financial-datasets 0.1.18

Financial datasets for LLMs

Financial Datasets 🧪

Financial Datasets is an open-source Python library that lets you create question & answer financial datasets using Large Language Models (LLMs). With this library, you can easily generate realistic financial datasets from a 10-K, 10-Q, PDF, and other financial texts.

Usage

Example generated dataset:

[
  {
    "question": "What was Airbnb's revenue in 2023?",
    "answer": "$9.9 billion",
    "context": "In 2023, revenue increased by 18% to $9.9 billion compared to 2022, primarily due to a 14% increase in Nights and Experiences Booked of 54.5 million combined with higher average daily rates driving a 16% increase in Gross Booking Value of $10.0 billion."
  },
  {
    "question": "By what percentage did Airbnb's net income increase in 2023 compared to the prior year?",
    "answer": "153%",
    "context": "Net income in 2023 increased by 153% to $4.8 billion, compared to the prior year, driven by our revenue growth, increased interest income, discipline in managing our cost structure, and the release of a portion of our valuation allowance on deferred tax assets of $2.9 billion."
  }
]

Example #1 - generate from any text

Most flexible option. Generates dataset using a list of string texts. Colab code example here.

from financial_datasets.generator import DatasetGenerator

# Your list of texts
texts = ...

# Create dataset generator
generator = DatasetGenerator(model="gpt-4-turbo", api_key="your-openai-key")

# Generate dataset from texts
dataset = generator.generate_from_texts(
    texts=texts,
    max_questions=100,
)

Example #2 - generate from PDF

Generate a dataset using a PDF url only. Colab code example here.

from financial_datasets.generator import DatasetGenerator

# Create dataset generator
generator = DatasetGenerator(model="gpt-4-turbo", api_key="your-openai-key")

# Generate dataset from PDF url
dataset = generator.generate_from_pdf(
    url="https://www.berkshirehathaway.com/letters/2023ltr.pdf",
    max_questions=100,
)

Example #3 - generate from 10-K

Generate a dataset using a ticker and year. Colab code example here.

from financial_datasets.generator import DatasetGenerator

# Create dataset generator
generator = DatasetGenerator(model="gpt-4-turbo", api_key="your-openai-key")

# Generate dataset from 10-K
dataset = generator.generate_from_10K(
    ticker="AAPL",
    year=2023,
    max_questions=100,
    item_names=["Item 1", "Item 7"],  # optional - specify Item names to use
)

Installation

Using pip

You can install the Financial Datasets library using pip:

pip install financial-datasets

Using Poetry

If you prefer to use Poetry for dependency management, you can add Financial Datasets to your project:

poetry add financial-datasets

From the Repository

If you want to install the library directly from the repository, follow these steps:

1. Clone the repository:

   git clone https://github.com/virattt/financial-datasets.git
   

2. Navigate to the project directory:

   cd financial-datasets
   

3. Install the dependencies using Poetry:

   poetry install
   

4. You can now use the library in your Python projects.

Contributing

Contributions are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request.

License

This project is licensed under the MIT License.

Contributors