def PLSCrossValidation(n_components, trainSet, validationSet): pls = PLSRegression(n_components=n_components) pls.fit(trainSet[predictorList], trainSet['Apps']) predictPls = pls.predict(validationSet[predictorList]) different = predictPls.flat - validationSet['Apps'] error_rate = np.mean(different ** 2) return error_rate
def fit(self, predictors, predictands, locations, log=False, **kwargs): self.locations = locations self.models = [] self.n = predictors['n'] id = 0 for location in locations: X = extract_n_by_n(predictors, location, **kwargs) Y = predictands[:,id] if log: Y = np.log(Y) #pca = PCA(n_components='mle', whiten=True) model = PLSRegression(n_components=2) model = model.fit(X,Y) #components = pca.components_ #pca.components_ = components self.models.append(model) print "pls: ", location, model.score(X, Y), model.x_loadings_.shape, np.argmax(model.x_loadings_, axis=0) id += 1
def build_model(X, y): # gbr = GradientBoostingRegressor(learning_rate= 0.03, n_estimators=2000, max_depth=8, subsample=0.9) # rf = RandomForestRegressor(n_estimators=200) # lr = LinearRegression(fit_intercept=True) # knr = KNeighborsRegressor(n_neighbors=10, weights='uniform') # svr = SVR(C=5.0, kernel='linear') pls = PLSRegression(n_components=35) return pls.fit(X, y)
def Training(df,seed, yratio, xratio, index = 1): snp_matrix = np.array(df.values) xdim, ydim = snp_matrix.shape ydimlist = range(0,ydim) xdimlist = range(0,xdim) random.seed(seed) random.shuffle(ydimlist) # shuffle the individuals random.shuffle(xdimlist) # shuffle the SNPs accuracy = 0 snp_matrix_shuffle = np.copy(snp_matrix[:,ydimlist]) snp_matrix_shuffle = np.copy(snp_matrix[xdimlist,:]) snp_matrix_train = snp_matrix_shuffle[:,0:int(ydim*yratio)] snp_matrix_test = snp_matrix_shuffle[:,int(ydim*yratio):] snp_matrix_train_x = snp_matrix_train[0:int(xdim*xratio),:] snp_matrix_test_x = snp_matrix_test[0:int(xdim*xratio),:] for i in range(int(xdim*xratio), xdim): snp_matrix_train_y = snp_matrix_train[i,:] snp_matrix_test_y = snp_matrix_test[i,:] if index != 7: if index == 1: clf = AdaBoostClassifier(n_estimators= 100) elif index == 2: clf = RandomForestClassifier(n_estimators=100) elif index == 3: clf = linear_model.LogisticRegression(C=1e5) elif index == 4: clf = svm.SVC(kernel = 'rbf') elif index == 5: clf = svm.SVC(kernel = 'poly') else: clf = svm.SVC(kernel = 'linear') clf = clf.fit(snp_matrix_train_x.T, snp_matrix_train_y) Y_pred = clf.predict(snp_matrix_test_x.T) prediction = snp_matrix_test_y - Y_pred wrong = np.count_nonzero(prediction) tmp = 1 - (wrong + 0.0) / len(prediction) print tmp accuracy += tmp accuracy = accuracy / (xdim - int(xdim*xratio)) if index == 7: pls2 = PLSRegression(n_components = 50, scale=False, max_iter=1000) snp_matrix_train_y = snp_matrix_train[int(xdim*xratio):,:] pls2.fit(snp_matrix_train_x.T,snp_matrix_train_y.T) snp_matrix_test_x = snp_matrix_test[0:int(xdim*xratio),:] snp_matrix_test_y = snp_matrix_test[int(xdim*xratio):,:] Y_pred = transform(pls2.predict(snp_matrix_test_x.T)) prediction = snp_matrix_test_y - Y_pred.T xdim, ydim = prediction.shape wrong = np.count_nonzero(prediction) accuracy = 1 - wrong / (xdim * ydim + 0.0) return accuracy
def fit(predictors, predictands, log=False, **kwargs): model = PLSRegression(n_components=2) try: model.fit(predictors, predictands) except: return None return model
def get_correlations(param, spec, wave):
'''Returns correlations between spec and params by wavelengths'''
# using PLS
pls = PLSRegression(10)
pls.fit(spec, param)
#get corretalions
nparam = param.shape[1]
cor = pls.coefs*np.asarray([pls.x_std_]*nparam).T
cor /= np.tile(pls.y_std_, (cor.shape[0],1))
return cor
class PLSPredictor:
def __init__(self):
self.pls2 = PLSRegression(n_components=2,
scale=True,
max_iter=500,
tol=1e-06,
copy=True)
def predict(self, values):
self.pls2.predict(values)
def train(self, measured_values, screen_points):
self.pls2.fit(measured_values, screen_points)
def do_pls(X, Y):
pls2 = PLSRegression(n_components=2)
pls2.fit(X,Y)
out = pls2.transform(X)
print(out)
print(out.shape)
plt.title("PLS2")
plt.xlabel("PL1")
plt.ylabel("PL2")
plt.grid();
plt.scatter(out[:, 0], out[:, 1], c=Y, cmap='viridis')
plt.savefig('pls.png', dpi=125)
def pls_approach():
from sklearn.cross_decomposition import PLSRegression
(X, Y), cities = pull_xy_data()
pls = PLSRegression()
pls.fit(X, Y)
plsX, plsY = pls.transform(X, Y)
plot(plsX, cities, ["Lat01", "Lat02", "Lat03"], ellipse_sigma=1)
return "OK What Now?"
def __one_pls(self, cat):
np.seterr(all='raise')
lcat = np.zeros(self.train_set['labels'].size)
lcat[self.train_set['labels'] != cat] = -1
lcat[self.train_set['labels'] == cat] = +1
pls = PLSRegression(n_components=2, scale=False)
pls.fit(self.train_set['data'], lcat)
return pls
def fit_base_model(classifiers, fully, dummyY, trainx, testx):
""" Takes a list of classifiers and/or PLS regression and
does dimension reduction by returning the predictions of the classifiers
or first two scores of the PLS regression on bootstrapped subsamples of
the data."""
trainProbs = []
testProbs = []
iterations = 0
for clf in classifiers:
for i in range(clf[1]):
iterations += 1
print(iterations)
print(clf[0])
train_rows = np.random.choice(trainx.shape[0],
round(trainx.shape[0] * base_prop),
True)
oob_rows = list(set(range(trainx.shape[0])) - set(train_rows))
print(len(train_rows))
print(len(oob_rows))
x = trainx[train_rows, :]
if clf[0] == 'PLS':
y = dummyY[train_rows, :]
mod = PLSRegression().fit(x, y)
trainscores = mod.transform(trainx)
testscores = mod.transform(testx)
trainProbs.append(trainscores[:, 0])
trainProbs.append(trainscores[:, 1])
testProbs.append(testscores[:, 0])
testProbs.append(testscores[:, 1])
else:
y = fully[train_rows]
print('\t Fitting model...')
mod = clf[0].fit(x, y)
print('\t Predicting training results...')
tpreds = mod.predict_proba(trainx)
trainProbs.append(list(tpreds[:, 1]))
print('\t Predicting test results...')
testProbs.append(list(mod.predict_proba(testx)[:, 1]))
print('\t OOB score: ' + str(log_loss(fully[oob_rows],
tpreds[oob_rows, :])))
return trainProbs, testProbs
def pls_regr(x, y):
from sklearn.cross_decomposition import PLSRegression
n = len(x[0])
if n < 2:
raise TypeError
score = -999999999999
pls = None
'''
for i in range(3, n):
pls2 = PLSRegression(n_components=i)
pls2.fit(x,y)
cscore = pls2.score(x, y)
#print i, cscore
if cscore > score:
pls = pls2
score = cscore
'''
pls = PLSRegression(n_components=5)
pls.fit(x,y)
return pls
def train_PLSR(x_filename, y_filename, model_filename, n):
"""
Train a PLSR model and save it to the model_filename.
X and Y matrices are read from x_filename and y_filename.
The no. of PLSR components is given by n.
"""
X = loadMatrix(x_filename)[0].todense()
Y = loadMatrix(y_filename)[0].todense()
if X.shape[0] != Y.shape[0]:
sys.stderr.write("X and Y must have equal number of rows!\n")
raise ValueError
sys.stderr.write("Learning PLSR...")
startTime = time.time()
pls2 = PLSRegression(copy=True, max_iter=10000, n_components=n, scale=True, tol=1e-06)
pls2.fit(X, Y)
model = open(model_filename, 'w')
pickle.dump(pls2, model, 1)
model.close()
endTime = time.time()
sys.stderr.write(" took %ss\n" % str(round(endTime-startTime, 2)))
pass
def lex_function_learning( class_name, hyper_vec ) : #pls2 = KernelRidge( kernel = "rbf", gamma= 100) #pls2 = KernelRidge( ) pls2 = PLSRegression(n_components=50, max_iter=5000) X = extract_postive_features ( train_dataset[class_name][0], train_dataset[class_name][1] ) Y = [] for hypo_vec in X : sub = hyper_vec-hypo_vec Y.append(sub) # Target = difference vector ( Hypernym_vector - Hyponym_vector ) #Y.append(hyper_vec) # Target = Hypernym vector pls2.fit( X, Y) train_acc = pls2.score(X, Y) print "class = ", class_name, "train len = ", len(X) return pls2, train_acc, len(X)
def reduce_PLS(dataframe):
PLS_file="data/pls_structure.pickle"
selectedcolumn=[x for x in dataframe.columns if x not in ["id","click","device_id","device_ip"]]
X=np.array(dataframe[selectedcolumn])
y=np.array(dataframe["click"])
if os.path.exists(PLS_file):
stand_PLS=pickle.load(open(PLS_file,'rb'))
print "PLS structure is loaded."
else:
stand_PLS=PLSRegression(n_components=10,scale=True)
stand_PLS.fit(X, y[:,np.newaxis])
stand_PLS.y_scores_=None
stand_PLS.x_scores_=None
pickle.dump(stand_PLS,open(PLS_file,"wb"))
print "PLS transform structure is stored."
T=stand_PLS.transform(X)
print "PLS transformation is performed."
return T
sc_X = StandardScaler()
X_train = sc_X.fit_transform(X_train)
X_test = sc_X.transform(X_test)
X_atual = sc_X.transform(X_atual)
sc_y = StandardScaler()
y = y.reshape(-1, 1)
sc_y.fit(y)
y_train = y_train.reshape(-1, 1)
y_train = sc_y.transform(y_train)
y_test = y_test.reshape(-1, 1)
y_test = sc_y.transform(y_test)
# Regressão linear
from sklearn.cross_decomposition import PLSRegression
pls2 = PLSRegression(n_components=3, scale=False)
pls2.fit(X_train, y_train)
# Predicting the Test set results
y_pred = pls2.predict(X_test)
y_pred_atual = pls2.predict(X_atual)
# coeficientes
mean_squared_error(y_test, y_pred)
r2_score(y_test, y_pred)
df_atual = df[df['atletas.rodada_id'] == CURRENT_ROUND]
df_atual = df_atual[['atletas.apelido', 'atletas.clube.id.full.name', 'atletas.nome', 'atletas.posicao_id', \
'atletas.rodada_id', 'atletas.preco_num','atletas.pontos_num_sum_last5', 'atletas.media_num' ]]
df_atual['pred_score'] = y_pred_atual
df_atual.to_csv('predictions/predict-PLS.csv', encoding='utf-8')
# plt.show()
#Partial Least Squares Regression
from sklearn.cross_decomposition import PLSRegression
from sklearn.preprocessing import scale
X_train_scaled = scale(X_train)
X_test_scaled = scale(X_test)
#Performing Cross_Validation for PLS
mse = []
n= len(X_train_scaled)
kf_10 = cross_validation.KFold(n,n_folds=10, shuffle=True, random_state=0)
for i in np.arange(1,17):
plsregr = PLSRegression(n_components=i, scale=False)
plsregr.fit(X_train_scaled,y_train)
score = -1*cross_validation.cross_val_score(plsregr, X_train_scaled, y_train, cv=kf_10, scoring='mean_squared_error').mean()
mse.append(score)
plt.plot(np.arange(1,17), np.array(mse), '-v')
plt.title("PLS: MSE vs. Principal Components")
plt.xlabel('Number of principal components in PLS regression')
plt.ylabel('MSE')
plt.xlim((-0.2, 17.2))
#Based off of the plot, 12 principal components minimized MSE
plsregr_test = PLSRegression(n_components=12, scale=False)
plsregr_test.fit(X_train_scaled, y_train)
MSE_PLS = np.mean((plsregr_test.predict(X_test_scaled) - y_test) ** 2)
# print "Mean Squared Error: ", MSE_PLS
def bestpls(vipMatrix, X, Y, V):
###########################
#bestR2 = -10000
#lv_best = 1
#position = 1
###########################
bestR2 = vipMatrix[0][1]
lv_best = vipMatrix[0][3]
position = 0
###########################
#for i in range (len(vipMatrix)):
# print vipMatrix[i]
for entries in range (len(vipMatrix)):
#print vipMatrix[entries][1], "=?=", bestR2 #############
if vipMatrix[entries][1] > bestR2:
position = entries
bestR2 = vipMatrix[entries][1]
lv_best = vipMatrix[entries][3]
#################################################################################################qq
variables = []
for i in range (1, position): # not position + 1, as the vipMatrix[position] holds the next variable to be removed
variables.append(vipMatrix[i][0])
#print "VAR TO BE REMOVED: ", variables
V_new_Indices = []
for i in variables: # removed variable names in random order
V_new_Indices.append(V.index(i))
#if V == sorted(V):
# print "\nV ok!\n"
# keep names == separate
V_new = deepcopy(V)
for i in variables:
V_new.remove(i)
X_new = []
for i in range (len(X)):
X_new.append([])
variables_sent = [] ####
for i in range (len(X)):
for j in range (len(V)):
if j not in V_new_Indices:
#if V[j] not in variables_sent: ####
# variables_sent.append(V[j])####
X_new[i].append(X[i][j])
# epic test
if not V_new == sorted(V_new):
return base64.b64encode("tobulo"), [], [], 0
#else:
