def trainsvr (self, train, trainlabel, seed, Cmin, Cmax, numC, rmin, rmax, numr, degree=3,\
verbose = 1, method = 'roc_auc', rad_stat =2):
C_range = np.logspace(Cmin, Cmax, num=numC, base=2, endpoint=True)
gamma_range = np.logspace(rmin, rmax, num=numr, base=2, endpoint=True)
scr = SVR(kernel=seed)
# mean_score=[]
df_C_gamma = pd.DataFrame({'gamma_range': gamma_range})
# df_this = DataFrame({'gamma_range':gamma_range})
count = 0
for C in C_range:
score_C = []
# score_C_this = []
count = count + 1
for gamma in gamma_range:
scr.C = C
scr.gamma = gamma
scr.degree = degree
scr.random_state = rad_stat
this_scores = cross_val_score(scr, train, trainlabel, scoring=method, cv=10, n_jobs=-1 \
)
score_C.append(np.mean(this_scores))
#score_C_this.append(np.mean(this_scores))
if verbose == 1:
print(np.mean(score_C))
print("%r cycle finished, %r left" % (count, numC - count))
df_C_gamma[C] = score_C
#df_this[C] = score_C_this
return df_C_gamma
def connectWidgets(self):
svr = SVR()
svr.kernel = 'rbf'
svr.degree = 3
svr.gamma = 'auto'
svr.coef0 = 0.0
svr.tol = 1e-3
svr.C = 1.0
svr.epsilon = 0.1
svr.shrinking = True
svr.cache_size = 200
svr.verbose = False
svr.max_iter = -1
self.cDoubleSpinBox.setValue(svr.C)
self.epsilonDoubleSpinBox.setValue(svr.epsilon)
self.defaultComboItem(self.kernelComboBox, svr.kernel)
self.degreeSpinBox.setValue(svr.degree)
self.defaultComboItem(self.gammaComboBox, svr.gamma)
self.coeff0DoubleSpinBox.setValue(svr.coef0)
self.shrinkingCheckBox.setChecked(svr.shrinking)
self.toleranceDoubleSpinBox.setValue(svr.tol)
self.cacheSizeSpinBox.setValue(svr.cache_size)
self.verboseCheckBox.setChecked(svr.verbose)
self.maxIterationsSpinBox.setValue(svr.max_iter)
def train_regress (self, train, trainlabel, seed, Cmin, Cmax, numC, rmin, rmax, numr, degree=3, method = 'rrmse', rad_stat =2):
C_range=np.logspace(Cmin, Cmax, num=numC, base=2,endpoint= True)
gamma_range=np.logspace(rmin, rmax, num=numr, base=2,endpoint= True)
svc = SVR(kernel=seed)
# mean_score=[]
df_C_gamma= DataFrame({'gamma_range':gamma_range})
# df_this = DataFrame({'gamma_range':gamma_range})
count = 0
for C in C_range:
score_C=[]
# score_C_this = []
count=count+1
for gamma in gamma_range:
svc.epsilon = 0.00001
svc.C = C
svc.gamma = gamma
svc.degree = degree
svc.random_state = rad_stat
this_scores = cross_val_score(svc, train, trainlabel, scoring=method, cv=10, n_jobs=-1 \
)
score_C.append(np.mean(this_scores))
#score_C_this.append(np.mean(this_scores))
print (np.mean(score_C) )
print ("%r cycle finished, %r left" %(count, numC-count))
df_C_gamma[C]= score_C
#df_this[C] = score_C_this
return df_C_gamma
def test_energy_model(X,
y,
epsilon=0.0841395,
C=0.122,
seed=None,
silent=False):
# best eps = 0.08413951416
# best C = 0.122
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.33,
random_state=seed)
svr = SVR()
svr.epsilon = epsilon
svr.C = C
svr.fit(X_train, y_train)
p_train = svr.predict(X_train)
p_test = svr.predict(X_test)
mse_train = np.mean((p_train - y_train)**2)
mse_test = np.mean((p_test - y_test)**2)
mean_abs_err_train = np.mean(np.abs(p_train - y_train))
mean_abs_err_test = np.mean(np.abs(p_test - y_test))
err_rel_train = np.mean(relative_err(p_train, y_train))
err_rel_test = np.mean(relative_err(p_test, y_test))
score_train = r2_score(y_train, p_train)
score_test = r2_score(y_test, p_test)
results = {
'mse_train': mse_train,
