Create README.md

README.md

@@ -0,0 +1,47 @@
+---
+license: mit
+datasets:
+- zeroshot/twitter-financial-news-sentiment
+---
+
+# Financial Sentiment Analysis with FinBERT
+
+This repository contains a financial sentiment analysis model fine-tuned on `ProsusAI/finbert`. The model classifies financial text (like tweets or news headlines) into three categories: **Bullish**, **Bearish**, or **Neutral**.
+
+The project includes scripts for data preprocessing, model training with hyperparameter optimization, and a Streamlit web application for interactive predictions.
+
+## Model Card
+
+### Model Description
+
+This model is a `BertForSequenceClassification` based on the `ProsusAI/finbert` architecture. It has been fine-tuned to predict the sentiment of financial text. The model was trained on a dataset of financial tweets and headlines, and it outputs one of three labels: `Bullish`, `Bearish`, or `Neutral`.
+
+```python
+from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
+
+MODEL_PATH = "path to your model"
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
+model = AutoModelForSequenceClassification.from_pretrained(MODEL_PATH)
+
+pipe = pipeline("text-classification", model=model, tokenizer=tokenizer)
+
+# Analyze sentiment
+results = pipe("Adobe price target raised to $350 vs. $320 at Canaccord")
+print(results)
+# [{'label': 'Bullish', 'score': 0.9...}]
+```
+
+### Training Data
+
+The model was trained on the [Twitter Financial News Sentiment](https://huggingface.co/datasets/zeroshot/twitter-financial-news-sentiment) dataset. The text data undergoes a comprehensive cleaning process (`data_preprocessing.py`) which includes:
+
+### Training Procedure
+
+The model was trained using the `transformers` library in PyTorch. The training script (`model_development.py`) includes the following features:
+
+- **Hyperparameter Optimization**: Optuna was used to find the best learning rate and batch size.
+- **Optimizer**: AdamW with a linear learning rate scheduler and warmup.
+- **Early Stopping**: Training stops if the validation accuracy does not improve for a set number of epochs.
+- **Mixed-Precision Training**: `torch.amp` was used for faster training.
+- **Gradient Accumulation**: To simulate a larger batch size.