nexgml 1.5.1

NexGML — Next Generation Machine Learning (educational ML utilities)

Nexarians - Open source machine learning repository for community

🔬 NexGML PyPI package: Transparent, Modular, and Easy to use

NexGML (Next Generation Machine Learning) is a custom Machine Learning utility package built for educational and research purposes, emphasizing code transparency and high performance with efficient python implementation.

Key Features & Technology Stack

• Modular Helpers: Separates complex logic into focused helper modules (ForLinear, ForTree, Indexing, Metrics, Guardians) for easy customization.
• Sparse Data Ready: Full support for scipy.sparse matrices (CSR/CSC) for memory efficiency.

Installation

pip install nexgml

1. Classifiers (The Models)

The primary model is the Gradient Supported Intense Classifier (GSIC).

from nexgml.gradient_supported import IntenseClassifier
import numpy as np

# Load data X, y...

model = IntenseClassifier(
    optimizer='adamw', 
    lr_scheduler='plateau', 
    batch_size=32, 
    penalty='elasticnet'
)
model.fit(X_train, y_train)

print(f"Final Training Loss: {model.loss_history[-1]:.6f}")

2. Regressors (The Models)

The primary model is the Gradient Supported Intense Regressor (GSIR).

from nexgml.gradient_supported import IntenseRegressor
import numpy as np

# Load data X, y...

model = IntenseClassifier(
    optimizer='adamw', 
    lr_scheduler='plateau', 
    batch_size=32, 
    penalty='elasticnet'
)
model.fit(X_train, y_train)

print(f"Final Training Loss: {model.loss_history[-1]:.6f}")

2. Helper Modules (Performance Backbone)

These modules contain the high-speed math used internally.

Module	Purpose	Example Usage
nexgml.amo.forlinear	Linear Criteria. Activation/Loss functions (Softmax, CCE, RMSE).	forlinear.softmax(logits)
nexgml.amo.fortree	Tree Criteria. Impurity measures (Gini, Entropy, Friedman MSE).	fortree.gini_impurity(labels)
nexgml.indexing	Data Utilities. One-hot encoding, smart feature slicing (standard_indexing).	indexing.standard_indexing(n_features, 'sqrt')
nexgml.metrics	Model Metrics. Regressor and classifier models metrics computation (R^2, F1, Accuracy Score)	accuracy_score(y_true, pred)
nexgml.guardians	Numerical stability. Value clipping, invalid value detecting (safe_array, hasinf, hasnan)	safe_array(array)

ConceptDesign

ConceptDesign (CoDe) is a folder where the design of a concept that is material for community research is defined and developed via files with markdown (md) existence.

ProtoConceptDesign

ProtoConceptDesign (ProCoDe) is a folder that stores the main ideas, basic theorems, main ideas and context definitions of a concept design which are in the ConceptDesign folder. The purpose of separating CoDe and ProCoDe is to provide clarity between concept definitions and context definitions.

DesignCache

DesignCache is a folder that contains concept designs that are deemed to have failed, the failure in question could be an error in the main idea of ​​the concept from the start, a refutation of the theory on which a concept is based, and other things that can affect the interpretability of the definition and meaning of a concept.

📝 Documentation & Exploration

This repository is dedicated to experimentation, learning, and personal research, primarily in the following fields:

• 🤖 Artificial Intelligence and Machine Learning
• 💻 Python development and performance optimization
• 📖 Technical documentation and concept notes
• 🥼 Designing and developing concepts, theories and hypotheses. Before finally forming its implementation

⚠️ This project is intended for exploration and learning purposes only.

If you find this repo helpful or interesting, feel free to fork, star, or open a pull request.
This is a learning space—no pressure, just passion! 😄