BTCWavelet:

An attempt to predict Bitcoin bubble crashes.

Implementation of Log-Periodic Power Law fit on Bitcoin hourly data series
Crawling hourly Bitcoin data Using Cryptocompare Historical Data API
Reconstructing Bitcoin Time series using Discrete Wavelet Transform using PyWavelets
Implementation of the Filimonov & Sornette Epsilon Drawdown algorithm
Some Genetic Algorithm attempt using SAWADA Takahiro code

Content:

1. Creating Anaconda Environment:

The easiest way to replicate the Python environment needed to run this code is to use the file conda-spec.txt to create an identical local conda environment by invoking the following command from the repository directory:

conda create --name tensorflow --file conda-spec.txt

For more information on how conda environments work refer to Conda user guide: Building Identical Conda Environments

2. Running the code:

All the experiments starting points are in lppl_ga.py. The code will default to run a 1000 Bitcoin trials on hourly data with DWT enabled. I might create a command-line interface for the different experiments in the future.

3. Code Files:

File Name	Contained Classes
Lppl_ga.py	Pipeline, Nonlinear_Fit
Epsilon.py	Data_Wrapper, Epsilon_Drawdown
Decomposition.py	Wavelet_Wrapper
Crawler.py	BaseCrawler, Crawler, … exchange specific crawlers were not used

4. Classes Description:

Class name	Description
BaseCrawler	Basic http, requesting API endpoint functions
Crawler	Cryptocompare crawler. The Bitcoin hourly data source.
Data_Wrapper	Reads local Bitcoin and other financial assets data and returns them on a dataframe with a standard column formatting
Epsilon_Drawdown	Implementation of (Gerlach, et al., 2018) Epsilon Drwadown algorithm
Nonlinear_Fit	Implementation of LPPL non-linear optimization. The main algorithm used across experiments is Basin Hopping.
Pipeline	The model pipeline from crawling to bubble prediction. Incomplete
Wavelet_Wrapper	Implementation of DWT reconstruction using PyWavelets6 library