---
license: apache-2.0
language:
- en
metrics:
- accuracy
pipeline_tag: tabular-regression
tags:
- code
---

# RRF-Savant Meta-State Logistic Regression

## Model Summary

This repository contains a lightweight **logistic regression** classifier implemented with **scikit-learn**.  
The model operates on **15-dimensional RRF-Savant meta-state features**, derived from the RRF / SavantEngine pipeline, and outputs a binary prediction with associated probabilities.

It is designed as a fast, interpretable decision layer on top of the richer RRF-Savant embedding and resonance machinery.

---

## Model Details

- **Model type:** Logistic Regression (binary classifier)  
- **Framework:** scikit-learn  
- **Input dimensionality:** 15  
- **Source notebook:** `RRFSavant_AGI_Core_Colab.ipynb`  
- **File format (recommended):** `joblib` (`.joblib`)

### Input Features

Each input is a 15-dimensional feature vector:

- **RRF-Savant meta-state features**, including:
  - φ / Φ phase indicators (RRF-Savant “phi” level)
  - Ω / omega dynamics or cycle index
  - Global coherence / resonance scores
  - Spectral features:  
    - `S_RRF`: spectral smoothness  
    - `C_RRF`: spectral concentration  
  - Energy-like measures (e.g. `E_H`)
  - Dominant frequency or harmonic index
  - One-hot encoded Φ nodes / states

> Exact semantics and preprocessing are defined in the source notebook  
> `RRFSavant_AGI_Core_Colab.ipynb`.

### Outputs

- `y_pred`: binary class label (e.g. `0` vs `1`)  
- `proba`: probability estimates for each class via `predict_proba`

The precise interpretation of class `0` and class `1` (e.g. baseline vs. “RRF-aligned” state, safe vs. risky, etc.) should be documented alongside your use-case.

---

## Intended Use

- **Primary use:**  
  - As a **meta-controller** for RRF-Savant systems, mapping high-level meta-state features to a simple decision (binary label).
  - As a **fast screening / routing head** deciding whether to:
    - escalate to a heavier RRF/Savant pipeline,
    - trigger a specific operating mode,
    - log / flag certain states.

- **Not intended for:**  
  - Standalone critical decision-making (medical, legal, safety-critical applications) without human oversight.
  - Direct real-world risk scoring without proper calibration and validation.

---

## Training

- **Training framework:** scikit-learn `LogisticRegression`
- **Data source:**  
  - Internal RRF-Savant meta-state dataset, generated and curated in  
    `RRFSavant_AGI_Core_Colab.ipynb`.
- **Preprocessing (typical):**
  - Numeric features scaled (e.g. `StandardScaler`)  
  - Categorical / discrete Φ nodes one-hot encoded  
  - Train/validation split performed inside the notebook

> For exact data splits, preprocessing, and hyperparameters, refer to the Colab notebook.

---

## Evaluation

Typical metrics for this model family include:

- Accuracy
- ROC-AUC
- Precision / Recall / F1
- Calibration of probabilities

You should log and report:

- Metrics on a **held-out test set**
- Any **class imbalance** handling performed (e.g. `class_weight="balanced"`)

---

## How to Use Assuming model is loaded 
import joblib
import numpy as np

# Load model
clf = joblib.load("rrf_savant_meta_logit.joblib")

# Example: single feature vector (15 dims)
x = np.array([
    0.85087634, 0.67296168, 0.74652746,
    0.03735409, 0.72399869, 0.66076596,
    0.30312352, 0.69585885, 0.98531076,
    0.28866375, 0.99602791, 0.69072907,
    0.05884264, 0.74298728, 0.75928443
]).reshape(1, -1)

# Prediction
y_pred = clf.predict(x)[0]
proba = clf.predict_proba(x)[0]  # [P(class 0), P(class 1)]

print("Predicted label:", y_pred)
print("Probabilities:", proba)


### Install Dependencies

```bash
pip install scikit-learn numpy joblib