Model Card for emotion-distilbert-7class
Model Details
Model Description
This is a fine-tuned Hugging Face transformers model for emotion classification. The model is based on DistilBERT and trained on a synthetic dataset containing 7 emotions.
- Shared by: fakhrulfaiz201
- Model type: DistilBertForSequenceClassification
- Language(s) (NLP): English
- License: Apache 2.0 (inherited from DistilBERT base model)
- Finetuned from model: distilbert-base-uncased
Model Sources
Uses
Direct Use
This model can be used for classifying text into one of seven emotions: Love, Sad, Anger, Fun, Hate, Surprise, Happiness.
Out-of-Scope Use
The model was trained on a synthetic dataset. Its performance on real-world, nuanced, or informal text might vary. It should not be used for critical applications without further validation.
Bias, Risks, and Limitations
The model's performance is limited by the synthetic nature of its training data. It might not generalize well to diverse linguistic styles, cultural contexts, or sarcasm/irony.
Recommendations
Users should test the model thoroughly on their specific use cases and consider fine-tuning on more diverse and real-world datasets if necessary.
How to Get Started with the Model
Use the code below to get started with the model.
1from transformers import AutoModelForSequenceClassification, AutoTokenizer
2import torch
3
4model_name = "fakhrulfaiz201/emotion-distilbert-7class"
5tokenizer = AutoTokenizer.from_pretrained(model_name)
6model = AutoModelForSequenceClassification.from_pretrained(model_name)
7
8id2label = {
9 0: 'Love', 1: 'Sad', 2: 'Anger', 3: 'Fun',
10 4: 'Hate', 5: 'Surprise', 6: 'Happiness'
11}
12
13def predict(text):
14 device = next(model.parameters()).device
15 inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True).to(device)
16 with torch.no_grad():
17 outputs = model(**inputs)
18 logits = outputs.logits
19 predicted_class_id = logits.argmax().item()
20 return id2label[predicted_class_id]
21
22print(predict("I feel amazing today!"))
Training Details
Training Data
The model was trained on the "synthetic_emotions.csv" dataset (available on Kaggle as prashanthan24/synthetic-emotions-dataset-14k-texts-7-emotions). This dataset contains 13,970 text samples labeled with one of seven emotions: Love, Sad, Anger, Fun, Hate, Surprise, Happiness. The dataset was split into training (60%), validation (20%), and test (20%) sets.
Training Procedure
Preprocessing
The text data was tokenized using DistilBertTokenizer. Input sequences were truncated and padded to a maximum length of 128 tokens.
Training Hyperparameters
- Learning Rate: 2e-5
- Per Device Train Batch Size: 16
- Per Device Eval Batch Size: 16
- Number of Train Epochs: 4
- Weight Decay: 0.01
- Metric for best model: macro_f1 (greater is better)
- Training regime: fp32
Evaluation
Testing Data, Factors & Metrics
Testing Data
The model was evaluated on a held-out test set (20% of the original dataset) from the "synthetic_emotions.csv" dataset.
Metrics
The evaluation metrics used were accuracy and macro F1-score.
Results
Summary
The model achieved the following performance on the test set:
- Test Loss: 0.0259
- Test Accuracy: 0.9946
- Test Macro F1-score: 0.9946