# print "v_new ok!"
#validity tests
#for i in range (len (variables_sent)):
# if variables_sent[i] == V_new[i]:
# print "ok", i
#print "var: ", len(V), "selected: ", len(V_new), "data (var) init length: ", len(X[0]), "data (var) now length: ", len(X_new[0])
"""
# PREVIOUS
variables = []
for i in range (1, position):
variables.append(vipMatrix[i][0])
V_new = deepcopy(V)
for i in variables:
V_new.remove(i) ################ remove by index??? CHECK!!!!
X_new = []
for i in range (len(X)):
X_new.append([])
for i in range (len(X)):
for j in range (len(V_new)): ####### HERE ALSO
X_new[i].append(X[i][j])
"""
#################################################################################################qq
#print V_new, "OOOO\n\n" #var names == cool
#print "\n\nNumber of variables ", len(V_new), " and latent: ", lv_best
#best_pls = PLSCanonical(n_components = lv_best)
best_pls = PLSRegression(n_components = lv_best)
best_pls.fit(X_new, Y)
saveas = pickle.dumps(best_pls)
encoded = base64.b64encode(saveas)
return encoded, X_new, V_new, lv_best
General Linear Model -- Elastic Net ''' clf = linear_model.ElasticNet(alpha=0.2, l1_ratio=0.01) clf.fit(x_scaled, y_scaled) print(clf.coef_) yvalid_scaled = clf.predict(xvalid_scaled) err1= MAPE(y, scalery.inverse_transform(clf.predict(x_scaled)).reshape(-1,1)) err = MAPE(yvalid, scalery.inverse_transform(yvalid_scaled).reshape(-1,1)) ''' General Linear Model -- Elastic Net ''' from sklearn.cross_decomposition import PLSRegression pls = PLSRegression(n_components=20) pls.fit(x_scaled, y_scaled) print(pls.coef_) yvalid_scaled = pls.predict(xvalid_scaled) err1= MAPE(y, scalery.inverse_transform(pls.predict(x_scaled)).reshape(-1,1)) err = MAPE(yvalid, scalery.inverse_transform(yvalid_scaled).reshape(-1,1)) from sklearn.decomposition import PCA reduced_data = PCA(n_components=2).fit_transform(xtrain_minmax) pca = PCA(n_components=2) pca.fit(xtrain_minmax) print(pca.explained_variance_ratio_)
def partial_least_squares(X_train, y_train, X_pred, store_settings, mod_params=None, metric=None): lr = PLSRegression(n_components=1, max_iter=1000, tol=1e-04) return lr.fit(X_train, y_train).predict(X_pred) return X_pred
def plot_pcr_vs_pls():
rng = np.random.RandomState(0)
n_samples = 500
cov = [[3, 3], [3, 4]]
X = rng.multivariate_normal(mean=[0, 0], cov=cov, size=n_samples)
pca = PCA(n_components=2).fit(X)
plt.scatter(X[:, 0], X[:, 1], alpha=.3, label='samples')
for i, (comp,
var) in enumerate(zip(pca.components_, pca.explained_variance_)):
comp = comp * var # scale component by its variance explanation power
plt.plot([0, comp[0]], [0, comp[1]],
label=f"Component {i}",
linewidth=5,
color=f"C{i + 2}")
plt.gca().set(aspect='equal',
title="2-dimensional dataset with principal components",
xlabel='first feature',
ylabel='second feature')
plt.legend()
plt.show()
y = X.dot(pca.components_[1]) + rng.normal(size=n_samples) / 2
fig, axes = plt.subplots(1, 2, figsize=(10, 3))
axes[0].scatter(X.dot(pca.components_[0]), y, alpha=.3)
axes[0].set(xlabel='Projected data onto first PCA component', ylabel='y')
axes[1].scatter(X.dot(pca.components_[1]), y, alpha=.3)
axes[1].set(xlabel='Projected data onto second PCA component', ylabel='y')
plt.tight_layout()
plt.show()
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=rng)
pcr = make_pipeline(StandardScaler(), PCA(n_components=1),
LinearRegression())
pcr.fit(X_train, y_train)
pca = pcr.named_steps['pca'] # retrieve the PCA step of the pipeline
pls = PLSRegression(n_components=1)
pls.fit(X_train, y_train)
fig, axes = plt.subplots(1, 2, figsize=(10, 3))
axes[0].scatter(pca.transform(X_test),
y_test,
alpha=.3,
label='ground truth')
axes[0].scatter(pca.transform(X_test),
pcr.predict(X_test),
alpha=.3,
label='predictions')
axes[0].set(xlabel='Projected data onto first PCA component',
ylabel='y',
title='PCR / PCA')
axes[0].legend()
axes[1].scatter(pls.transform(X_test),
y_test,
alpha=.3,
label='ground truth')
axes[1].scatter(pls.transform(X_test),
pls.predict(X_test),
alpha=.3,
label='predictions')
axes[1].set(xlabel='Projected data onto first PLS component',
ylabel='y',
title='PLS')
axes[1].legend()
plt.tight_layout()
plt.show()
print(f"PCR r-squared {pcr.score(X_test, y_test):.3f}")
print(f"PLS r-squared {pls.score(X_test, y_test):.3f}")
