Medial Tools
This section provides a list of applications, tools, and executables built on the AlgoMedical library framework:
- Guide for Common Actions
- Using the Flow App: A versatile application with multiple switches, each triggering a specific action. Below are some of the key functionalities:
- Load New Repository
- Train a model
- Apply a model to generate predictions.
- Extract feature matrices from the model pipeline.
- Print specific patient data or signal distributions.
- Feature Importance with Shapley Values Analysis
- more...
- Bootstrap App: A tool for bootstrap analysis of prediction and outcome files.
- Bootstrap Legend - The bootsrap output file result legend
- Extending Bootstrap:
- Utility Tools for Processing Bootstrap Results.
- Optimizer: A tool for hyperparameter optimization.
- TestModelExternal: Compares repositories or samples by building a propensity model to identify differences.
- Change Model: Modifies a model without retraining, such as enabling verbose mode for outliers or limiting memory batch size.
- Adjust Model: Adds components like rep_processors or post_processors to a model. Some components may require training with MedSamples and a repository.
- Iterative Feature Selector: Iteratively selects features or groups of features in a top-down or bottom-up manner and generates a report.
- Fairness Extraction: Calculates fairness metrics.
- Model Signals Importance: Evaluates the importance of signals in an existing model and measures the impact of including or excluding them on performance.
- Simulator: Simulates performance by controlling variables such as target population age distribution and key covariates. It also evaluates the impact of signal availability and existence.
- AutoTest - Pipeline scripts to run common tests for model development, validation. This is mainly shell, python scripts that utilize above tools.
Examples of Simple Applications
To learn how to create your own applications, clone the MR_Tools repository. Navigate to the MedProcessUtils directory and explore the following examples:
learn- Application.getMatrix- Application.predict- Application.
Retired Applications
- Action Outcome Effect: A tool for estimating the average treatment or action effect on outcomes. It also supported feature selection, model selection, training, and bootstrap analysis with recovery and step-skipping capabilities.
- Signals Dependencies: Identifies statistical dependencies between registry and signal values within a time window. Useful for selecting categorical signals.
- Create Registry:
- Find Required Signals.
- Compare AUCs.
- GANs for Imputing Matrices: