This repository contains the solution for team HODL from the Signate Tokyo Stock Exchange Fundamentals Modeling Competition.

1. Create conda environment signate_jpx with environment.yml file.

   conda env create -f environment.yml

2. Activate signate_jpx conda env

   conda activate signate_jpx

3. Add csv.gz files to data/raw folder

4. To begin an end to end training ensure all data is in data/raw and run the following. Trained models should be outputted to the models folder.

   sh start_pipeline.sh baseline_model 

To update environment.

conda env update --file environment.yml --prune

The below contains a description of parameters in config.yml which guide the overall training pipeline. Depending on the setting of parameters, the behavior of the pipeline will change.

• test_model: Whether to use public or private mode, Options: ["public", "private"]
• random_seed: random seed for models, Ex: 255
• low_memory_mode: Mode for debugging end to end pipeline quicker, uses only 10000 rows. Options: [True, False]
• use_fin_data: Whether to use the financial data, we have set to false as it did not help us much. Options: [True, False]
• data_date_limit: Maximum date for data, Ex: "2021-01-01"
• train_split_date: Maximum date for training data Ex: "2020-01-01"
• test_split_date: Starting date for test data Ex: "2020-01-01"
• drop_data: If we want to drop data from train and test --> Avoid Data leak -> drop 2 months data at end of dates Options: [True, False]
• drop_data_train_date: Drop data before this date. Ex: "2016-02-01"
• drop_data_test_date: Drop test data after this date. Ex: "2020-01-01"
• cross_validation: For internal validation, performs cross validation when training. Options: [True, False]
• lgb_model: Whether to use LightGBM model. Options: [True]
• use_test_as_validation: Whether to use the test data as validation set. Options: [True, False]
• train_with_all_data: Used for training with all data, this was used for final submission as using more data is beneficial for improving model performance. Options: [True, False]
• lgb_params: LightGBM model hyperparameters