pca_2 = make_pipeline(PCA(n_components=2), LinearRegression())
pca_2.fit(X_train, y_train)
print(f"PCR r-squared with 2 components {pca_2.score(X_test, y_test):.3f}")
fold_number = min(fold_number, len(autoscaled_y_train))
autoscaled_y_train = pd.Series(autoscaled_y_train)
y_test = pd.Series(y_test)
autoscaled_x_train = pd.DataFrame(autoscaled_x_train)
autoscaled_x_test = pd.DataFrame(autoscaled_x_test)
plt.rcParams['font.size'] = 18 # 横軸や縦軸の名前の文字などのフォントのサイズ
for method in regression_methods:
print(method)
if method == 'pls': # Partial Least Squares
pls_components = np.arange(1, min(np.linalg.matrix_rank(autoscaled_x_train) + 1, max_pls_component_number + 1), 1)
r2all = list()
r2cvall = list()
for pls_component in pls_components:
pls_model_in_cv = PLSRegression(n_components=pls_component)
pls_model_in_cv.fit(autoscaled_x_train, autoscaled_y_train)
calculated_y_in_cv = np.ndarray.flatten(pls_model_in_cv.predict(autoscaled_x_train))
estimated_y_in_cv = np.ndarray.flatten(
model_selection.cross_val_predict(pls_model_in_cv, autoscaled_x_train, autoscaled_y_train, cv=fold_number))
"""
plt.figure(figsize=figure.figaspect(1))
plt.scatter( y, estimated_y_in_cv)
plt.xlabel("Actual Y")
plt.ylabel("Calculated Y")
plt.show()
"""
r2all.append(float(1 - sum((autoscaled_y_train - calculated_y_in_cv) ** 2) / sum(autoscaled_y_train ** 2)))
r2cvall.append(float(1 - sum((autoscaled_y_train - estimated_y_in_cv) ** 2) / sum(autoscaled_y_train ** 2)))
plt.plot(pls_components, r2all, 'bo-')
for j in range(len(random_split)):
test = random_split[j]
training_list = random_split[0:j] + random_split[j +
1:len(random_split)]
training = pd.concat(training_list)
X_train = training.drop(
training.columns[dsloader.RESPONSE_COLUMN_INDEX_NO_PCA], axis=1)
Y_train = training.iloc[:,
dsloader.RESPONSE_COLUMN_INDEX_NO_PCA].values
X_test = test.drop(test.columns[dsloader.RESPONSE_COLUMN_INDEX_NO_PCA],
axis=1)
Y_test = label_encoder.fit_transform(
test.iloc[:, dsloader.RESPONSE_COLUMN_INDEX_NO_PCA].apply(
transform_numeric_to_y))
model = PLSRegression(n_components=f_num_candidate)
model.fit(X_train, Y_train)
predictions = model.predict(X_test)
predictions = label_encoder.fit_transform(
np.array([
transform_numeric_to_y(prediction)
for prediction in predictions
]))
current_auc += roc_auc_score(Y_test, predictions)
current_auc /= len(random_split)
print("step {}: {} - {}".format(i + 1, f_num_candidate, current_auc))
results_df.iloc[0, i] = f_num_candidate
results_df.iloc[1, i] = current_auc
i += 1
class VM_Process2_시뮬레이터:
metric = 0
def __init__(self, A, d, C, F, p_lambda, p_VM, p_ACT, seed):
self.pls = PLSRegression(n_components=6,
scale=False,
max_iter=50000,
copy=True)
np.random.seed(seed)
self.A = A
self.d = d
self.C = C
self.F = F
self.p_lambda = p_lambda
self.p_VM = p_VM
self.p_ACT = p_ACT
self.real_ACT = [] # Process-1을 반영하는 실제 Actual 값
def sampling_up(self):
# u1 = np.random.normal(0.4, np.sqrt(0.2))
# u2 = np.random.normal(0.6, np.sqrt(0.2))
u1 = np.random.normal(0.2, np.sqrt(0.1))
u2 = np.random.normal(0.1, np.sqrt(0.05))
u = np.array([u1, u2])
return u
def sampling_vp(self):
v1 = np.random.normal(-0.4, np.sqrt(0.2))
v2 = 2 * v1
v3 = np.random.uniform(0.2, 0.6)
v4 = 3 * v3
v5 = np.random.uniform(0, 0.4)
v = np.array([v1, v2, v3, v4, v5])
return v
def sampling_ep(self):
e1 = np.random.normal(0, np.sqrt(0.05))
e2 = np.random.normal(0, np.sqrt(0.1))
e = np.array([e1, e2])
return e
def sampling(self,
k,
uk=np.array([0, 0]),
vp=np.array([0, 0, 0, 0, 0]),
ep=np.array([0, 0]),
p_VM=np.array([0, 0]),
p_ACT=np.array([0, 0]),
isInit=True):
u1 = uk[0]
u2 = uk[1]
u = uk
v1 = vp[0]
v2 = vp[1]
v3 = vp[2]
v4 = vp[3]
v5 = vp[4]
v = vp
e = ep
k1 = k % 150
k2 = k
eta_k = np.array([[k1], [k2]])
if isInit == True:
e = np.array([0, 0]) #DoE는 Sampling Actual이기 때문에 e가 없다.
fp = p_ACT # DoE에서는 Process-1 Act 값 사용
else:
fp = p_VM # VM에서는 Process-1 VM 값 사용
psi = np.array([u1, u2, v1, v2, v3, v4, v5, k1, k2])
if fp is not None: # Process-1의 입력값이 있다면
# 이건 왜 해놓은지 모르겠다.. R2R에서 결과를 도출하기 위해서인지, y를 Paremeter로 학습목적인지.. 추후 검증필요
if k % 10 == 0:
f = p_ACT
else:
f = p_VM
if isInit == True:
f = p_ACT
psi = np.r_[psi, f]
# VM이든 DoE든 계산한다.
y = u.dot(self.A) + v.dot(self.C) + np.sum(
eta_k * self.d, axis=0) + f.dot(self.F) + e
if isInit == False: #VM이 아닌 실제 ACT값을 별도로 계산한다.
