metadata

license: mit
datasets:
  - dair-ai/emotion
language:
  - en
library_name: transformers
widget:
  - text: I am so happy with the results!
  - text: I am so pissed with the results!
tags:
  - debarta
  - debarta-v3-small
  - emotions-classifier

Fast Emotion-X: Fine-tuned DeBERTa V3 Small Based Emotion Detection

This model is a fine-tuned version of microsoft/deberta-v3-small for emotion detection using the dair-ai/emotion dataset.

Overview

Fast Emotion-X is a state-of-the-art emotion detection model fine-tuned from Microsoft's DeBERTa V3 Small model. It is designed to accurately classify text into one of six emotional categories. Leveraging the robust capabilities of DeBERTa, this model is fine-tuned on a comprehensive emotion dataset, ensuring high accuracy and reliability.

Model Details

Model Name: AnkitAI/deberta-v3-small-base-emotions-classifier
Base Model: microsoft/deberta-v3-small
Dataset: dair-ai/emotion
Fine-tuning: The model is fine-tuned for emotion detection with a classification head for six emotional categories: anger, disgust, fear, joy, sadness, and surprise.

Emotion Labels

Anger
Disgust
Fear
Joy
Sadness
Surprise

Usage

You can use this model directly with the provided Python package or the Hugging Face transformers library.

Installation

Install the package using pip:

pip install emotionclassifier

Basic Usage

Here's an example of how to use the emotionclassifier to classify a single text:

from emotionclassifier import EmotionClassifier

# Initialize the classifier with the default model
classifier = EmotionClassifier()

# Classify a single text
text = "I am very happy today!"
result = classifier.predict(text)
print("Emotion:", result['label'])
print("Confidence:", result['confidence'])

Batch Processing

You can classify multiple texts at once using the predict_batch method:

texts = ["I am very happy today!", "I am so sad."]
results = classifier.predict_batch(texts)
print("Batch processing results:", results)

Visualization

To visualize the emotion distribution of a text:

from emotionclassifier import plot_emotion_distribution

result = classifier.predict("I am very happy today!")
plot_emotion_distribution(result['probabilities'], classifier.labels.values())

Command-Line Interface (CLI) Usage

You can also use the package from the command line:

emotionclassifier --model deberta-v3-small --text "I am very happy today!"

DataFrame Integration

Integrate with pandas DataFrames to classify text columns:

import pandas as pd
from emotionclassifier import DataFrameEmotionClassifier

df = pd.DataFrame({
    'text': ["I am very happy today!", "I am so sad."]
})

classifier = DataFrameEmotionClassifier()
df = classifier.classify_dataframe(df, 'text')
print(df)

Emotion Trends Over Time

Analyze and plot emotion trends over time:

from emotionclassifier import EmotionTrends

texts = ["I am very happy today!", "I am feeling okay.", "I am very sad."]
trends = EmotionTrends()
emotions = trends.analyze_trends(texts)
trends.plot_trends(emotions)

Fine-tuning

Fine-tune a pre-trained model on your own dataset:

from emotionclassifier.fine_tune import fine_tune_model

# Define your training and validation datasets
train_dataset = ...
val_dataset = ...

# Fine-tune the model
fine_tune_model(classifier.model, classifier.tokenizer, train_dataset, val_dataset, output_dir='fine_tuned_model')

Using transformers Library

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model_name = "AnkitAI/deberta-v3-small-base-emotions-classifier"
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Example usage
def predict_emotion(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=128)
    outputs = model(**inputs)
    logits = outputs.logits
    predictions = logits.argmax(dim=1)
    return predictions

text = "I'm so happy with the results!"
emotion = predict_emotion(text)
print("Detected Emotion:", emotion)

Training

The model was trained using the following parameters:

Learning Rate: 2e-5
Batch Size: 4
Weight Decay: 0.01
Evaluation Strategy: Epoch

Training Details

Evaluation Loss: 0.0858
Evaluation Runtime: 110070.6349 seconds
Evaluation Samples/Second: 78.495
Evaluation Steps/Second: 2.453
Training Loss: 0.1049
Evaluation Accuracy: 94.6%
Evaluation Precision: 94.8%
Evaluation Recall: 94.5%
Evaluation F1 Score: 94.7%

Model Card Data

Parameter	Value
Model Name	microsoft/deberta-v3-small
Training Dataset	dair-ai/emotion
Number of Training Epochs	20
Learning Rate	2e-5
Per Device Train Batch Size	4
Evaluation Strategy	Epoch
Best Model Accuracy	94.6%

License

This model is licensed under the MIT License.