def gener_init_pq(self): # 生成初始搜索点(p0,q0)
rgs = BassEstimate(self.s)
P0, Q0 = rgs.optima_search()[1:3] # SABM最优点（P0,Q0）
p_range = np.linspace(0.4 * P0, P0, num=3)
q_range = np.linspace(0.2 * Q0 / self.k, 0.6 * Q0 / self.k, num=3)
to_fit = {}
params_cont = []
for p in p_range: # 取9个点用于确定参数与估计值之间的联系
for q in q_range:
diffu = Diffuse(p,
q,
g=self.G,
num_runs=self.s_len,
multi_proc=self.m_p)
s_estim = diffu.repete_diffuse()
s_estim_avr = np.mean(s_estim, axis=0)
rgs_1 = BassEstimate(s_estim_avr)
P, Q = rgs_1.optima_search()[1:3]
params_cont.append([p, q, P, Q])
to_fit = pd.DataFrame(params_cont, columns=['p', 'q', 'P', 'Q'])
result_p = smf.ols('p~P+Q-1', data=to_fit).fit()
result_q = smf.ols('q~P+Q-1', data=to_fit).fit()
p0 = result_p.params['P'] * P0 + result_p.params['Q'] * Q0
q0 = result_q.params['P'] * P0 + result_q.params['Q'] * Q0
return round(p0, 5), round(q0, 5) # 保留5位小数位，防止出错
def add_data(self, p, q):
diff = Diffuse(p, q, g=self.g, alpha=self.alpha, sigma=self.sigma)
x = np.mean(diff.repete_diffuse(), axis=0)
max_idx = np.argmax(x)
s = x[:(max_idx + 2)]
para_range = [[1e-6, 0.1], [1e-5, 0.8], [sum(s), 4*self.num_nodes]]
bassest = BassEstimate(s, para_range)
bassest.t_n = 1000
res = bassest.optima_search(c_n=200, threshold=10e-6)
return res[1:3] # P, Q
def r2_mse_abm(params, S=S):
p, q = params
T = len(S)
diffuse = Diffuse(p, q, num_runs=T, multi_proc=True)
diffuse_cont = diffuse.repete_diffuse(repetes=10)
X = np.mean(diffuse_cont, axis=0)
a = np.sum(np.square(X)) / np.sum(S) # 除以np.sum(S)是为减少a的大小
b = -2*np.sum(X*S) / np.sum(S)
c = np.sum(np.square(S)) / np.sum(S)
sigma = -b/(2*a) # 倍数
mse = np.sqrt(np.sum(S) * (4*a*c - b**2) / (4*a*T))
hat_S = list(X*sigma)
tse = np.sum(np.square(S - hat_S))
mean_y = np.mean(S)
ssl = np.sum(np.square(S - mean_y))
R_2 = (ssl - tse) / ssl
return R_2, mse, sigma
def fitness(individual, S=S):
p, q = individual
T = len(S)
if p <= 0:
raise Exception(f"p={p}小于0!")
elif q <= 0:
raise Exception(f"q={q}小于0!")
else:
diffuse = Diffuse(p, q, num_runs=T, multi_proc=True)
diffuse_cont = diffuse.repete_diffuse(repetes=10)
X = np.mean(diffuse_cont, axis=0)
a = np.sum(np.square(X)) / np.sum(S) # 除以np.sum(S)是为减少a的大小
b = -2*np.sum(X*S) / np.sum(S)
c = np.sum(np.square(S)) / np.sum(S)
mse = np.sqrt(np.sum(S) * (4*a*c - b**2) / (4*a*T))
# sigma = -b/(2*a) # 倍数
return mse
def get_M(self, p, q): # 获取对应扩散率曲线的最优潜在市场容量
if p <= 0:
raise Exception(f"p={p}小于0!")
elif q <= 0:
raise Exception(f"q={q}小于0!")
