Cryptocurrency Price Forecasting using Variational AutoEncoder with Versatile Quantile Modeling

This repository is the official implementation of Cryptocurrency Price Forecasting using Variational AutoEncoder with Versatile Quantile Modeling (CIKM, 2024).

NOTE: This repository supports WandB MLOps platform!

project homepage: https://crypto-vae.streamlit.app/

Training & Evaluation

1. Training Proposed Method

python main.py --model <model>

• <model> options: GLD_finite, GLD_infinite, LSQF, ExpLog
• detailed configuration files can be found in configs folder

2. Training Benchmark Methods

python main.py --model 'TLAE'
python main.py --model 'ProTran'
python benchmarks/deepar.py 
python benchmarks/gp_copula.py 
python benchmarks/mqrnn.py 
python benchmarks/sqf_rnn.py 
python benchmarks/tft.py
python benchmarks/benchmark_eval.py --model 'TFT' --tau 1

• detailed configuration files can be found in configs folder for TLAE and ProTran
• pretrained weights for DeepAR, GP-copula, MQRNN, SQF-RNN, TFT can be found in /assets folder

codes for evaluation

• step-by-step evaluation for our proposed method: infernce.py
• step-by-step evaluation for benchmark methods: benchmarks/benchmark_eval.py

Directory and Codes

.
+-- assets (includes visualization results and pretrained weights of benchmark methods)
+-- benchmarks (includes codes for training benchmark methods)
+-- config (includes detailed configuration files)
+-- module 
+-- inference.py
+-- main.py
+-- LICENSE
+-- README.md

Citation