HeimdallX 0.3

A Machine Learning and Artificial Intelligence application package

HEIMDALL v0.3 - Updated 17/04/24

High-Resolution [----] Identification and [----] Discriminator [----]

1. ML_meta Class:

Purpose: Acts as a coordinator, managing the application of various machine learning algorithms. Key Methods: apply_all_models(flag=True): Applies multiple ML models and compares their scores. apply_neural_net(): Applies a feedforward neural network (FFNN). apply_single_model(cm=False, save_model=False, save_model_name=False): Applies a specified model, with options for confusion matrix plotting and model saving. split_data(encode_categorical=True, y='target'): Splits data into features (X) and target (y), with optional encoding of categorical features. call_ML(): Instantiates the ML class to access model implementation details.

2. ML Class:

Houses the implementation of various machine learning algorithms, including:

• Support Vector Machine (SVM)
• Naive Bayes (NB)
• Decision Tree (DT)
• Random Forest (RF)
• k-Nearest-Neighbour (kNN)
• Logistic Regression (LR)
• Multi-Layered Perceptron (MLP)
• Ensemble Classifier (EC)
• Gradient Boosted Classifier (GBC)
• Ada Boosted Classifier (ABC)

PLANNED: MultiLayeredPerceptron (MLP) bug fixes

3. FFNN Class:

Implements a feedforward neural network.

4. BasePredictor Class:

Provides a base class for prediction-related functionality.

5. CNN Class:

Implements a convoluted neural network architecture.

6. YOLOv8 Object identifier

Provides the functionality for a user-trained YOLO identifier to predict on either pre-recorded or live video stream

7. Pipeline Class:

Reads in YAML configuration file to streamline the implementation and usage of the package

---

Usage

Use case 1: Applying Multiple Models and Comparing Performance:

import ML_meta

model_coordinator = ML_meta.ML_meta() # Instantiate the coordinator model_coordinator.apply_all_models(flag=True) # Apply all models and generate a report

Use case 2: Applying a Specific Model with Confusion Matrix and Model Saving:

model_coordinator.apply_single_model("RF", cm=True, save_model=True, save_model_name="best_rf.pkl")

Use case 3: Applying a Feedforward Neural Network:

model_coordinator.apply_neural_net()

Use case 4: Customising Model Application:

Split data with custom settings X_train, X_test, y_train, y_test = model_coordinator.split_data(encode_categorical=False, y="my_target")

Access specific models and control parameters ml_instance = model_coordinator.call_ML() ml_instance.apply_decision_tree(X_train, X_test, y_train, y_test, max_depth=5)

The ML_meta class serves as the primary entry point for users. Methods like apply_all_models, apply_single_model, and apply_neural_net provide streamlined model application. Access to individual models and their parameters is available through call_ML. Customise data splitting and encoding using split_data.

---

TO-DO

• Refine YAML input capability
• Fully-refine the Pipeline class to be fully functional
• Explore model explainability techniques (e.g., SHAP, LIME) to understand model behavior better
• Employ techniques for handling imbalanced datasets if applicable
• Consider model deployment strategies for real-world applications
• Add a Setup.cfg file
• Add a Setup.py file
• Add UnitTests for the package

Copyright © 2024 . All rights reserved.