Metadata-Version: 2.4
Name: tonneanalyzer-ankit
Version: 0.2.2
Summary: Hybrid rule + ML based tone analysis library
Author: Ankit Ghosh
License-Expression: MIT
Requires-Python: >=3.8
Description-Content-Type: text/markdown
Requires-Dist: numpy
Requires-Dist: scikit-learn
Requires-Dist: joblib

# tonneanalyzer-ankit

A **hybrid rule + machine-learning tone analysis library**.

This library analyzes **spoken or written text** across multiple **independent communicative tone axes** using:

- deterministic linguistic rules (precision)
- lightweight ML classifiers (robustness)
- explicit rule–model fusion
- negation-aware suppression

The system is designed to be:
- modular
- interpretable
- reproducible
- production-safe

This is **not** a large language model.

---

## Table of Contents

- Installation
- Python Version
- Package Name vs Import Name
- Project Structure
- Core Concept
- Supported Axes
- Required Artifacts
- Quick Start
- Full Usage
- Output Format
- Negation Handling
- Rule–Model Fusion
- Design Constraints
- What This Library Is / Is Not
- Versioning
- License

---

## Installation

```bash
pip install tonneanalyzer-ankit
This installs the library code only.

⚠️ Trained model artifacts are not downloaded automatically.
You must provide them explicitly (see below).

Python Version
Python 3.8 or higher

Tested on Python 3.9

Package Name vs Import Name (IMPORTANT)
Purpose	Name
pip install name	tonneanalyzer-ankit
Python package	tone
Main class	tone.analyzer.ToneAnalyzer

This separation is intentional and standard in Python packaging.

Project Structure (after installation)
The installed package provides:

arduino
Copy code
tone/
├── __init__.py
├── analyzer.py        # public API
├── fusion/            # rule–model fusion logic
├── rules/             # deterministic linguistic rules
└── models/            # ML classifier definitions
You do not import from fusion, rules, or models directly.

Core Concept
Each tone axis is handled independently.

For a given axis:

A rule function produces a binary signal (0 or 1)

A trained ML model produces a probability

Both are combined using a weighted fusion rule

Negation is applied when linguistically valid

There is no shared label space and no shared classifier.

Supported Axes (v1)
Axis	Meaning
Persuasive	Attempts to influence or convince
Questioning	Interrogative intent
Formal	Formal vs informal language
Assertive	Forceful vs submissive tone
Authoritative	Commanding vs collaborative
Narrative	Story-like vs explanatory

Each axis returns a boolean decision.

Required Artifacts (MANDATORY)
You must provide trained model artifacts in Joblib format.

Required directory layout
Copy code
artifacts/
├── persuasive_model.joblib
├── questioning_model.joblib
├── formal_model.joblib
├── assertive_model.joblib
├── authoritative_model.joblib
└── narrative_model.joblib
These files are loaded at runtime.

The library does not train models automatically.

Quick Start
python
Copy code
from tone.analyzer import ToneAnalyzer

analyzer = ToneAnalyzer("path/to/artifacts")

result = analyzer.analyze(
    "I strongly recommend using this approach."
)

print(result)
Full Usage (Step by Step)
1. Import
python
Copy code
from tone.analyzer import ToneAnalyzer
2. Initialize Analyzer
python
Copy code
analyzer = ToneAnalyzer("path/to/artifacts")
The path must exist

All required .joblib files must be present

Missing files raise FileNotFoundError

3. Analyze Text
python
Copy code
output = analyzer.analyze(
    "I do not recommend this approach."
)
Output Format
The output is a Python dictionary:

python
Copy code
{
  'persuasive': False,
  'questioning': False,
  'formal': False,
  'assertive': False,
  'authoritative': False,
  'narrative': False
}
Each value is a boolean

True means the axis is detected

False means it is not detected

Negation Handling
Negation is applied selectively, only where linguistically valid.

Example:

graphql
Copy code
"I recommend this."
→ persuasive = True

"I do not recommend this."
→ persuasive = False
Negation is not applied to all axes blindly.

Sarcasm is not handled in v1.

Rule–Model Fusion
For each axis:

ini
Copy code
score = α * rule_signal + (1 − α) * model_probability
decision = score ≥ threshold
Where:

α is axis-specific

rules have higher priority than models

models smooth noisy or ambiguous cases

Fusion is deterministic.

Design Constraints (Intentional)
No deep learning models

No transformer dependencies

No shared embeddings across axes

No global confidence score

No auto-downloaded data

No hidden state

This is by design.

What This Library IS
A tone / intent analyzer

Deterministic and inspectable

Suitable for production pipelines

Lightweight and fast

Modular and extensible

What This Library Is NOT
❌ A large language model

❌ A sentiment analyzer

❌ A dialogue agent

❌ A text generator

❌ A black-box classifier

Versioning
0.1.x → documentation and stability

0.2.x → new axes or features

1.0.0 → stable public API

PyPI versions are immutable.

License
MIT License.

Author
Ankit Ghosh

yaml
Copy code

---

### What to do next (exact steps)

1. Save this as `README.md`
2. Bump version in `pyproject.toml` (e.g. `0.1.1`)
3. Run:
   ```bash
   python -m build
   python -m twine upload dist/*