'mse_test': mse_test,
'err_rel_train': err_rel_train,
'err_rel_test': err_rel_test,
'mean_abs_err_train': mean_abs_err_train,
'mean_abs_err_test': mean_abs_err_test,
'score_train': score_train,
'score_test': score_test,
'y_train': y_train,
'p_train': p_train,
'y_test': y_test,
'p_test': p_test,
}
if not silent:
print(results)
return results
def test_energy_model_cv(X, y, epsilon=0.0841395, C=0.122, cv=5, silent=False):
svr = SVR()
svr.epsilon = epsilon
svr.C = C
cv_score = cross_val_score(svr, X, y, cv=cv)
y_pred = cross_val_predict(svr, X, y, cv=cv)
cv_mse = mean_squared_error(y, y_pred)
cv_r2 = r2_score(y, y_pred)
if not silent:
print('cv_score', cv_score)
print('cv_r2', cv_r2)
print('cv_mse', cv_mse)
return {'cv_score': cv_score, 'cv_r2': cv_r2, 'cv_mse': cv_mse}
def train_regress(self,
train,
trainlabel,
seed,
Cmin,
Cmax,
numC,
rmin,
rmax,
numr,
degree=3,
method='rrmse',
rad_stat=2):
C_range = np.logspace(Cmin, Cmax, num=numC, base=2, endpoint=True)
gamma_range = np.logspace(rmin, rmax, num=numr, base=2, endpoint=True)
svc = SVR(kernel=seed)
# mean_score=[]
df_C_gamma = DataFrame({'gamma_range': gamma_range})
# df_this = DataFrame({'gamma_range':gamma_range})
count = 0
for C in C_range:
score_C = []
# score_C_this = []
count = count + 1
for gamma in gamma_range:
svc.epsilon = 0.00001
svc.C = C
svc.gamma = gamma
svc.degree = degree
svc.random_state = rad_stat
this_scores = cross_val_score(svc, train, trainlabel, scoring=method, cv=10, n_jobs=-1 \
)
score_C.append(np.mean(this_scores))
#score_C_this.append(np.mean(this_scores))
print(np.mean(score_C))
print("%r cycle finished, %r left" % (count, numC - count))
df_C_gamma[C] = score_C
#df_this[C] = score_C_this
return df_C_gamma
def connectWidgets(self):
svr = SVR()
svr.kernel = 'rbf'
svr.degree = 3
svr.gamma = 'auto'
svr.coef0 = 0.0
svr.tol = 1e-3
svr.C = 1.0
svr.epsilon = 0.1
svr.shrinking = True
svr.cache_size = 200
svr.verbose = False
svr.max_iter = -1
self.cLineEdit.setText(str(svr.C))
self.epsilonLineEdit.setText(str(svr.epsilon))
self.kernel_list.setCurrentItem(self.kernel_list.findItems('Radial Basis Function', QtCore.Qt.MatchExactly)[0])
self.degreeLineEdit.setText(str(svr.degree))
self.coeff0LineEdit.setText(str(svr.coef0))
self.shrinking_list.setCurrentItem(self.shrinking_list.findItems(str(svr.shrinking), QtCore.Qt.MatchExactly)[0])
self.toleranceLineEdit.setText(str(svr.tol))
self.maxIterationsLineEdit.setText(str(svr.max_iter))
def main():
df_train = pd.read_csv("data/train.csv")
df_test = pd.read_csv("data/test.csv")
y = np.log(np.array(df_train["revenue"]))
plt.hist(df_train['P28'])
plt.show()
return
#test_id=df_test["Id"]
df = pd.concat([df_train, df_test])
df = processing(df)
df_train = df[0:137, :]
df_test = df[137:, :]
X = df_train
print X.shape, y.shape
X_test = df_test
model = SVR(kernel='linear')
#samplesubmit.head()
if (1):
selector = SelectPercentile(f_regression, percentile=100)
selector.fit(X, y)
print selector.pvalues_
scores = -np.log10(selector.pvalues_)
scores /= scores.max()
print scores
feature_index_0 = scores > 0.1
#model.C=1.e4
#model.gamma=0.2
X = X[:, feature_index_0]
#print X.shape
#model.fit(X,y)
#y_pred=model.predict(X)
#meanres=np.median(y-y_pred)
#stdres=np.median(np.abs(y-y_pred-meanres))
#print meanres,stdres
#outlierindex=np.abs(y-y_pred-meanres)>5.*stdres
#print len(y[outlierindex])
#plt.plot(y-y_pred,'.')
