---
library_name: transformers
license: apache-2.0
base_model: distilbert-base-uncased
tags:
- generated_from_trainer
metrics:
- accuracy
- precision
- recall
- f1
model-index:
- name: creative-energy-analyzer
  results: []
---

# 🎨 Creative Energy Sentiment Classifier

This model is fine-tuned to detect **creative emotional states** in text. It predicts one of **six nuanced sentiment labels** that represent different dimensions of creative energy: from inspiration and flow to burnout and doubt.

---

## 🧠 Labels

The model classifies text into one of the following six labels:

| Label      | Description |
|------------|-------------|
| **inspired** | Bursting with ideas, energized to create |
| **expressive** | In the flow of articulating or experimenting freely |
| **curious** | Exploring new ideas, researching, or discovering |
| **stuck** | Blocked from creating despite the desire |
| **doubtful** | Feeling unsure of one’s ideas, skill, or worth |
| **drained** | Creatively exhausted, lacking energy or motivation |

---

## 🧭 States of Creative Energy

Each label is mapped to a **creative state** — a broader dimension of how energy shows up in the creative process:

| State        | Description | Labels |
|--------------|-------------|--------|
| **momentum** | 🔥 Spark vs. paralysis | inspired ↔ stuck |
| **voice**    | 🎤 Flow vs. self-criticism | expressive ↔ doubtful |
| **exploration** | 🌱 Wonder vs. exhaustion | curious ↔ drained |

These contrasts help reveal how creativity shifts between energized and blocked states.

---

## 📊 Training

- **Base Model**: `distilbert-base-uncased`
- **Fine-tuned on**: 1,200 labeled examples (200 per class)
- **Format**: JSONL with `"text"`, `"label"`, and `"state"`
- **Split**: 80/10/10 (train/val/test)

---

## 📦 Dataset

This model was trained on the custom King-8/creative-energy-sentiment 
dataset (https://huggingface.co/datasets/King-8/creative-energy-sentiment), containing 1,200 examples crafted and categorized into 6 emotion-based labels.
 
---

## 💡 Inspiration

This project aims to go beyond typical positive/negative sentiment and capture the emotional complexity of the creative process — to better support artists, writers, students, and thinkers navigating their creative energy.

---

## 🧪 Example Usage

```python
from transformers import pipeline

classifier = pipeline("text-classification", model="King-8/creative-energy-sentiment")

classifier("I’ve been experimenting with new textures all morning — it's so fun!")
# [{'label': 'expressive', 'score': 0.91}]
```
  
---

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall | F1     |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:---------:|:------:|:------:|
| 1.2884        | 1.0   | 120  | 1.2956          | 0.5      | 0.4050    | 0.5    | 0.4412 |
| 0.8774        | 2.0   | 240  | 0.9885          | 0.6      | 0.6306    | 0.6    | 0.5836 |
| 0.515         | 3.0   | 360  | 0.8751          | 0.6417   | 0.6590    | 0.6417 | 0.6466 |
| 0.3956        | 4.0   | 480  | 0.8428          | 0.6583   | 0.6679    | 0.6583 | 0.6564 |
| 0.2319        | 5.0   | 600  | 0.8588          | 0.65     | 0.6633    | 0.65   | 0.6443 |

---

### Evaluation (Validation Set)

- Loss: 0.6559
- Accuracy: 0.7667
- Precision: 0.7630
- Recall: 0.7667
- F1: 0.7618

---

### Framework versions

- Transformers 4.56.1
- Pytorch 2.8.0+cu126
- Datasets 4.1.1
- Tokenizers 0.22.0