def train_anf(anfis, membership, mf, x_train, y_train, x_val, y_val, epc=10): best_result = np.Inf mfc = membership.membershipfunction.MemFuncs(mf) anf = anfis.ANFIS(x_train, y_train, mfc) anf.trainHybridJangOffLine(epochs=epc) prev = anfis.predict(anf, x_val) mse = MSE(y_val, prev) return anf, mse
def training(self):
ts = numpy.loadtxt("anfis/traindaya.txt", usecols=[0, 1, 2])
X = ts[:, 0:2]
Y = ts[:, 2]
mf = [[['gaussmf', {
'mean': -11.,
'sigma': 5.
}], ['gaussmf', {
'mean': -8.,
'sigma': 5.
}], ['gaussmf', {
'mean': -14.,
'sigma': 20.
}], ['gaussmf', {
'mean': -7.,
'sigma': 7.
}]],
[['gaussmf', {
'mean': -10.,
'sigma': 20.
}], ['gaussmf', {
'mean': -20.,
'sigma': 11.
}], ['gaussmf', {
'mean': -9.,
'sigma': 30.
}], ['gaussmf', {
'mean': -10.5,
'sigma': 5.
}]]]
mfc = membership.membershipfunction.MemFuncs(mf)
self.anf = anfis.ANFIS(X, Y, mfc)
self.anf.trainHybridJangOffLine(epochs=self.epochs)
# print round(anf.consequents[-1][0],6)
# print round(anf.consequents[-2][0],6)
# print round(anf.fittedValues[9][0],6)
if round(self.anf.consequents[-1][0], 6) == -5.275538 and round(
self.anf.consequents[-2][0], 6) == -1.990703 and round(
self.anf.fittedValues[9][0], 6) == 0.002249:
print 'test is good'
[
['gaussmf',{'mean':0.,'sigma':1.}],
['gaussmf',{'mean':-1.,'sigma':2.}],
['gaussmf',{'mean':-4.,'sigma':10.}],
['gaussmf',{'mean':-7.,'sigma':7.}]
],
[
['gaussmf',{'mean':1.,'sigma':2.}],
['gaussmf',{'mean':2.,'sigma':3.}],
['gaussmf',{'mean':-2.,'sigma':10.}],
['gaussmf',{'mean':-10.5,'sigma':5.}]
]
]
mfc = membership.membershipfunction.MemFuncs(mf)
anf = anfis.ANFIS(X, Y, mfc)
anf.trainHybridJangOffLine(epochs=20)
print round(anf.consequents[-1][0], 6)
print round(anf.consequents[-2][0], 6)
print round(anf.fittedValues[9][0], 6)
if round(anf.consequents[-1][0], 6) == -5.275538 and round(
anf.consequents[-2][0], 6) == -1.990703 and round(
anf.fittedValues[9][0], 6) == 0.002249:
print 'test is good'
anf.plotErrors()
anf.plotResults()
# Matt's test
""" mf = \
[
[
def test():
ts = pd.read_csv('event_driven_model.csv',
usecols=[1, 2, 3, 4, 5, 6, 7, 8],
encoding='gbk')
# 处理异常值
ts.fillna(0, inplace=True)
# 设置日期的起点。
dcounter = partial(tradedays_count, datetime(2010, 12, 30))
# 输入时间规则化
days = [datetime.strptime(str(time), '%Y%m%d') for time in ts['time']]
days = dcounter(np.asarray(days))
ts['time'] = days
# 其他输入变量预处理
ts['second_exist'] = ts['second_exist'].astype(np.int32)
ts.ix[ts['second_exist'] == 0, [5, 6, 7]] = 0
for col in range(ts.shape[1]):
if col == 0:
continue
ts.ix[:, col] = ts.ix[:, col].astype(np.float32)
ts['gap_time'] /= 360.0 # 归一化
for col in [0, 1, 2, 3]:
ts.ix[:, col] = (ts.ix[:, col] - ts.ix[:, col].min()) / (
ts.ix[:, col].max() - ts.ix[:, col].min())
print(ts)
ts = np.asarray(ts)
X = ts[:, 0:4] # 输入变量
Y = ts[:, 4:8] # 输出变量
# 设置隶属度函数
# mf_time = [['sigmf', {'b': dcounter(datetime(2014, 1, 1)), 'c': -0.05}],
# ['gaussmf', {'mean': dcounter(datetime(2015, 1, 1)), 'sigma': 100}],
# ['sigmf', {'b': dcounter(datetime(2015, 6, 30)), 'c': 0.2}],
# ['gaussmf', {'mean': dcounter(datetime(2015, 11, 15)), 'sigma': 30}]]
mf_time = [['gaussmf', {
'mean': 0.87,
'sigma': 0.1
}], ['gaussmf', {
'mean': 0.92,
'sigma': 0.3
}]]
mf_class = [['gaussmf', {
'mean': 0.1,
'sigma': 0.05
}], ['gaussmf', {
'mean': 0.3,
'sigma': 0.1
}], ['gaussmf', {
'mean': 0.6,
'sigma': 0.1
}], ['gaussmf', {
'mean': 0.9,
'sigma': 0.1
}]]
# mf_low = [['gaussmf', {'mean': 1, 'sigma': 5}],
# ['gaussmf', {'mean': 10, 'sigma': 10}],
# ['sigmf', {'b': 60, 'c': 0.01}]]
mf_low = [['gaussmf', {
'mean': 0.2,
'sigma': 0.1
}], ['gaussmf', {
'mean': 0.8,
'sigma': 0.1
}]]
mf_risepercent = [['gaussmf', {
'mean': 0.1,
'sigma': 0.05
}], ['gaussmf', {
'mean': 0.5,
'sigma': 0.3
}]]
# mf_risepercent = [['gaussmf', {'mean': 0.1, 'sigma': 0.1}],
# ['gaussmf', {'mean': 0.5, 'sigma': 0.5}],
# ['sigmf', {'b': 1, 'c': 0.02}]]
mf = [mf_time, mf_class, mf_low, mf_risepercent]
