Dyna-Mo is a "dynamical model" simulation framework to simualate Physics based time-series on Networks. Given an input model, initial conditions and input graph, Dyna-mo simulates the trajectory of the dynamical system. We are also including two key statistical features.
- Understanding the statistics of the underlying graph.
- Simple statistics based forecasting (Physics-agnostic)
- input model using symbolic math (will be built on sympy)
- Time-series analysis -- statistics (tbd)
- Advanced analysis -- Times Net/Prophet/TSlib etc
To get started clone the dynamo repository. After two months this will be pushed to PyPi as a standalone package.
git clone git@github.com:CMSC720-Foundation-in-Deep-Learning/dynamo.git
Simulating dynamical model for given input pamaters and model
init = 1. + np.random.uniform(0.,30.,size=(10,))
tspan = np.arange(0,100,1)
adjacency_matrix = np.random.randint(2, size=(10,10))
physics_model = PhysicsModel(name='kuramoto2', time=tspan) # Assigning tspan to time
omega_value = 1 + 10 * np.random.rand(10,) # np.random needs to be np.random.rand
y = odeint(physics_model.dynamical_model, init, tspan, args=(physics_model.adjacency_matrix,omega_value))Plotting the time series and the underlying network.
plot_timeseries(y, t)
plot_boolean_graph(A):
Started by Raunak Dey. Open to contribution from April'24 onwards.