maryjis / autoNeuro

Pipeline to run the experiments with tabular data.

autoNeuro

Pipeline to run the experiments with tabular data for classification tasks (Major Depressive Disorder vs Healthy Control, Schisophrenia vs Healthy Control).

Pipeline consist of several steps

1. GridSearchBase - search for the best ML method with best feature selection algorithm from (XGBClassifier, SVC, RandomForestClassifier(), "lr" : LogisticRegression() ) and feature selection methods (SelectKBest, RandomForestClassifier (select top important features), LogisticRegression (select top important features) ).It is located in core/gridcv.py module. You can configure the parameters for grid search algorithm for each model in GRID_CONFIG_MODELS in core/constants.py module.
2. ExperimentsInfo - calculate the most important features for the best methods, build roc curves and other metrics, save important features as DataFrame to EXPERIMENTS_PATH (experiments.py module). It is located in core/metrics.py module.
3. FeaturesStats - calculate the distributions and post-hoc t-test for the the most important features calculated in the previos step. It is located in core/stats.py

How to run

You can run code using experiments.py or see examples with jupyter notebooks in experiments package.

  X - pd.DataFrame with numeric data
  y - pd.Series with target values
  experiment_name - name of experiment (results are saved in directiry with this name) 

  best_result,best_f1 =run(X,y, experiment_name=file.stem, repeats =10, topN=1,scaling=False)