gdnb 0.0.1

Implement the GDNB algorithm

GDNB

This is a light package to implement the GDNB (giant-component-based dynamic network biomarker/marker) algorithm, which is a data-driven and model-free method and can be used to detect the tipping points of the phase transition in complex systems.

---

Install

pip install gdnb

---

Demonstrate the intuition of GDNB with 2D Ising model

1. import packages

import gdnb
gdnb.set_default_style()

2. perform MC simulations under 9 temperatures (from 1 to 5 with an interval of 0.5)

lattice = 10 # 10 x 10 2D Ising model
sim_steps = 100000 # simulation length
seed = 1 # random number seed

traj = gdnb.load_ising(lattice=lattice, sim_steps=sim_steps, seed=seed)

visualize the simulations

import numpy as np
import matplotlib.pyplot as plt

def plot_traj(traj, c='b'):
    '''plot the simulated trajectory.'''
    avg_mag = np.mean(traj, axis=1)
    ts = np.arange(1, 5.1, .5)
    fig, ax = plt.subplots(len(traj), 1, figsize=(6, 10),
                           sharex=True, sharey=True)
    for i, mag in enumerate(avg_mag):
        ax[i].plot(mag, c=c)
        ax[i].set_ylim(-1.1, 1.1)
        ax[i].set_yticks(np.arange(-1, 1.1, .5))
        ax[i].set_ylabel('T={}'.format(ts[i]), fontsize=15)
        ax[i].tick_params(direction='in', labelsize=13)
    ax[-1].set_xlabel('Snapshot', fontsize=15)
    fig.tight_layout()    
    return fig, ax

fig, ax = plot_traj(traj)

This figure shows that the average magnetizations, defined by the average absolute value of all spins, change over time at different temperatures.

def cal_mean_std(traj):
    '''calculate the average magnetization and corresponding SD.'''
    means = []
    stds = []
    for i, state in enumerate(traj):    
        tmp = np.sum(state, axis=0) / len(state)
        means.append(abs(np.mean(tmp)))
        stds.append(np.std(tmp))
    return means, stds

def plot_magnetization(means, stds, c='teal'):
    '''plot the average magnetization with errorbar.'''       
    fig, ax = plt.subplots(figsize=(6, 4))
    ax.errorbar(range(len(means)), means, yerr=stds, fmt='-o', capsize=5,
                capthick=2, c=c)
    ax.tick_params(direction='in', labelsize=13)
    ax.grid(axis='y')
    ax.set_xticks(np.arange(9))
    ax.set_xticklabels(np.arange(1., 5.1, .5))
    ax.set_yticks(np.arange(-.6, 1.21, .3))
    ax.set_xlabel('Temperature', fontsize=15)
    ax.set_ylabel('Magnetization', fontsize=15)
    fig.tight_layout()
    return fig, ax

means, stds = cal_mean_std(traj)
fig, ax = plot_magnetization(means, stds)

It can be seen that our simulations can properly describe the phase transition of the Ising model.

3.apply the GDNB algorithm to this simulated data

3.1 instantiation

The core class of this package is GDNBData. Using a 3D array (mxnxs) to instantiate, where m, n and s are the numbers of observations, variables and repeats, respectively.

mydata = gdnb.GDNBData(traj)

3.2 Use .fit() to perform analysis

pv_cut = 0.05
pcc_cut = 0.6
mydata.fit(pv_cut=pv_cut, pcc_cut=pcc_cut)

3.3 visualize the results of GDNB

fig, ax = mydata.plot_index(xticklabels=np.arange(1, 5.1, 0.5), xlabel='Temperature')

The composite index (CI) has an obvious peak at T=2.5, indicating the transition temperature of this system is 2.5, which agrees well with the theoretical critical temperature (2.27) among all 9 temperatures. To show the transition more clear, the transition core is mapped on the Ising model.

def plot_ising_maps(data, lattice=10, s=100, ncol=3, figsize=(12, 12), 
                    c_list=['wheat','gray','red']):
    points = np.array([
        [j, i] for i in range(lattice) for j in range(lattice)])
    m = len(data.sele_index)
    Ts = np.arange(1, 5.1, .5)
    fig, ax = plt.subplots(int(m/ncol), ncol, figsize=figsize)
    ax=ax.flatten()

    for t in range(m):
        colors=[c_list[0]] * lattice ** 2
        for i in range(lattice ** 2):
            if i in data.sele_index[t][
                    np.where(
                        data.cls_index[t] == data.largest_cls_index[t])]:
                colors[i] = c_list[2]
            elif i in data.sele_index[t]:
                colors[i] = c_list[1]
            else:
                continue
        ax[t].scatter(points[:,0], points[:,1], color=colors, s=s)
        ax[t].set_title('T={:2.1f}'.format(Ts[t]), fontsize=18)
        ax[t].set_axis_off()        
    return fig, ax

fig, ax = plot_ising_maps(mydata)

As shown in this figure, The giant component (red dots) only appears at T=2.5.

---

Reproduce the results in the paper

To reproduce the results in the paper, check the notebook files and results in paper_examples.

---

License

MIT License