pyFTS/pyFTS/data/rossler.py
2021-08-16 15:01:37 -03:00

51 lines
1.7 KiB
Python

"""
O. E. Rössler, Phys. Lett. 57A, 397 (1976).
dx/dt = -z - y
dy/dt = x + ay
dz/dt = b + z( x - c )
"""
import numpy as np
import pandas as pd
def get_data(var: str, a: float = 0.2, b: float = 0.2, c: float = 5.7, dt: float = 0.01,
initial_values: np.ndarray = [0.001, 0.001, 0.001], iterations: int=5000) -> np.ndarray:
"""
Get a simple univariate time series data.
:param var: the dataset field name to extract
:return: numpy array
"""
return get_dataframe(a, b, c, dt, initial_values, iterations)[var].values
def get_dataframe(a: float = 0.2, b: float = 0.2, c: float = 5.7, dt: float = 0.01,
initial_values: np.ndarray = [0.001, 0.001, 0.001], iterations: int=5000) -> pd.DataFrame:
'''
Return a dataframe with the multivariate Rössler Map time series (x, y, z).
:param a: Equation coefficient. Default value: 0.2
:param b: Equation coefficient. Default value: 0.2
:param c: Equation coefficient. Default value: 5.7
:param dt: Time differential for continuous time integration. Default value: 0.01
:param initial_values: numpy array with the initial values of x,y and z. Default: [0.001, 0.001, 0.001]
:param iterations: number of iterations. Default: 5000
:return: Panda dataframe with the x, y and z values
'''
x = [initial_values[0]]
y = [initial_values[1]]
z = [initial_values[2]]
for t in np.arange(0, iterations):
dxdt = - (y[t] + z[t])
dydt = x[t] + a * y[t]
dzdt = b + z[t] * x[t] - z[t] * c
x.append(x[t] + dt * dxdt)
y.append(y[t] + dt * dydt)
z.append(z[t] + dt * dzdt)
return pd.DataFrame({'x': x, 'y':y, 'z': z})