else:
diffu = Diffuse(p,
q,
g=self.G,
num_runs=self.s_len,
multi_proc=self.m_p)
s_estim = diffu.repete_diffuse()
x = np.mean(s_estim, axis=0)
a = np.sum(np.square(x)) / np.sum(
self.s) # 除以np.sum(self.s)是为减少a的大小
b = -2 * np.sum(x * self.s) / np.sum(self.s)
c = np.sum(np.square(self.s)) / np.sum(self.s)
mse = np.sqrt(
sum(self.s) * (4 * a * c - b**2) / (4 * a * self.s_len))
sigma = -b / (2 * a)
m = sigma * self.G.number_of_nodes()
return [mse, p, q, m], list(x * sigma)
def gener_init_params(S, g=nx.gnm_random_graph(10000, 30000)): # 生成初始搜索点(p0,q0)
rgs = BassEstimate(S)
P0, Q0 = rgs.optima_search()[1 : 3] # BASS最优点（P0,Q0）
p_range = np.linspace(0.3*P0, P0, num=3)
q_range = np.linspace(0.06*Q0, 0.3*Q0, num=3)
to_fit = {}
params_cont = []
for p in p_range: # 取9个点用于确定参数与估计值之间的联系
for q in q_range:
diffu = Diffuse(p, q, g=g, num_runs=len(S), multi_proc=True)
s_estim = diffu.repete_diffuse()
s_estim_avr = np.mean(s_estim, axis=0)
rgs_1 = BassEstimate(s_estim_avr)
P, Q = rgs_1.optima_search()[1: 3]
params_cont.append([p, q, P, Q])
to_fit = pd.DataFrame(params_cont, columns=['p', 'q', 'P', 'Q'])
result_p = smf.ols('p~P+Q-1', data=to_fit).fit()
result_q = smf.ols('q~P+Q-1', data=to_fit).fit()
p0 = result_p.params['P']*P0 + result_p.params['Q']*Q0
q0 = result_q.params['P']*P0 + result_q.params['Q']*Q0
return round(p0, 5), round(q0, 5) # 保留5位小数位，防止出错
def func_1(p, q, sigma):
p = round(p, 5)
q = round(q, 5)
diff = Diffuse(p=p, q=q, sigma=sigma, num_runs=35, multi_proc=True)
x = np.mean(diff.repete_diffuse(), axis=0)
return str([p, q]), list(np.concatenate(([p, q], x)))
def func(p, q, sigma):
diff = Diffuse(p, q, sigma=sigma, num_runs=40)
x = np.mean(diff.repete_diffuse(), axis=0)
return np.concatenate(([p, q], x))
prj = db.indivHeter
#%%
def func(p, q, sigma):
diff = Diffuse(p, q, sigma=sigma, num_runs=40)
x = np.mean(diff.repete_diffuse(), axis=0)
return np.concatenate(([p, q], x))
#%%[markdown]
#### 扩散示例
#%%
sigma = 0.1
p, q = 0.001, 0.05
t1 = time.perf_counter()
diff = Diffuse(p, q, sigma=sigma)
diff_cont = diff.repete_diffuse()
S = np.mean(diff_cont, axis=0)
print(f"Time elasped:{time.perf_counter()-t1:.2f}s")
print(f"网络:{sigma}")
print(f"最大采纳量{np.max(S):.0f}, 最大时间步:{np.argmax(S)}")
#%%
m_idx = np.argmax(S)
s = S[:m_idx + 2]
t1 = time.process_time()
para_range = [[1e-5, 0.1], [1e-5, 0.8], [sum(s), 10*sum(s)]]
bassest = BassEstimate(s, para_range)
mse, P, Q, M = bassest.optima_search(c_n=100, threshold=10e-6)
r_2 = bassest.r2([P, Q, M])
print(f'Time elapsed: {(time.process_time() - t1):.2f}s')
# coding=utf-8
import numpy as np
import networkx as nx
import time
import random
from abmdiffuse import Diffuse
if __name__ == '__main__':
t1 = time.perf_counter()
p, q, alpha = 0.01, 0.5, 1
diffu = Diffuse(p, q, alpha=alpha)
diffu_cont = diffu.repete_diffuse(repetes=10)
print(
f"参数设置: p--{p}, q--{q}, alpha--{alpha} network--{diffu.g.number_of_nodes()}"
)
print(f"用时{time.perf_counter() - t1:.2f}秒")
def func(p, q, g):
p = round(p, 5)
q = round(q, 5)
diff = Diffuse(p, q, g=g, num_runs=35)
x = np.mean(diff.repete_diffuse(), axis=0)
return str([p, q]), list(np.concatenate(([p, q], x)))
for line in diff_list:
pl.plot(line[2:])
#%%
for line in diff_list[:15]:
pl.plot(line[2:])
#%%
param_bound = ws0["param_boundary"]
print(param_bound)
#%%
t1 = time.perf_counter()
g = nx.watts_strogatz_graph(10000, 6, 0)
p, q = 0.001, 0.1
diff = Diffuse(p, q, g=g)
diff_cont = diff.repete_diffuse(20)
S = np.mean(diff_cont, axis=0)
print(f"Time elasped:{time.perf_counter()-t1:.2f}s")
print(f"网络: nx.watts_strogatz_graph(10000, 6, 0)")
print(f"最大采纳量{np.max(S):.0f}, 最大时间步:{np.argmax(S)}")
m_idx = np.argmax(S)
s = S[:m_idx + 2]
t1 = time.process_time()
para_range = [[1e-5, 0.1], [1e-5, 0.8], [sum(s), 10 * sum(s)]]
bassest = BassEstimate(s, para_range)
mse, P, Q, M = bassest.optima_search(c_n=100, threshold=10e-6)
r_2 = bassest.r2([P, Q, M])
print(f'Time elapsed: {(time.process_time() - t1):.2f}s')
print(f'P:{P:.4f} Q:{Q:.4f} M:{M:.0f}\nr^2:{r_2:.4f}')
# coding=utf-8
import numpy as np
import networkx as nx
import time
import random
from abmdiffuse import Diffuse
if __name__ == '__main__':
import pylab as pl
t1 = time.perf_counter()
p, q = 0.001, 0.05
diffu = Diffuse(p=p, q=q, sigma=0.1)
diffu_cont = diffu.repete_diffuse(repetes=20)
print(f"参数设置: p--{p}, q--{q} sigma--{diffu.sigma}")
print(f"用时{time.perf_counter() - t1:.2f}秒")
fig = pl.figure(figsize=(12, 6))
ax = fig.add_subplot(1, 1, 1)
for line in diffu_cont:
ax.plot(line, 'k-', lw=0.5, alpha=0.5)
ax.plot(np.mean(diffu_cont, axis=0), 'r-', lw=2)
pl.show()