FactorLib

A python library for creating custom factors, and running walk-forward optimization with various portfolio optimization options.

Features

• Parallel processing for custom factor data creation with highly customizable interface.
• Walk forward optimization with various portfolio optimization options.
• Performance visualization and sharpe ratios for benchmarking.
• Integration with SHAP values for model interpretability.

Prerequisites

The following software needs to be installed to use FactorLib:

• python
• pandas
• numpy
• scikit-learn
• scipy
• xgboost
• ray
• tqdm
• jupyter
• shap
• catboost
• lightgbm
• QuantStats
• matplotlib
• pyarrow
• fastparquet
• ipywidgets
• yfinance
• prettytable

To install all of the required dependencies run:

pip install -r requirements.txt

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Installation

1. Clone the repo

   git clone https://github.com/your_username_/Project-Name.git

2. Install pip packages

   pip install -r requirements.txt

Usage

factorlib

The factorlib directory contains all of the source code for the FactorLib library. The most important files to get started are:

- **factor.py**: Defines the `Factor` class, which is used to store and prepare data to later be passed into
 a `FactorModel`.
- **factor\_model.py**: Defines the `FactorModel` class, which is responsible for training and evaluating a model
using the provided factors. 

The recommended workflow for using FactorLib is as follows:

1. Prepare your factor data in a pandas DataFrame. The DataFrame should have a multi-index with levels named 'date' and 'ticker'. Each column in the DataFrame represents a different factor.
2. Instantiate a Factor object, passing in the factor data and other relevant parameters, such as the factor name, the time interval for the data (e.g. daily, monthly), and any transformations to apply to the data.
3. Instantiate a FactorModel object, passing in the desired model type (e.g. LightGBM), tickers used for training and evaluation, and the time interval for the data.
4. Add the Factor object to the FactorModel using the add_factor() method.
5. Call the wfo() (walk-forward optimization) method on the FactorModel object. This will train and evaluate the model over the specified time period, using a rolling window approach. The wfo() method returns a Statistics object that contains various performance metrics for the model.
6. Analyze the results using the various methods provided by the Statistics object, such as snapshot(), stats_report, beeswarm_shaps() and waterfall_shaps().

The file debug.py is an example of how to use the Factor and FactorModel classes in conjunction to train a model and visualize the results.