#plt.plot((y-y_pred)[outlierindex],'r.')
#plt.show()
#return
#X=X[~outlierindex,:]
#y=y[~outlierindex]
#if(1):
# selector=SelectPercentile(f_regression,percentile=100)
# selector.fit(X,y)
# print selector.pvalues_
# scores=-np.log10(selector.pvalues_)
# scores/=scores.max()
# print scores
# feature_index_1=scores>0.1
##return
if 0:
#gammas=np.logspace(-7,7,50)
#gammas=[0.1]
#C_s=np.array([1])
#return
#model=SVR()
#X=X[:,feature_index_1]
C_s = np.logspace(-7, 2, 50)
#C_s=np.logspace(0,5,50)
scores_mean = []
scores_std = []
best_score = 1.e20
best_C = 1
best_gamma = 1
for C in C_s:
if (1):
#for gamma in gammas:
model.C = C
#model.gamma=gamma
scores = cross_val_score(model,
X,
y,
cv=10,
scoring='mean_squared_error')
scores_mean.append(np.sqrt(-np.mean(scores)))
print np.sqrt(-np.mean(scores))
#if np.sqrt(-np.mean(scores))<best_score:
# best_C=C
# best_gamma=gamma
# best_score=np.sqrt(-np.mean(scores))
scores_std.append(np.std(scores))
#plt.semilogx(gammas,scores_mean,'.')
#plt.plot(X[:,0],y-y_pred,'.')
#plt.plot(X[:,1],y-y_pred,'.')
#plt.plot(X[:,2],y-y_pred,'.')
#plt.plot(X[:,3],y-y_pred,'.')
plt.semilogx(C_s, np.array(scores_mean) / 1.e6, '.')
plt.show()
#print best_C,best_gamma,best_score
return
#
#model=SVR()
#X=X[:,feature_index_1]
#X=X[:,feature_index_0]
#model.gamma=0.1
#model.C=1.5e3
#model=SVR()
#model.C=1.38949549437
model.C = 0.01
#model.gamma=0.193069772888
model.fit(X, y)
#y_pred=model.predict(X)
#plt.plot(y-y_pred,'.')
#plt.show()
#return
X_test = X_test[:, feature_index_0]
#X_test=X_test[:,feature_index_1]
y_pred = model.predict(X_test)
samplesubmit = pd.read_csv("data/sampleSubmission.csv")
samplesubmit["Prediction"] = np.exp(y_pred)
#samplesubmit.to_csv
#samplesubmit.to_csv("data/submit_fistsvr.csv",index=False)
#samplesubmit.to_csv("data/submit_linearsvr_fregression_age.csv",index=False)
samplesubmit.to_csv("data/submit_svr_logrevenue.csv", index=False)
#samplesubmit.to_csv("data/submit_linearsvr_fregression_outlierreject.csv",index=False)
return
plt.xlabel('Energies (after)')
#plt.show()
#Support Vector Regression
from sklearn.svm import SVR
from sklearn.metrics import mean_squared_error, mean_absolute_error
train_mse = []
val_mse = []
Cs = np.logspace(-3, 5, 20)
print(Cs)
for C in Cs:
model = SVR()
model.C = C
model.fit(designmatrix_train_std, Energies_train_std)
Energies_train_pred = Energies_mu + model.predict(
designmatrix_train_std) * Energies_std
Energies_val_pred = Energies_mu + model.predict(
designmatrix_val_std) * Energies_std
train_mse.append(
mean_squared_error(y_true=Energies_train, y_pred=Energies_train_pred))
val_mse.append(
mean_squared_error(y_true=Energies_val, y_pred=Energies_val_pred))
plt.figure()
plt.plot(Cs, train_mse, label='Training set MSE')
plt.plot(Cs, val_mse, label='Validation set MSE')
plt.ylabel('Mean Squared Error')
def main():
df_train = pd.read_csv("data/train.csv")
df_test = pd.read_csv("data/test.csv")
