Skip to content

Contributing to RAPID

Thank you for your interest in contributing to the RAPID Pipeline!

To maintain consistency, scientific integrity, and performance across biostatistical analyses and machine learning models, we follow a strict architectural pattern. All contributors must adhere to the standards outlined below.

Development Environment Setup

1. Clone and Install

git clone git clone https://github.com/noispuc/isaric3.0-rapid.git
cd rapid-pipeline
python -m venv .venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
pip install -e ".[dev]"

2. Project Structure

src/isaric/ ├── base/ # Abstract base classes ├── pipelines/ # Model implementations ├── preprocessing/ # Data cleaning and imputation ├── modeling/ # Statistical models ├── evaluation/ # Metrics and diagnostics ├── validation/ # Cross-validation and bootstrapping └── visualization/ # Plotting functions

The RAPID_Pipeline Base Class

To ensure seamless integration, all model classes must inherit from RAPID_Pipeline. This guarantees a common interface across all analytical tools.

from isaric.base import RAPID_Pipeline

class MyNewModel(RAPID_Pipeline):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)

Abstract Methods (Required)

All new pipelines must implement the following core methods:

Method Purpose Return
fit() Train the model with preprocessing self
summary() Generate diagnostics and results None (prints/displays)

Example Template

class MyNewModel(RAPID_Pipeline):
    def __init__(self, data, param1=default1, param2=default2):
        super().__init__(data=data)
        self.param1 = param1
        self.param2 = param2

    def fit(self, **kwargs):
        # 1. Validate input data
        # 2. Preprocess
        # 3. Train model
        # 4. Store results in self.summary_df, self.metrics_df, etc.
        return self

    def summary(self, plots=None, **kwargs):
        # 1. Print results table
        # 2. Generate requested plots
        pass

Factory Integration

All models must be registered with the RAPID_PipelineFactory to enable the unified create() interface:

# In your model file
from isaric.pipelines.factory import RAPID_PipelineFactory

@RAPID_PipelineFactory.register("my_model")
class MyNewModel(RAPID_Pipeline):
    ...

This allows users to instantiate your model with: 

model = factory.create("my_model", data=df, param1=value)

Naming Conventions

Element Convention Example
Variables/Functions snake_case dependent_var, calculate_auc()
Classes PascalCase LogisticRegression, SurvivalCox
Constants UPPER_SNAKE_CASE DEFAULT_THRESHOLD
Private methods _leading_underscore _validate_data()

Return Types and Attributes

All models should expose results via standardized attributes:

Attribute Type Content
summary_df pd.DataFrame Coefficients, CI, p-values
performance_metrics_df pd.DataFrame Model fit metrics
assumption_metrics_df pd.DataFrame Diagnostic tests
vif_df pd.DataFrame Multicollinearity (if applicable)

Documentation

Update the following files when adding a new model:

File What to Add
docs/user_guide/guide_*.md Full tutorial with theory
docs/quick_guide/quick_*.md Quick reference table
docs/index.md Link in "Choose Your Path"
mkdocs.yml Nav entry under User Guide and Quick Guide

Pull Request Checklist

Before submitting a PR, ensure:

  • [ ] Class inherits from RAPID_Pipeline
  • [ ] Registered with RAPID_PipelineFactory
  • [ ] fit() and summary() methods implemented
  • [ ] Standard attributes populated (summary_df, performance_metrics_df)
  • [ ] Type hints included for all arguments and return values
  • [ ] Docstrings follow PEP 257
  • [ ] Code follows PEP 8
  • [ ] Tests added in tests/
  • [ ] Documentation updated (User Guide + Quick Guide)

Questions?