# mf = [mf_time, mf_class]
and_func = ['mamdani', 'T-S']
mfc = membership.membershipfunction.MemFuncs(mf)
anf = anfis.ANFIS(X, Y, mfc, andfunc=and_func[0])
anf.trainHybridJangOffLine(epochs=15, eta=0.00001)
print(anf.fittedValues)
print('mf list:')
print(anf.memFuncs)
anf.plotErrors()
# In[112]:
from membership import membershipfunction
mfc = membershipfunction.MemFuncs(mf)
# In[113]:
power_to_be_generated = df.pop('power_to_be_generated')
# In[114]:
import anfis
anf = anfis.ANFIS(df, power_to_be_generated, mfc)
train = anf.trainHybridJangOffLine(epochs=20)
# In[115]:
anf.plotErrors()
# In[116]:
anf.plotResults()
# In[104]:
print(train)
# In[ ]:
'gaussmf', {
'mean': np.mean(np.arange(45, 76)),
'sigma': np.std(np.arange(45, 76))
}
],
[
'gaussmf', {
'mean': np.mean(np.arange(55, 86)),
'sigma': np.std(np.arange(55, 86))
}
]]]
from membership import membershipfunction
mfc = membershipfunction.MemFuncs(mf)
import anfis
anf = anfis.ANFIS(train_data, train_label, mfc)
pred_train = anf.trainHybridJangOffLine(epochs=20)
train_label = np.reshape(train_label, [1, len(train_label)])
test_label = np.reshape(test_label, [1, len(test_label)])
print(train_label.shape)
print(test_label.shape)
error = np.mean((pred_train - train_label)**2)
print(error)
anf.plotErrors()
anf.plotResults()
c = np.asarray(datase["Result"])
img = ax.scatter(x, y, z, c=c, cmap=plt.hot())
fig.colorbar(img)
plt.show()
"""preparing membership function """
mf = [[['gaussmf',{'mean':0.,'sigma':1.}],['gaussmf',{'mean':-1.,'sigma':2.}],['gaussmf',{'mean':-4.,'sigma':10.}],['gaussmf',{'mean':-7.,'sigma':7.}]],
[['gaussmf',{'mean':1.,'sigma':2.}],['gaussmf',{'mean':2.,'sigma':3.}],['gaussmf',{'mean':-2.,'sigma':10.}],['gaussmf',{'mean':-10.5,'sigma':5.}]],
[['gaussmf',{'mean':0.,'sigma':1.}],['gaussmf',{'mean':-1.,'sigma':2.}],['gaussmf',{'mean':-4.,'sigma':10.}],['gaussmf',{'mean':-7.,'sigma':7.}]],
[['gaussmf',{'mean':1.,'sigma':2.}],['gaussmf',{'mean':2.,'sigma':3.}],['gaussmf',{'mean':-2.,'sigma':10.}],['gaussmf',{'mean':-10.5,'sigma':5.}]]]
mfc=membershipfunction.MemFuncs(mf)
"""training of the model"""
anf=anfis.ANFIS(dataset, target, mfc)
out=anf.trainHybridJangOffLine(epochs=10)
"""graphical plot of errors"""
anf.plotErrors()
"""actual vs trained graph"""
anf.plotResults()
tt=np.asarray(target)
"""cleaning of out values"""
ll=[]
for i in out:
}], ['gaussmf', {
'mean': -7.,
'sigma': 7.
}]],
[['gaussmf', {
'mean': -10.,
'sigma': 20.
}], ['gaussmf', {
'mean': -20.,
'sigma': 11.
}], ['gaussmf', {
'mean': -9.,
'sigma': 30.
}], ['gaussmf', {
'mean': -10.5,
'sigma': 5.
}]]]
mfc = membership.membershipfunction.MemFuncs(mf)
anf = anfis.ANFIS(all_inputs, all_targets, mfc)
anf.trainHybridJangOffLine(epochs=10)
print round(anf.consequents[-1][0], 6)
print round(anf.consequents[-2][0], 6)
print round(anf.fittedValues[9][0], 6)
if round(anf.consequents[-1][0], 6) == -5.275538 and round(
anf.consequents[-2][0], 6) == -1.990703 and round(
anf.fittedValues[9][0], 6) == 0.002249:
print 'test is good'
anf.plotErrors()
anf.plotResults()
"""
args_PSO = (args[0], args[1])
learners = args[2]
nPop = theta.shape[0]
J = np.zeros(nPop)
for i in range(nPop):
J[i] = learners[i].create_model(theta[i, :], args_PSO)
return J
# Init learners (one for each particle)
learners = []
for i in range(nPop):
learners.append(anf.ANFIS(n_mf=n_mf, n_outputs=n_outputs, problem=problem))
# Always normalize inputs
Xn_tr, norm_param = utl.normalize_data(X_tr)
Xn_te = utl.normalize_data(X_te, norm_param)
# Build boundaries using heuristic rules
LB, UB = utl.bounds_pso(Xn_tr,
n_mf,
n_outputs,
mu_delta=mu_delta,
s_par=s_par,
c_par=c_par,
A_par=A_par)
# Scale output(s) in continuous problems to reduce the range in <A_par>