#print('f.dot(self.F) : ', f.dot(self.F), 'p_ACT.dot(self.F) : ', p_ACT.dot(self.F))
temp = u.dot(self.A) + v.dot(self.C) + np.sum(
eta_k * self.d, axis=0) + p_ACT.dot(self.F) + e
self.real_ACT.append(np.array([temp[0], temp[1]]))
else: # Process-1의 입력값이 없고, 향후 Process-2 VM만 하고 싶을 때
y = u.dot(self.A) + v.dot(self.C) + np.sum(eta_k * self.d,
axis=0) + e
rows = np.r_[psi, y]
idx_end = len(rows)
idx_start = idx_end - 2
return idx_start, idx_end, rows
def pls_update(self, V, Y):
self.pls.fit(V, Y)
return self.pls
def setDoE_Mean(self, DoE_Mean):
self.DoE_Mean = DoE_Mean
def getDoE_Mean(self):
return self.DoE_Mean
def setPlsWindow(self, PlsWindow):
self.PlsWindow = PlsWindow
def getPlsWindow(self):
return self.PlsWindow
def DoE_Run(self, lamda_PLS, Z, M, f):
N = Z * M
DoE_Queue = []
for k in range(1, N + 1): # range(101) = [1, 2, ..., 120])
if f is not None:
fp = f[k - 1, 0:2]
else:
fp = None
idx_start, idx_end, result = self.sampling(k, self.sampling_up(),
self.sampling_vp(),
self.sampling_ep(),
None, fp, True)
DoE_Queue.append(result)
initplsWindow = DoE_Queue.copy()
npPlsWindow = np.array(initplsWindow)
plsWindow = []
# Process-1의 lamda_PLS는 이미 반영되어서 넘어오기 때문에, 중복되어 lamda_PLS를 반영할 필요가 없다.
for z in np.arange(0, Z):
if f is not None:
npPlsWindow[z * M:(z + 1) * M - 1, 0:idx_start -
2] = lamda_PLS * npPlsWindow[z * M:(z + 1) * M - 1,
0:idx_start - 2]
npPlsWindow[z * M:(z + 1) * M - 1, idx_start -
2:idx_start] = self.p_lambda * npPlsWindow[
z * M:(z + 1) * M - 1, idx_start - 2:idx_start]
npPlsWindow[z * M:(z + 1) * M - 1,
idx_start:idx_end] = lamda_PLS * (npPlsWindow[
z * M:(z + 1) * M - 1, idx_start:idx_end])
else:
npPlsWindow[z * M:(z + 1) * M - 1,
0:idx_start] = lamda_PLS * npPlsWindow[
z * M:(z + 1) * M - 1, 0:idx_start]
npPlsWindow[z * M:(z + 1) * M - 1,
idx_start:idx_end] = lamda_PLS * (npPlsWindow[
z * M:(z + 1) * M - 1, idx_start:idx_end])
for i in range(len(npPlsWindow)):
plsWindow.append(npPlsWindow[i])
npDoE_Queue = np.array(plsWindow)
DoE_Mean = np.mean(npDoE_Queue, axis=0)
plsModelData = npDoE_Queue - DoE_Mean
V0 = plsModelData[:, 0:idx_start]
Y0 = plsModelData[:, idx_start:idx_end]
pls = self.pls_update(V0, Y0)
y_prd = pls.predict(V0) + DoE_Mean[idx_start:idx_end]
y_act = npDoE_Queue[:, idx_start:idx_end]
#print("Init DoE VM Mean squared error: %.4f" % metrics.mean_squared_error(y_act[:,1:2], y_prd[:,1:2]))
#print("Init DoE VM r2 score: %.4f" % metrics.r2_score(y_act[:,1:2], y_prd[:,1:2]))
#print("pls : ", pls.coef_)
self.setDoE_Mean(DoE_Mean)
self.setPlsWindow(plsWindow)
# self.plt_show1(N, y_act[:,0:1], y_prd[:,0:1])
def VM_Run(self, lamda_PLS, Z, M):
N = Z * M
## V0, Y0 Mean Center
DoE_Mean = self.getDoE_Mean()
idx_end = len(DoE_Mean)
idx_start = idx_end - 2
meanVz = DoE_Mean[0:idx_start]
meanYz = DoE_Mean[idx_start:idx_end]
M_Queue = []
ez_Queue = []
mape_Queue = []
ez_Queue.append([0, 0])
y_act = []
y_prd = []
VM_Output = []
ACT_Output = []
plsWindow = self.getPlsWindow()
for z in np.arange(0, Z):
for k in np.arange(z * M + 1, ((z + 1) * M) + 1):
if self.p_VM[k - 1] is not None:
idx_start, idx_end, result = self.sampling(
k, self.sampling_up(), self.sampling_vp(),
self.sampling_ep(), self.p_VM[k - 1],
self.p_ACT[k - 1], False)
else:
idx_start, idx_end, result = self.sampling(
k, self.sampling_up(), self.sampling_vp(),
self.sampling_ep(), None, None, False)
psiK = result[0:idx_start]
psiKStar = psiK - meanVz
y_predK = self.pls.predict(psiKStar.reshape(
1, idx_start)) + meanYz
rows = np.r_[result, y_predK.reshape(2, )]
M_Queue.append(rows)
y_prd.append(rows[idx_end:idx_end + 2])
y_act.append(rows[idx_start:idx_end])
del plsWindow[0:M]
ez = M_Queue[M - 1][idx_start:idx_end] - M_Queue[
M - 1][idx_end:idx_end + 2]
ez_Queue.append(ez)
if z == 0:
ez = np.array([0, 0])
npVM_Queue = np.array(M_Queue)
npACT_Queue = np.array(M_Queue)
