AUTOML Training Repo

The pipeline integrates gemini API via gemini.R package into the auto-ml pipeline. The integration allows for automated interpretation of key outputs.

Illustrations

• DGH-DSWB Hackathon 2025 (2025 Aug 06 (Wed))
• DSWB AGM 2025 (2025 Feb 22 - 23)
• Data Community meeting at APHRC on 2024 Jun 20 (Thur)

Requirements

R and RStudio

Download and install R.

• Windows
  • For Windows users, download Rtools
• Mac
• Linux

Next step is to download Rstudio and install

• Rstudio

R packages

• Run R code:

source("requirements.R")

Gemini API

• Get Gemini API
• Copy and paste the API key into the R function below

Download Rautoml package

• Either via pkgs or requirements.R

Rautoml::set_api("GEMINE_API_KEY", "YOUR_API_KEY")

Usage

• Use your own data or dowmload sample data from here.
• Open simple_ml_pipeline.R from your preferred editor (Rstudio recommended) and run line by line for default usage. Once you familiarize with the codebase, you can change the data via data_path, etc.
• Edit upload_data.R to specify data
• Edit model_config.R to change default parameters outline above.
• Edit the data_management.R file to add any data transformations needed.
  • data_management - TRUE/FALSE. Set to FALSE for default.

Key components

• You can modify the following variables depending on the task:
  • experiment_name - This gives a tag to the current experiment. The output folder which contains all the files related to the current experiment will be tagged this, i.e., outputs_your_experiment_name.
  • study_description - The description of the study or current dataset. You can also include study objective.
  • data_path - URL or path to the dataset. Supports .csv, .xlsx, .xls, .dta, .rda, .rds, .spss files.
  • problem_type - Whether "classification" or "regression" problem
  • outcome_var - The outcome of interest. The variable should be in the dataset
  • train_test_ratio - Training-test data ratio for partitioning
  • corr_threshold - Correlation threshold, ranges [0, 1], for bivariate comparison of variables. One of the highly correlated variables is dropped based on the threshold
  • handle_missing - How to hand missing values. It can either be "omit", "missing_mean", "missing_median", "mode_median", "mode _mean", "bag", "knn", or "knn_linear"
  • cv_method - Cross-validation method. It can be "CV ", "repeatedcv ", "adaptive_cv", "LOOCV ", "LGOCV ", "adaptive_LGOCV ", "boot", "boot632", "optimism_boot", "boot_all", or "adaptive_boot"
  • n_folds - Number of cross-validation folds. Either 3, 5, or 10.
  • cv_repeats - Number of repeats in cross-validation
  • performance_metric - Measure of model performance
    • For regression: "RMSE", "Rsquared", "MAE"
    • For classification: "AUCROC", "Accuracy", "AUCRecall", "Sens", "Spec", "Precision", "Recall", "F"
  • report_metric - Depending on the task, it can be any of the above

Advance pipeline

• See AUTOML Pipeline