y = np.log(np.array(df_train["revenue"]))
plt.hist(df_train["P28"])
plt.show()
return
# test_id=df_test["Id"]
df = pd.concat([df_train, df_test])
df = processing(df)
df_train = df[0:137, :]
df_test = df[137:, :]
X = df_train
print X.shape, y.shape
X_test = df_test
model = SVR(kernel="linear")
# samplesubmit.head()
if 1:
selector = SelectPercentile(f_regression, percentile=100)
selector.fit(X, y)
print selector.pvalues_
scores = -np.log10(selector.pvalues_)
scores /= scores.max()
print scores
feature_index_0 = scores > 0.1
# model.C=1.e4
# model.gamma=0.2
X = X[:, feature_index_0]
# print X.shape
# model.fit(X,y)
# y_pred=model.predict(X)
# meanres=np.median(y-y_pred)
# stdres=np.median(np.abs(y-y_pred-meanres))
# print meanres,stdres
# outlierindex=np.abs(y-y_pred-meanres)>5.*stdres
# print len(y[outlierindex])
# plt.plot(y-y_pred,'.')
# plt.plot((y-y_pred)[outlierindex],'r.')
# plt.show()
# return
# X=X[~outlierindex,:]
# y=y[~outlierindex]
# if(1):
# selector=SelectPercentile(f_regression,percentile=100)
# selector.fit(X,y)
# print selector.pvalues_
# scores=-np.log10(selector.pvalues_)
# scores/=scores.max()
# print scores
# feature_index_1=scores>0.1
##return
if 0:
# gammas=np.logspace(-7,7,50)
# gammas=[0.1]
# C_s=np.array([1])
# return
# model=SVR()
# X=X[:,feature_index_1]
C_s = np.logspace(-7, 2, 50)
# C_s=np.logspace(0,5,50)
scores_mean = []
scores_std = []
best_score = 1.0e20
best_C = 1
best_gamma = 1
for C in C_s:
if 1:
# for gamma in gammas:
model.C = C
# model.gamma=gamma
scores = cross_val_score(model, X, y, cv=10, scoring="mean_squared_error")
scores_mean.append(np.sqrt(-np.mean(scores)))
print np.sqrt(-np.mean(scores))
# if np.sqrt(-np.mean(scores))<best_score:
# best_C=C
# best_gamma=gamma
# best_score=np.sqrt(-np.mean(scores))
scores_std.append(np.std(scores))
# plt.semilogx(gammas,scores_mean,'.')
# plt.plot(X[:,0],y-y_pred,'.')
# plt.plot(X[:,1],y-y_pred,'.')
# plt.plot(X[:,2],y-y_pred,'.')
# plt.plot(X[:,3],y-y_pred,'.')
plt.semilogx(C_s, np.array(scores_mean) / 1.0e6, ".")
plt.show()
# print best_C,best_gamma,best_score
return
#
# model=SVR()
# X=X[:,feature_index_1]
# X=X[:,feature_index_0]
# model.gamma=0.1
# model.C=1.5e3
# model=SVR()
# model.C=1.38949549437
model.C = 0.01
# model.gamma=0.193069772888
model.fit(X, y)
# y_pred=model.predict(X)
# plt.plot(y-y_pred,'.')
# plt.show()
# return
X_test = X_test[:, feature_index_0]
# X_test=X_test[:,feature_index_1]
y_pred = model.predict(X_test)
samplesubmit = pd.read_csv("data/sampleSubmission.csv")
samplesubmit["Prediction"] = np.exp(y_pred)
# samplesubmit.to_csv
# samplesubmit.to_csv("data/submit_fistsvr.csv",index=False)
# samplesubmit.to_csv("data/submit_linearsvr_fregression_age.csv",index=False)
samplesubmit.to_csv("data/submit_svr_logrevenue.csv", index=False)
# samplesubmit.to_csv("data/submit_linearsvr_fregression_outlierreject.csv",index=False)
return
