def FA_ELM(X_train, Y_train, n_hidden=64):
## FA优化参数
# 参数向量 parameters [n N_iteration alpha betamin gamma]
# n为种群规模，N_iteration为迭代次数
time1 = time.time()
para = [20, 50, 0.3, 0.2, 1]
#待优化参数上下界 Simple bounds/limits for d-dimensional problems
d = 2
# 待优化参数个数
#weight(-1,1) bias(0,1)
Lb = np.array([-1, 0])
# 下界
Ub = np.array([1, 1])
# 上界
# 迭代寻优 Display results
[bestsolutio, bestojb] = ffa_mincon_elm(d, Lb, Ub, para, X_train, Y_train)
## 打印参数选择结果
bestw = bestsolutio[1]
bestb = bestsolutio[0]
#以最佳参数训练ELM
rhl = RandomLayer(n_hidden=64, alpha=1.0) #表示_use_mlp_input=true
elmr = GenELMRegressor(hidden_layer=rhl)
elmr.fit(X_train, Y_train, bias=bestb, weights=bestw)
time2 = time.time()
useTime = time2 - time1
print("总用时：%d秒" % (useTime))
print(bestw)
print(bestb)
return elmr
def train_ann(X_train, targets, f_name):
# training MLP
# n_neurons = 16
# mlp = MLPRegressor(hidden_layer_sizes=(n_neurons, n_neurons, n_neurons), activation='tanh', solver='lbfgs', tol=1e-12) #, alpha=1e-01
# mlp.fit(X_train, targets)
# tr_mse = mean_squared_error(targets, predict(mlp, X_train))
# print("n iterations: ", mlp.n_iter_)
# print("training mse: ", tr_mse)
# # save model
# joblib.dump(mlp, f_name)
#
# return (mlp, tr_mse)
elm = pipeline.Pipeline([('rhl',
RandomLayer(n_hidden=256,
activation_func='multiquadric',
alpha=0.5,
rbf_width=0.3)),
('lr', LinearRegression(fit_intercept=True))])
elm.fit(X_train, targets)
tr_mse = mean_squared_error(targets, predict(elm, X_train))
print("training mse: ", tr_mse)
# save model
joblib.dump(elm, f_name)
return (elm, tr_mse)
def _create_random_layer(self):
"""Pass init params to RandomLayer"""
return RandomLayer(n_hidden=self.n_hidden,
alpha=self.alpha, random_state=self.random_state,
activation_func=self.activation_func,
activation_args=self.activation_args,
user_components=self.user_components,
rbf_width=self.rbf_width)
def train_ann(X_train, targets):
elm = pipeline.Pipeline([('rhl', RandomLayer(n_hidden=85, activation_func='tanh', alpha=0.65)),
('lr', LinearRegression(fit_intercept=False))])
elm.fit(X_train, targets)
tr_elm = mean_squared_error(targets, predict(elm, X_train))
#print("training mse: ", tr_elm)
# save model
joblib.dump(elm, 'models/6_12_17/elm/te_pc/elm_XX_te_pc.pkl')
return (elm, tr_elm)
def train_ann(X_train, targets, n_h_neurons):
elm = pipeline.Pipeline([('rhl',
RandomLayer(n_hidden=n_h_neurons,
activation_func='multiquadric',
alpha=0.5)),
('lr', LinearRegression(fit_intercept=False))])
elm.fit(X_train, targets)
tr_mse = mean_squared_error(targets, predict(elm, X_train))
return (elm, tr_mse)
def train_ann(X_train, targets):
# training ELM
elm = pipeline.Pipeline([('rhl', RandomLayer(n_hidden=120, activation_func='multiquadric', alpha=0.7)),
('lr', LinearRegression(fit_intercept=False))])
elm.fit(X_train, targets)
tr_mse = mean_squared_error(targets, predict(elm, X_train))
#print("training mse: ", tr_mse)
# save model
joblib.dump(elm, 'models/6_12_17/elm/sc5/elm_XX_sc5_pc.pkl')
return (elm, tr_mse)
def train_ann(X_train, targets, f_name):
# training ELM
elm = pipeline.Pipeline([('rhl',
RandomLayer(n_hidden=240,
activation_func='multiquadric')),
('lr', LinearRegression(fit_intercept=False))])
elm.fit(X_train, targets)
