confound without removal2020-10-05T15:53:10+02:00s.hamdanConfound without removalCurrently, confounds are always ignored for transformers,
but they are not removed if there is no confound removal.
Therefore, they will be no used as feature in the Model.
This is inconsistent. Rather it should:
* Be ignored for al...Currently, confounds are always ignored for transformers,
but they are not removed if there is no confound removal.
Therefore, they will be no used as feature in the Model.
This is inconsistent. Rather it should:
* Be ignored for all steps including the model
* Added as a normal featurehttps://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/11More Metrics2020-10-02T14:52:48+02:00s.hamdanMore MetricsCurrently, we use the default metrics for scoring.
sklearn's `cross_validate` can support multiple metrics.
We will have to add multiple metrics as a selection, but maybe also multiple ones as default.
I would show by default for:
...Currently, we use the default metrics for scoring.
sklearn's `cross_validate` can support multiple metrics.
We will have to add multiple metrics as a selection, but maybe also multiple ones as default.
I would show by default for:
A) Regression:
- r2
- adj. r2
- mae
- mse
B) Classification:
- accuracy
- balanced accuracy
- auc_roc
- f1https://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/10Save Model2020-10-02T11:37:47+02:00s.hamdanSave ModelCreate a joblib file to save the model and use it later again.
Therefore, we also need a function for using such a model on new data.
It should have a retrain tag if people do not only want to predict, but also retrain starting from th...Create a joblib file to save the model and use it later again.
Therefore, we also need a function for using such a model on new data.
It should have a retrain tag if people do not only want to predict, but also retrain starting from the weights of the saved modelhttps://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/8Tests2020-10-01T19:31:50+02:00s.hamdanTestshttps://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/7More Flexible Data Input2020-10-02T11:15:32+02:00s.hamdanMore Flexible Data InputArgument df has to be replaced with a data argument in user facing api.
This argument should be able to work with multiple data structures.
CSV, Dictionary, Matlab Structure or Data LadArgument df has to be replaced with a data argument in user facing api.
This argument should be able to work with multiple data structures.
CSV, Dictionary, Matlab Structure or Data Ladhttps://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/6Better Input Validation2020-10-01T19:51:03+02:00s.hamdanBetter Input ValidationCurrently, the `validate_input` function is not very useful.
We need to give users good feedback when input is wrong.
Not all the inputs are final until now, so we might first need to discuss what parts will definitely stay.Currently, the `validate_input` function is not very useful.
We need to give users good feedback when input is wrong.
Not all the inputs are final until now, so we might first need to discuss what parts will definitely stay.https://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/5Cross Validation2020-10-01T18:56:42+02:00s.hamdanCross ValidationCurrently, we are always doing grid search and always KFold CV, but with different numbers of folds and repeats.
LOO has to be added. But what else and what do you think would be a good way to specify in the user facing api.
Should we...Currently, we are always doing grid search and always KFold CV, but with different numbers of folds and repeats.
LOO has to be added. But what else and what do you think would be a good way to specify in the user facing api.
Should we always do Gridsearch even when there are few parameters ( so results are consistent) or only when it is needed?https://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/2Documentation2020-10-01T18:41:59+02:00s.hamdanDocumentationProper docstrings are missing.
This will be added as soon as some of the discussion points are solved.Proper docstrings are missing.
This will be added as soon as some of the discussion points are solved.https://jugit.fz-juelich.de/s.hamdan/ml-juelich/-/issues/1Code Style + CI Pipeline2020-10-01T18:40:34+02:00s.hamdanCode Style + CI PipelineCurrently, the code is @s.hamdan style.
It will be updated to default flake8.
Furthermore, a CI pipline checking whether pushes are compatible will be created.Currently, the code is @s.hamdan style.
It will be updated to default flake8.
Furthermore, a CI pipline checking whether pushes are compatible will be created.