# creating dataframe without overwriting (so additing)
# if you start a new set of data, make sure you start with a fresh file
filename = 'data_2-regular.csv'
overwrite = False
if overwrite == True:
output = pd.DataFrame()
else:
output = load_csv(filename)
p_max = 10 # maxdepth
n_min, n_max = 3, 20
N = 1 # number of graphs per node number n
d = 2 # degree of regular graphs
for n in range(n_min, n_max + 1):
for s in range(N):
G = nx.cycle_graph(n)
graph_type = str(d) + '-regular_' + str(n) + '-nodal'
results = interp_pyquil(G, p_max)
for i, results_i in results.items():
results_i['graph_name'] = graph_type
results_i['seed'] = s
results_i['GW_cutvalue'] = goemans_williamson(G)
output = output.append(results_i, ignore_index=True)
# saving every round so the process can be stopped prematurely without losing all progress
output.to_csv(filename)
# plt.ylabel(r'$F_p$')
# plt.xlabel('p')
# plt.xlim([0.8,p_max+0.2])
r_sum = np.zeros(p_max)
label = True
df = df[(df['n_nodes'] == n) & (df['seed'] < seed_max)]
r_GW_mean = 0
for i in range(0, len(df), p_max):
DF = df[df['n_nodes'] == n].iloc()
Fp = np.array([DF[i + j]['Fp'] for j in range(p_max)])
Cmax = np.array([DF[i + j]['Cmax'] for j in range(p_max)])
GW = np.mean(
[goemans_williamson(DF[i]['graph']) / Cmax[0] for _ in range(10)])
print(GW)
r_GW_mean += GW / len(range(0, len(df), p_max))
r = Fp / Cmax
r_sum += r
xticks = np.arange(1, p_max + 1)
#plt.title(r"$F_p$ for 12-nodal 3-regular graphs for $p = 1...10$")
plt.yscale("log")
if label:
plt.plot(xticks, 1 - r, color='peachpuff', linestyle='dashed')
plt.plot(xticks,
1 - r,
color='darkorange',
@author: joost
"""
# graphs
import networkx as nx
# solvers
from PSO import PSO
from GW import goemans_williamson
from INTERP import INTERP
from FOURIER import FOURIER
G = nx.random_regular_graph(3, 10)
p = 4
# Goemans Williamson
cut_value_GW = goemans_williamson(G)
print("GW", cut_value_GW)
# Particle swarm
pso = PSO(G)
gamma_pso, beta_pso = pso.get_angles_PSO(p, debug=True, maxiter=10)
cut_value_pso = pso.best_sampled(gamma_pso, beta_pso)
print("PSO", cut_value_pso)
# INTERP
interp = INTERP(G)
gamma_int, beta_int = interp.get_angles_INTERP(p)
cut_value_interp = interp.best_sampled(gamma_int, beta_int)
print("INT", cut_value_interp)
# FOURIER
plt.grid('on')
plt.ylabel(r'$r$')
plt.xlabel('p')
# figure 2
plt.figure()
r_sum = np.zeros(10)
label = True
r_GW_mean = 0
for i in range(0,280,10):
Fp = [df['Fp'][i+j] for j in range(10)]
Cmax = [ceil(df['Fp'][i+9]) for j in range(10)]
GW = goemans_williamson(df['graph'][i])
r_GW_mean += GW/Cmax[0]/len(range(0,len(df),p_max))
r = np.array([Fp[i]/Cmax[i] for i in range(10)])
r_sum += r
xticks = np.arange(1,10+1)
#plt.title(r"$F_p$ for 12-nodal 3-regular graphs for $p = 1...10$")
plt.yscale("log")
if label:
plt.plot(xticks,1-r, color = 'peachpuff', linestyle = 'dashed')
plt.plot(xticks,1-r, color = 'darkorange', linestyle = 'None', marker = 'o', label = 'random instance')
label = False
else:
plt.plot(xticks,1-r, color = 'peachpuff', linestyle = 'dashed')
#plt.title(r"$F_p$ for 12-nodal 3-regular graphs for $p = 1...10$")
plt.plot(xticks,r, color = 'lightgreen', linestyle = 'dashed')
plt.plot(xticks,r, color = 'green', linestyle = 'None', marker = 'o')
plt.grid('on')
plt.ylabel(r'$F_p$')
plt.xlabel('p')
plt.xlim([0.8,p_max+0.2])
plt.figure()
r_sum = np.zeros(p_max)
label = True
r_GW_mean = 0
for i in range(0,len(df),p_max):
Fp = np.array([df['Fp'][i+j] for j in range(p_max)])
Cmax = np.array([df['Cmax'][i+j] for j in range(p_max)])
GW = np.mean([goemans_williamson(df['graph'][i]) for _ in range(10)])/Cmax[0]
print(GW)
r_GW_mean += GW/len(range(0,len(df),p_max))
r = Fp/Cmax
r_sum += r
xticks = np.arange(1,p_max+1)
#plt.title(r"$F_p$ for 12-nodal 3-regular graphs for $p = 1...10$")
plt.yscale("log")
if label:
plt.plot(xticks,1-r, color = 'peachpuff', linestyle = 'dashed')
plt.plot(xticks,1-r, color = 'darkorange', linestyle = 'None', marker = 'o', label = 'random instance')
label = False
else:
plt.plot(xticks,1-r, color = 'peachpuff', linestyle = 'dashed')
df = load_csv(pre + filename + post)
r_sum = np.zeros(p_max)
r_GW = 0
label = True
# looping over the different graphs
for k, i in enumerate(range(0, len(df), p_max)):
print(k, i)
Fp = np.array([df['Fp'][i + j] for j in range(p_max)])
Cmax = np.array([df['Cmax'][i + j] for j in range(p_max)])
G = df['graph'][i]
r = Fp / Cmax
r_GW += np.mean([goemans_williamson(G) for i in range(10)]) / Cmax[0]
r_sum += r
xticks = np.arange(1, p_max + 1)
r_mean = r_sum / len(range(0, len(df), p_max))
r_GW = r_GW / len(range(0, len(df), p_max))
# Mean, including a fit
from scipy import optimize
def test_func(p, a, b):
if not SQR:
return a * np.exp(-p / b)
if SQR:
return a * np.exp(-np.sqrt(p / b))