distilbert-base-turkish-cased-emotion

Model description:

Distilbert-base-turkish-cased finetuned on the emotion dataset (Translated to Turkish via Google Translate API) using HuggingFace Trainer with below Hyperparameters

 learning rate 2e-5, 
 batch size 64,
 num_train_epochs=8,

Model Performance Comparision on Emotion Dataset from Twitter:

Model	Accuracy	F1 Score	Test Sample per Second
Distilbert-base-turkish-cased-emotion	83.25	83.17	232.197

How to Use the model:

from transformers import pipeline
classifier = pipeline("text-classification",
                       model='zafercavdar/distilbert-base-turkish-cased-emotion',
                       return_all_scores=True)
prediction = classifier("Bu kütüphaneyi seviyorum, en iyi yanı kolay kullanımı.", )
print(prediction)

"""
Output:
[
  [
    {'label': 'sadness', 'score': 0.0026786490343511105},
    {'label': 'joy', 'score': 0.6600754261016846},
    {'label': 'love', 'score': 0.3203163146972656},
    {'label': 'anger', 'score': 0.004358913749456406},
    {'label': 'fear', 'score': 0.002354539930820465},
    {'label': 'surprise', 'score': 0.010216088965535164}
  ]
]

"""

Dataset:

Twitter-Sentiment-Analysis.

Eval results

{
 'eval_accuracy': 0.8325,
 'eval_f1': 0.8317301441160213,
 'eval_loss': 0.5021793842315674,
 'eval_runtime': 8.6167,
 'eval_samples_per_second': 232.108,
 'eval_steps_per_second': 3.714
}