douroucoulis douroucoulis is a fun and practical library designed to help with model selection and building predictive models using the latest machine learning algorithms. It follows an Information-Theoretic framework, mainly focused on AIC-based model selection, and includes functionality for a variety of data science tasks like data exploration, model evaluation, and hyperparameter optimization.

This library is specifically helpful when you need to perform multi-model inference and use an ensemble of best-ranked models for more accurate estimates. It provides a complete pipeline from data cleaning to model selection and evaluation.

Features General Workflow Data Cleaning – Impute missing values, explore relationships between variables, and clean the dataset. Data Exploration – Visualize data and identify important relationships using heatmaps. Model Selection – Use AIC-based model selection to compare multiple models and find the best-fitting ones. Cross-validation – Assess model performance using cross-validation and tune hyperparameters. Model Averaging – Calculate model-averaged estimates for better predictive performance. Key Functions douroucoulis.instructions() Produces step-by-step instructions to guide you through the entire modeling process.

douroucoulis.test_dataset(n_samples, n_features, n_informative, random_state, regression) Generates a test dataset for exercises. Set regression=True for regression tasks, otherwise, it creates a classification dataset.

douroucoulis.check_data(data) Checks the dataset for missing values and provides feedback.

douroucoulis.impute_data(strategy) Imputes missing data using SimpleImputer. Choose a strategy like 'mean', 'median', or 'most_frequent' for categorical data.

douroucoulis.explore(data, cmap) Visualizes the relationships between explanatory variables (features) and the outcome variable (target) using a heatmap. The cmap argument allows you to specify color maps like 'rainbow', 'seismic', etc.

douroucoulis.aictable(model_set, model_names) Ranks a list of models based on their AIC values. Takes a list of models (model_set) and corresponding names (model_names).

douroucoulis.best_fit() Returns the name and statistics for the single best-fit model (AIC weight > 0.90). For multi-model inference, use douroucoulis.best_ranked().

douroucoulis.best_ranked() Returns the best-ranked models with cumulative AIC weight > 0.95. You can then use douroucoulis.mod_avg() for model-averaged parameter estimates.

douroucoulis.mod_avg() Computes model-averaged estimates for each parameter in the best-ranked models.

douroucoulis.cross_val(X, y, classification) Evaluates a model’s accuracy using cross-validation. Set classification=True for classification tasks.

douroucoulis.hyper(model) Tunes the hyperparameters of the provided model using GridSearchCV for the most accurate fit.

douroucoulis.best_predictions(new_data) Uses the best-fit and most hyperparameterized model to make predictions on a new dataset (new_data).

Fun Sound Functions for Debugging douroucoulis.tonalhoot(reps) Emits a tonal hoot, repeated for the specified number of times (reps). Useful for tracking model fitting progress and debugging.

douroucoulis.gruffhoot(reps) Emits a gruff hoot, repeated for the specified number of times (reps). Use for debugging and tracking model fitting.

douroucoulis.rwhoop(reps) Emits a resonant whoop, repeated for the specified number of times (reps). Also useful for debugging and progress tracking.

Installation You can install the library using pip: pip install douroucoulis

Example Usage Here's an example of how to use the douroucoulis library to perform model selection:

import douroucoulis as do

Generate test dataset

data = dc.test_dataset(n_samples=100, n_features=5, n_informative=3, random_state=42, regression=True)

Check for missing data

do.check_data(data)

Impute missing data

do.impute_data(strategy='mean')

Explore data relationships

do.explore(data, cmap='seismic')

Create a model set and rank them using AIC

model_set = [LinearRegression(), RidgeCV()] model_names = ['Linear Regression', 'RidgeCV'] aic_table = do.aictable(model_set, model_names)

Get the best-ranked models

best_models = do.best_ranked()

Model-averaged parameter estimates

averaged_params = do.mod_avg()

Perform cross-validation on the best model

X = data.drop(columns='target') y = data['target'] do.cross_val(X, y, classification=False)

License This project is licensed under the MIT License – see the LICENSE file for details.