time-series-analysis

Overview

Most events may be viewed, or converted to time-series. The purpose of this project is to provide tools to analyze time-series for the purpose of estimating "Roughness", "Fractal Dimension" and help create some intuition of the predictability and exposure to extremities.

Project Structure & Setup

- datasets/
    (folder holding datasets)
- generators/ 
    (Jupyter Notebooks for generating datasets)
- viewers/
    (folder holding various dataset viewers)
- analyzers/
    (folder holding various dataset analysis tools)

Install requirements by:

pip install -r requirements.txt

Generators

generator_randomwalk : generating a random walk datas-et. Data is written into the datasets folder, as CSV file. Filename include rw prefix and timestamp. The notebook includes basic visualization and analysis of the generated walk.
generator_whitenoise : generating a "white noise" data-set. Data is written into the datasets folder, as CSV file. Filename include wn prefix and timestamp. The notebook includes basic visualization and analysis of the generated series.
generator_basic_fractal: generating a basic "fractal" data-set. The Fractal is generated around a random trend-line to simulate financial data. Data is written into the datasets folder, as CSV file. Filename include bf prefix and timestamp. The notebook includes basic visualization and analysis of the generated series.
generator_pink_noise: generating a "pink noise" data-set. Data is written into the datasets folder, as CSV file. Filename include pn prefix and timestamp. The notebook includes basic visualization and analysis of the generated series.

Viewers

basic: Notebook w/ basic visualization for a data-set.
powers: Notebook for comparison between the data-set and various functions, viewing on various scales to provide intuition on the dataset power and behavior.
stationary-vs-non-stationary: calculates and presents the mean and variance of the timeseries over time. Identify weak-stationary-process using auto-covariance. The presentation should help forming an intuition of the timeseries is a result of a stationary or non-stationary process.
lag view (iid analysis): presents lag plot of the time-series, with lag 1, 2, 4, 8, 16 (1 & 2 lags also showed w log scale). The view is helpful to get some basic intuition on the independeance assumption of the time series, and the autocorrelation
deltas: present visualizations of the time-series deltas, as well as visualized the probability for a streak of upwards and downwards trends. For financial time-series, helps to form intuition on volatility and decision making processes.

Analysers