utils-functions

A collection of reusable Machine Learning and Deep Learning utility functions for faster experimentation and model development.
The repository is divided into three main modules:

• utils_ml.py → Classical ML utilities (data preprocessing, encoding, scaling, evaluation, visualization).
• utils_dl.py → TensorFlow/Keras Deep Learning utilities (training visualization, inference time, dataset loaders).
• utils_dl_pytorch.py → PyTorch Deep Learning utilities (training loops, validation, testing, visualization).

---

📦 Installation

Clone the repository:

git clone https://github.com/AhmAshraf1/utils-functions.git

---

🔧 Example Usage

from utils_ml import encode_data, evaluate_classification_models
from utils_dl import measure_inference_time, plot_loss_accuracy
from utils_dl_pytorch import train_model, test_model

---

📚 Modules and Functions

🔹 utils_ml.py (Machine Learning Utilities)

Data Cleaning & Preprocessing

• analyze_IQR_outliers(data, num_columns) – Detect outliers using IQR method
• visualize_outliers(outlier_data, plot_type) – Plot outlier counts/percentages
• replace_outliers(data, column, value_to_replace) – Replace outliers with a given value
• visualize_nulls(data, plot_type) – Show missing values
• encode_data(X_train, X_test, encoder_type, columns) – Encode categorical features
• encode_target(y_train, y_test, encoder_type) – Encode target labels
• scale_data(X_train, X_test, scaler_type, columns) – Scale features

Model Evaluation

• evaluate_regression_models(X_train, y_train, X_test, y_test, models) -> Evaluate regression models → returns metrics + trained models.
• evaluate_classification_models(X_train, y_train, X_test, y_test, models) -> Evaluate classification models.
• evaluate_classification_metrics(y_true, y_pred, target_names, display) -> Generate confusion matrix + classification report.
• evaluate_clustering_models(X_train, X_test, models) -> Evaluate clustering models with silhouette score.
• evaluate_models(X_train, y_train, X_test, y_test, models, task_type) -> General wrapper for regression/classification.
• get_voting(models_df, n_top, voting_type) -> Build voting ensemble (classifier/regressor).

Metrics & Visualization

• accuracy_and_rmse(y_test, prediction)
• precision_recall_f1(y_test, prediction)
• plot_roc_auc_curve(y_test, y_prob)
• plot_precision_recall_curve(y_true, y_pred_proba)

Model Tuning

• grid_search_classification_models(X, y)
• random_search_classification_models(X, y, n_iter)

---

🔹 utils_dl.py (Deep Learning Utilities – TensorFlow/Keras)

Training & Inference

• measure_inference_time(model, input_data) → Measure model inference time.
• plot_loss_accuracy(history) → Plot loss/accuracy curves.
• learning_curves_plot(tr_data, start_epoch)
• learning_curves_tuning(tr_data, start_epoch, history_fine, fine_tune_epoch)
• plot_history(history)

Dataset Handling

• train_val_test_data(batch_size, img_size, train_directory, validation_directory, test_directory)
• visualize_data(data)
• load_prep(img_path, img_title)

Predictions

• random_image_predict(model, test_dir, data, rand_class, cls_name)
• predict_img(img_path, model, data)

Helpers

• parse_image(filename)
• show(image, label)

---

🔹 utils_dl_pytorch.py (Deep Learning Utilities – PyTorch)

Reproducibility

• set_seed(seed) → Fix random seeds across libraries.

Training & Validation

• train_epoch(model, train_loader, criterion, optimizer, device) → Train for one epoch.
• validate_model(model, val_loader, criterion, device) → Validate for one epoch.
• train_model(model, train_loader, val_loader, criterion, optimizer, scheduler, num_epochs, device, best_val_loss, best_val_acc, best_model_state) -> Full training loop with history + checkpointing.

Testing & Evaluation

• test_model(model, test_loader, device) → Classification report & confusion matrix.
• visualize_results(model, test_loader, classes, num_images) → Visualize test predictions.

Visualization

• learning_curves_tuning(history, fine_tune_epoch) → Plot training curves.

Utilities

• EarlyStopping(patience, min_delta) → Custom early stopping (class with .step(val_loss)).
• CustomDataset(paths, transform, is_train) → Dataset class with augmentation.
• SimpleCNN(num_classes) → Example CNN model.