# for i in range(M): # VM_Output 구한다. lamda_pls 가중치를 반영하지 않는다.
# if i == M - 1:
# temp = npM_Queue[i:i + 1, idx_start:idx_end]
# else:
# temp = npM_Queue[i:i + 1, idx_end:idx_end + 2]
# VM_Output.append(np.array([temp[0, 0], temp[0, 1]]))
# Process-1의 lamda_PLS는 이미 반영되어서 넘어오기 때문에, 중복되어 lamda_PLS를 반영할 필요가 없다.
if self.p_VM[z - 1] is not None:
npVM_Queue[0:M - 1, 0:idx_start -
2] = lamda_PLS * npVM_Queue[0:M - 1,
0:idx_start - 2]
npVM_Queue[0:M - 1, idx_start -
2:idx_start] = self.p_lambda * npVM_Queue[
0:M - 1, idx_start - 2:idx_start]
npVM_Queue[0:M - 1, idx_start:idx_end] = lamda_PLS * (
npVM_Queue[0:M - 1, idx_end:idx_end + 2] + 0.5 * ez
) # + 0.5 * ez
npVM_Queue = npVM_Queue[:, 0:idx_end]
npACT_Queue[0:M - 1, 0:idx_start -
2] = lamda_PLS * npACT_Queue[0:M - 1,
0:idx_start - 2]
npACT_Queue[0:M - 1, idx_start -
2:idx_start] = lamda_PLS * npACT_Queue[0:M - 1,
idx_start -
2:idx_start]
npACT_Queue[0:M - 1,
idx_start:idx_end] = lamda_PLS * npACT_Queue[
0:M - 1, idx_start:idx_end]
npACT_Queue = npACT_Queue[:, 0:
idx_end] ##idx_start ~ end 까지 VM 값 정리
else:
npVM_Queue[0:M - 1,
0:idx_start] = lamda_PLS * npVM_Queue[0:M - 1,
0:idx_start]
npVM_Queue[0:M - 1, idx_start:idx_end] = lamda_PLS * (
npVM_Queue[0:M - 1, idx_end:idx_end + 2] + 0.5 * ez
) # + 0.5 * ez
npVM_Queue = npVM_Queue[:, 0:idx_end]
npACT_Queue[0:M - 1,
0:idx_start] = lamda_PLS * npACT_Queue[0:M - 1,
0:idx_start]
npACT_Queue[0:M - 1,
idx_start:idx_end] = lamda_PLS * npACT_Queue[
0:M - 1, idx_start:idx_end]
npACT_Queue = npACT_Queue[:, 0:
idx_end] ##idx_start ~ end 까지 VM 값 정리
for i in range(
M): #VM_Output 구한다. lamda_pls 가중치를 반영하여 다음 계산시 편리하게 한다.
if i == M - 1:
temp = npACT_Queue[i:i + 1, idx_start:idx_end]
else:
temp = npVM_Queue[i:i + 1, idx_start:idx_end]
VM_Output.append(np.array([temp[0, 0], temp[0, 1]]))
temp = npACT_Queue[i:i + 1, idx_start:idx_end]
ACT_Output.append(np.array([temp[0, 0], temp[0, 1]]))
for i in range(M):
plsWindow.append(npVM_Queue[i])
M_Mean = np.mean(plsWindow, axis=0)
meanVz = M_Mean[0:idx_start]
meanYz = M_Mean[idx_start:idx_end]
plsModelData = plsWindow - M_Mean
V = plsModelData[:, 0:idx_start]
Y = plsModelData[:, idx_start:idx_end]
self.pls_update(V, Y)
del M_Queue[0:M]
#y_act = np.array(y_act)
y_act = np.array(self.real_ACT)
y_prd = np.array(y_prd)
ez_all_run = y_act - y_prd
self.metric = metrics.explained_variance_score(y_act[:, 1:2],
y_prd[:, 1:2])
print("VM Mean squared error: %.4f" %
metrics.mean_squared_error(y_act[:, 1:2], y_prd[:, 1:2]))
print("explained_variance_score: %.4f" % self.metric)
print("VM r2 score: %.4f" %
metrics.r2_score(y_act[:, 1:2], y_prd[:, 1:2]))
#print("pls : ", self.pls.coef_)
ez_run = np.array(ez_Queue)
VM_Output = np.array(VM_Output)
ACT_Output = np.array(ACT_Output)
return VM_Output, ACT_Output, ez_run, y_act, y_prd, ez_all_run
import time
import pandas as pd
import matplotlib.pyplot as plt
import sys
n = sys.argv[1]
start = time.time()
with open(f"{n}.pickle", "rb") as f:
datas = pickle.load(f)
models = {}
models["LASSO"] = LassoCV(max_iter=10000, cv=5, n_jobs=-1)
models["RIDGE"] = RidgeCV(cv=5)
models["EN"] = ElasticNetCV(max_iter=10000, cv=5, n_jobs=-1)
models["PLS20"] = PLSRegression(n_components=20, scale=False)
results = {}
for key in models.keys():
model = models[key]
results[key] = {"metrics": {}, "data": {}}
for i in range(len(datas)):
data = datas[i]
q2 = cross_val_score(model,
data['train_X'],
data['train_Y'],
cv=5,
scoring='r2').mean()
model.fit(data["train_X"], data["train_Y"])
predict_Y = model.predict(data["test_X"])
def pls(self, x, y, param_info):
pls = PLSRegression(n_components=param_info.pls_compnum)
#pls = PLSRegression(n_components=param_info.pls_compnum, max_iter=1000000)
#pls = PLSSVD(n_components=param_info.pls_compnum)
self.learned_pls = pls.fit(x, y)
# scale all samples according to training set
scaler = preprocessing.MinMaxScaler().fit(train_hydrogens)
train_hydrogens_normalized = scaler.transform(train_hydrogens)
test_hydrogens_normalized = scaler.transform(test_hydrogens)
# one hot encode training labels for plsda
train_labels_one_hot = []
for i in np.ravel(train_labels):
if i == 0:
train_labels_one_hot.append([1, 0])
else:
train_labels_one_hot.append([0, 1])
train_labels_one_hot = np.array(train_labels_one_hot)
plsda = PLSRegression(n_components=30, scale=False)
plsda.fit(train_hydrogens_normalized, train_labels_one_hot)
test_pred_ = plsda.predict(test_hydrogens_normalized)
test_pred = np.array([np.argmax(x) for x in test_pred_]).reshape(-1, 1)
cm = confusion_matrix(test_labels, test_pred)
auroc = roc_auc_score(test_labels, test_pred_[:, 1])
auroc_folds.append(auroc)
precision, recall, thresh = precision_recall_curve(test_labels,
test_pred_[:, 1])
aupr = auc(recall, precision)
aupr_folds.append(aupr)
def stacklearning(self):
class sparseNorm(BaseEstimator, TransformerMixin):
def __init__(self):
pass
def fit(self, X, y=None):
return self
def transform(self, X):
from sklearn import preprocessing