tr_mse = mean_squared_error(targets, predict(elm, X_train))
print("training mse: ", tr_mse)
# save model
#joblib.dump(elm, f_name)
return (elm, tr_mse)
def objfun_elm(bw, x_train, y_train, n_hidden):
'''
优化所用的损失函数，返回误差值。
n_hidden 隐藏神经元数
wb为长度为2的横向量，即ELM中参数w和b的值
'''
bias = bw[0]
weights = bw[1]
rhl = RandomLayer(n_hidden=n_hidden, alpha=1.0) #表示_use_mlp_input=true
elmr = GenELMRegressor(hidden_layer=rhl)
elmr.fit(x_train, y_train, bias=bias, weights=weights)
score = elmr.score(x_train, y_train)
return score
def define_classification_model(h):
if config['model_type'] == 'linearSVM':
return LinearSVC(C=h)
elif config['model_type'] == 'ELM':
rl = RandomLayer(n_hidden=h, activation_func='reclinear', alpha=1)
return GenELMClassifier(hidden_layer=rl)
elif config['model_type'] == 'MLP':
return MLPClassifier(hidden_layer_sizes=(20, ),
max_iter=600,
verbose=10,
early_stopping=False)
elif config['model_type'] == 'linear':
return linear_model.SGDClassifier()
elif config['model_type'] == 'KNN':
return KNeighborsClassifier(n_neighbors=h)
def train_ann(X_train, targets):
elm = pipeline.Pipeline([('rhl',
RandomLayer(n_hidden=200,
activation_func='multiquadric',
alpha=0.69)),
('lr', LinearRegression(fit_intercept=False))])
#elmr = GenELMRegressor( hidden_layer = rl )
#elmr = ELMRegressor(n_hidden=98,random_state=0, alpha=0.8)
elm.fit(X_train, targets)
tr_mse = mean_squared_error(targets, predict(elm, X_train))
print("training mse: ", tr_mse)
# save model
joblib.dump(
elm, 'models/2018/23_06_18/elm/diel_diamond/elm_XX_diel_diamond.pkl')
return (elm, tr_mse)
mrx, mry = make_regression(n_samples=2000, n_targets=4)
mrx_train, mrx_test, mry_train, mry_test = train_test_split(mrx,
mry,
test_size=0.2)
xtoy, ytoy = make_toy()
xtoy, ytoy = stdsc.fit_transform(xtoy), stdsc.fit_transform(ytoy)
xtoy_train, xtoy_test, ytoy_train, ytoy_test = train_test_split(xtoy,
ytoy,
test_size=0.2)
plot(xtoy, ytoy)
# <codecell>
# RBFRandomLayer tests
for af in RandomLayer.activation_func_names():
print af,
elmc = ELMClassifier(activation_func=af)
tr, ts = res_dist(irx, iry, elmc, n_runs=200, random_state=0)
# <codecell>
elmc.classes_
# <codecell>
for af in RandomLayer.activation_func_names():
print af
elmc = ELMClassifier(activation_func=af, random_state=0)
tr, ts = res_dist(dgx, dgy, elmc, n_runs=100, random_state=0)
diabetes = load_diabetes()
dbx, dby = stdsc.fit_transform(diabetes.data), diabetes.target
dbx_train, dbx_test, dby_train, dby_test = train_test_split(dbx, dby, test_size=0.2)
mrx, mry = make_regression(n_samples=2000, n_targets=4)
mrx_train, mrx_test, mry_train, mry_test = train_test_split(mrx, mry, test_size=0.2)
xtoy, ytoy = make_toy()
xtoy, ytoy = stdsc.fit_transform(xtoy), stdsc.fit_transform(ytoy)
xtoy_train, xtoy_test, ytoy_train, ytoy_test = train_test_split(xtoy, ytoy, test_size=0.2)
plot(xtoy, ytoy)
# <codecell>
# RBFRandomLayer tests
for af in RandomLayer.activation_func_names():
print af,
elmc = ELMClassifier(activation_func=af)
tr,ts = res_dist(irx, iry, elmc, n_runs=200, random_state=0)
# <codecell>
elmc.classes_
# <codecell>
for af in RandomLayer.activation_func_names():
print af
elmc = ELMClassifier(activation_func=af, random_state=0)
tr,ts = res_dist(dgx, dgy, elmc, n_runs=100, random_state=0)