Y = preprocessing.normalize(sp.sparse.csc_matrix(X.values))
return Y
fm = sgd.FMRegression(
n_iter=4743,
init_stdev=0.1,
rank=100,
l2_reg_w=0,
l2_reg_V=0,
step_size=0.1,
)
fm = sgd.FMRegression(
n_iter=9943,
init_stdev=0.1,
rank=219,
l2_reg_w=0,
l2_reg_V=0.06454,
step_size=0.1,
)
pipe = make_pipeline(sparseNorm(), fm)
calcACC(pipe, X=X2)
xgb = xgboost.XGBRegressor(
n_estimators=100,
max_depth=7,
gamma=0,
colsample_bytree=0.1
)
lgbm = LGBMRegressor(
boosting_type='gbdt', num_leaves=367,
learning_rate=0.06,feature_fraction=0.14,
max_depth=28, min_data_in_leaf=8
)
rgf = RGFRegressor(
max_leaf=1211, algorithm="RGF", test_interval=100,
loss="LS", verbose=False, l2=0.93,
min_samples_leaf=2
)
rf = RandomForestRegressor(
max_depth=20, random_state=0,
n_estimators=56,min_samples_split=2,
max_features=0.21
)
rf = RandomForestRegressor()
ext = ExtraTreesRegressor(
n_estimators=384,max_features= 2228,
min_samples_split= 0.01,max_depth= 856,
min_samples_leaf= 1
)
svr = SVR(
gamma=9.5367431640625e-07,
epsilon=0.0009765625,
C= 2048.0
)
#test combination
desNew = make_pipeline(extdescriptorNew(),rf)
morNew = make_pipeline(extMorganNew(),rf)
kotNew = make_pipeline(extklekotaTothNew(),rf)
macNew = make_pipeline(extMACCSNew(),rf)
desMac = make_pipeline(extDescriptorMACCS(),rf)
morMac = make_pipeline(extMorganMACCS(),rf)
kotMac = make_pipeline(extKlekotaTothMACCS(),rf)
morKotNew = make_pipeline(extMorganKlekotaTothNew(),rf)
des = make_pipeline(extOnlyDescriptor(),rf)
mor = make_pipeline(extOnlyMorgan(),rf)
kot = make_pipeline(extOnlyklekotaToth(),rf)
mac = make_pipeline(extOnlyMACCS(),rf)
all = make_pipeline(extAll(),rf)
allwithoutNew = make_pipeline(extAllwithoutNew(),rf)
allwithoutMaccs = make_pipeline(extAllwithoutMaccs(),rf)
allwithoutDes = make_pipeline(extAllwithoutDescriptor(),rf)
testDic = {"Desc+New":desNew,"Mor+New":morNew,"kot+New":kotNew,"MACCS+New":macNew,"Des+MAC":desMac,"Morgan+Maccs":morMac,"Kot+MACCS":kotMac,"mor+kot+New":morKotNew,
"descriptor":des,"morgan":mor,"kot":kot,"MACCS":mac,"All":all,"All without "
"new":allwithoutNew,
"All without MACCS":allwithoutMaccs,"All without Des":allwithoutDes}
#10fold
cv = KFold(n_splits=10, shuffle=True, random_state=0)
#Fingerprinttest
resultDic={}
resultDic2={}
for name,model in testDic.items():
#model = StackingRegressor(regressors=[name], meta_regressor=rf,verbose=1)
#calcACC(model,X=X,y=y2,name=name)
Scores = cross_validate(model, X2, y2, cv=cv,scoring=myScoreFunc)
RMSETmp = Scores['test_RMSE'].mean()
CORRTmP = Scores['test_Correlation coefficient'].mean()
resultDic.update({name:[RMSETmp,CORRTmP]})
print(name,RMSETmp,CORRTmP)
#stacking
alldata = make_pipeline(extAll())
# random forest
#1.1546 0.70905
stack = StackingRegressor(regressors=[alldata], meta_regressor=rf,verbose=1)
# Light Gradient boosting
# 1.160732 0.703776
testmodel = StackingRegressor(regressors=[alldata], meta_regressor=lgbm,verbose=1)
# XGboost
# 1.1839805 0.689571
testmodel = StackingRegressor(regressors=[alldata], meta_regressor=xgb,verbose=1)
# Regularized greedily forest
# 1.17050 0.6992
testmodel = StackingRegressor(regressors=[alldata], meta_regressor=rgf,verbose=1)
#pls 22.808047774809697 0.6410026452910016 i=4
for i in np.arange(3,11,1):
pls = PLSRegression(n_components=i)
testmodel = StackingRegressor(regressors=[alldata], meta_regressor=pls,verbose=0)
calcACC(testmodel)
pls = PLSRegression(n_components=4)
#SVR
svr = SVR(gamma=9.5367431640625/10000000,C=1559.4918100725592,
epsilon=0.0009765625,)
svr = SVR(kernel='rbf',gamma=9.5367431640625e-07,epsilon=0.0009765625,C=2048.0)
testmodel = StackingRegressor(regressors=[alldata], meta_regressor=svr, verbose=1)
calcACC(svr)
#Extratree 1.157420824123527 0.7061010221224269
testmodel = StackingRegressor(regressors=[alldata], meta_regressor=ext, verbose=1)
calcACC(testmodel)
#k-NN
nbrs = KNeighborsRegressor(3)
##Linear regressions
#Stochastic Gradient Descenta
sgd = SGDRegressor(max_iter=1000)
# Ridge
for i in [1,10,100,1000]:
ridge = Ridge(alpha=i)
calcACC(ridge)
ridge = Ridge(alpha=45.50940042350705)
calcACC(ridge)
# multiple linear
lin = make_pipeline(forlinear(),LinearRegression(n_jobs=-1))
calcACC(lin)
#stacking
#0.69
testmodel = StackingRegressor(regressors=[alldata,nbrs,all], meta_regressor=rf,verbose=1)
#1.1532 0.70926
testmodel = StackingRegressor(regressors=[alldata,nbrs,all,xgb,lgbm,rgf], meta_regressor=rf,
verbose=1)
#1.16420 0.7041
testmodel = StackingRegressor(regressors=[alldata,alldata,all], meta_regressor=rf,verbose=1)
#1.16379 0.7044
stack1 = StackingRegressor(regressors=[alldata,nbrs,all,xgb,lgbm,rgf], meta_regressor=rf,verbose=1)
testmodel = StackingRegressor(regressors=[alldata,stack1,stack1], meta_regressor=rf,verbose=1)
#1.1535496740699531 0.7108839199109559
pcaFeature = make_pipeline(extPCA())
testmodel = StackingRegressor(regressors=[pcaFeature,alldata,nbrs,rf,xgb,lgbm,rgf]
,meta_regressor=rf,verbose=1)
#1.181801005432221 0.6889745579620922
testmodel = StackingRegressor(regressors=[pcaFeature,alldata,nbrs,rf,xgb,lgbm,rgf]