mrx, mry = make_regression(n_samples=2000, n_targets=4)
mrx_train, mrx_test, mry_train, mry_test = train_test_split(mrx,
mry,
test_size=0.2)
xtoy, ytoy = make_toy()
xtoy, ytoy = stdsc.fit_transform(xtoy), stdsc.fit_transform(ytoy)
xtoy_train, xtoy_test, ytoy_train, ytoy_test = train_test_split(xtoy,
ytoy,
test_size=0.2)
plot(xtoy, ytoy)
# <codecell>
# RBFRandomLayer tests
for af in RandomLayer.activation_func_names():
print af,
elmc = ELMClassifier(activation_func=af)
tr, ts = res_dist(irx, iry, elmc, n_runs=200, random_state=0)
# <codecell>
elmc.classes_
# <codecell>
for af in RandomLayer.activation_func_names():
print af
elmc = ELMClassifier(activation_func=af, random_state=0)
tr, ts = res_dist(dgx, dgy, elmc, n_runs=100, random_state=0)
dbx, dby, test_size=0.2)
mrx, mry = make_regression(n_samples=2000, n_targets=4)
mrx_train, mrx_test, mry_train, mry_test = train_test_split(
mrx, mry, test_size=0.2)
xtoy, ytoy = make_toy()
xtoy, ytoy = stdsc.fit_transform(xtoy), stdsc.fit_transform(ytoy)
xtoy_train, xtoy_test, ytoy_train, ytoy_test = train_test_split(
xtoy, ytoy, test_size=0.2)
plot(xtoy, ytoy)
# <codecell>
# RBFRandomLayer tests
for af in RandomLayer.activation_func_names():
print(af, end=' ')
elmc = ELMClassifier(activation_func=af)
tr, ts = res_dist(irx, iry, elmc, n_runs=200, random_state=0)
# <codecell>
elmc.classes_
# <codecell>
for af in RandomLayer.activation_func_names():
print(af)
elmc = ELMClassifier(activation_func=af, random_state=0)
tr, ts = res_dist(dgx, dgy, elmc, n_runs=100, random_state=0)
mrx, mry = make_regression(n_samples=2000, n_targets=4)
mrx_train, mrx_test, mry_train, mry_test = train_test_split(mrx,
mry,
test_size=0.2)
xtoy, ytoy = make_toy()
xtoy, ytoy = stdsc.fit_transform(xtoy), stdsc.fit_transform(ytoy)
xtoy_train, xtoy_test, ytoy_train, ytoy_test = train_test_split(xtoy,
ytoy,
test_size=0.2)
plot(xtoy, ytoy)
# <codecell>
# RBFRandomLayer tests
for af in RandomLayer.activation_func_names():
print(af, end=' ')
elmc = ELMClassifier(activation_func=af)
tr, ts = res_dist(irx, iry, elmc, n_runs=200, random_state=0)
# <codecell>
elmc.classes_
# <codecell>
for af in RandomLayer.activation_func_names():
print(af)
elmc = ELMClassifier(activation_func=af, random_state=0)
tr, ts = res_dist(dgx, dgy, elmc, n_runs=100, random_state=0)
plot(xtoy, ytoy)
# <codecell>
elmr = ELMRegressor(random_state=0, activation_func='gaussian', alpha=0.0)
elmr.fit(xtoy_train, ytoy_train)
print(elmr.score(xtoy_train, ytoy_train), elmr.score(xtoy_test, ytoy_test))
plot(xtoy, ytoy, xtoy, elmr.predict(xtoy))
# <codecell>
from sklearn import pipeline
from sklearn.linear_model import LinearRegression
elmr = pipeline.Pipeline([('rhl',
RandomLayer(random_state=0,
activation_func='multiquadric')),
('lr', LinearRegression(fit_intercept=False))])
elmr.fit(xtoy_train, ytoy_train)
print(elmr.score(xtoy_train, ytoy_train), elmr.score(xtoy_test, ytoy_test))
plot(xtoy, ytoy, xtoy, elmr.predict(xtoy))
# <codecell>
rhl = RandomLayer(n_hidden=200, alpha=1.0)
elmr = GenELMRegressor(hidden_layer=rhl)
tr, ts = res_dist(mrx, mry, elmr, n_runs=200, random_state=0)
scatter(tr, ts, alpha=0.1, marker='D', c='r')
# <codecell>
rhl = RBFRandomLayer(n_hidden=15, rbf_width=0.8)