,meta_regressor=lgbm,verbose=1)
#0.70613
testmodel = StackingRegressor(regressors=[pcaFeature,alldata,nbrs,rf,xgb,lgbm,rgf,ext]
,meta_regressor=xgb,verbose=1)
#0.71641717
testmodel = StackingRegressor(regressors=[pcaFeature,alldata,nbrs,rf,xgb,lgbm,rgf,ext]
,meta_regressor=rf,verbose=1)
#0.7146922
testmodel = StackingRegressor(regressors=[pcaFeature,alldata,nbrs,ridge,rf,xgb,lgbm,rgf,ext]
,meta_regressor=rf,verbose=1)
#new features
pcaFeature = make_pipeline(extPCA())
#old
pipe1 = make_pipeline(extMACCS(), rf)
pipe2 = make_pipeline(extMorgan(), rf)
pipe3 = make_pipeline(extDescriptor(), rf)
pipe4 = make_pipeline(extPCA(), rgf)
pipe7 =make_pipeline(extDescriptor(), rgf)
pipe8 =make_pipeline(extDescriptor(), rgf)
xgb = xgboost.XGBRegressor()
nbrs = KNeighborsRegressor(2)
svr = SVR(gamma='auto',kernel='linear')
pls = PLSRegression(n_components=4)
extMACCSdata = make_pipeline(extMACCS())
nbrsPipe = make_pipeline(extMorgan(), nbrs)
pipe6 = make_pipeline(extMACCS(), rgf)
alldata = make_pipeline(extAll())
ave = extAverage()
withoutdesc = make_pipeline(extMACCS())
meta = RandomForestRegressor(max_depth=20, random_state=0, n_estimators=400)
#stack1 = StackingRegressor(regressors=[rgf, nbrs, alldata], meta_regressor=rgf, verbose=1)
#0.70
stack = StackingRegressor(regressors=[pipe1,pipe2,pipe3,xgb,lgbm,rgf,rf], meta_regressor=ave, verbose=1)
#stack2 = StackingRegressor(regressors=[stack1,nbrs, svr,pls,rgf], meta_regressor=lgbm, verbose=1)
#0.69######################
stack1 = StackingRegressor(regressors=[pipe1,pipe2,pipe3], meta_regressor=rf, verbose=1)
#0.70
stack2 = StackingRegressor(regressors=[stack1,alldata,rgf,lgbm,xgb], meta_regressor=rf,verbose=1)
#0.71
stack3 = StackingRegressor(regressors=[stack2,pipe1], meta_regressor=ave, verbose=1)
###########################
###########################
stack1 = StackingRegressor(regressors=[pipe1,pipe2,pipe3], meta_regressor=rf, verbose=1)
stack2 = StackingRegressor(regressors=[stack1,withoutdesc,lgbm,rgf], meta_regressor=rf,verbose=1)
stack3 = StackingRegressor(regressors=[stack2,pipe1,xgb], meta_regressor=ave, verbose=1)
###########################
#stackingwithknn
stack1 = StackingRegressor(regressors=[pipe1,pipe2,pipe3], meta_regressor=rf, verbose=1)
stack2 = StackingRegressor(regressors=[stack1,nbrs,pipe1], meta_regressor=rf, verbose=1)
#stack3 = StackingRegressor(regressors=[rgf, nbrs, alldata], meta_regressor=ave, verbose=1)
cv = ShuffleSplit(n_splits=10, test_size=0.1, random_state=0)
cv = KFold(n_splits=10, shuffle=True, random_state=0)
St1Scores = cross_validate(stack1,X,y,cv=cv)
St1Scores['test_score'].mean()**(1/2)
St2Scores = cross_validate(stack2,X,y,cv=cv)
St2Scores['test_score'].mean()**(1/2)
St3Scores = cross_validate(stack3,X,y,cv=cv)
St3Scores['test_score'].mean()**(1/2)
stackScore = cross_validate(stack, X, y, cv=cv)
stackScore['test_score'].mean()**(1/2)
lgbmScores =cross_validate(lgbm,X,y,cv=cv)
lgbmScores['test_score'].mean()**(1/2)
rgfScores = cross_validate(rgf,X,y,cv=cv)
rgfScores['test_score'].mean()**(1/2)
RFScores = cross_validate(rf,X,y,cv=cv)
RFScores['test_score'].mean()**(1/2)
scores = cross_validate(stack2,X,y,cv=cv)
scores['test_score'].mean()**(1/2)
print("R^2 Score: %0.2f (+/- %0.2f) [%s]" % (scores['test_score'].mean(), scores['test_score'].std(), 'stacking'))
stack3.fit(X, y)
y_pred = stack3.predict(X_train)
y_val = stack3.predict(X_test)
#stack3.score(X_train, y_train)
exX = preprocess(extractDf, changeList)
valy = (10 **(stack3.predict(exX))).tolist()
print("Root Mean Squared Error train: %.4f" % calcRMSE(y_pred, y_train))
print("Root Mean Squared Error test: %.4f" % calcRMSE(y_val, y_test))
print('Correlation Coefficient train: %.4f' % calcCorr(y_pred, y_train))
print('Correlation Coefficient test: %.4f' % calcCorr(y_val, y_test))
stack1.fit(X, y)
valy = (10 **(stack1.predict(exX))).tolist()
sgd.fit(X,y)
valy = (10 **(sgd.predict(exX))).tolist()
rgfpipe = make_pipeline(extMACCS(), rf)
rgf.fit(X,y)
valy = (10 **(rgf.predict(exX))).tolist()
nbrs.fit(X,y)
valy = (10 **(nbrs.predict(exX))).tolist()
pipe = make_pipeline(extMACCS(), rf)
pipe.fit(X,y)
valy = (10 **(pipe.predict(exX))).tolist()
rf.fit(X, y)
y_pred = rf.predict(X_train)
y_val = rf.predict(X_test)
exX = preprocess(extractDf, changeList)
valy = (10 **(rf.predict(exX))).tolist()
print("Root Mean Squared Error train: %.4f" % calcRMSE(y_pred, y_train))
print("Root Mean Squared Error test: %.4f" % calcRMSE(y_val, y_test))
print('Correlation Coefficient train: %.4f' % calcCorr(y_pred, y_train))
print('Correlation Coefficient test: %.4f' % calcCorr(y_val, y_test))
lgbm.fit(X, y)
#y_pred = pipe1.predict(X_train)
#y_val = pipe1.predict(X_test)
exX = preprocess(extractDf, changeList)
valy = (10 **(lgbm.predict(exX))).tolist()
print("Root Mean Squared Error train: %.4f" % calcRMSE(y_pred, y_train))
print("Root Mean Squared Error test: %.4f" % calcRMSE(y_val, y_test))
print('Correlation Coefficient train: %.4f' % calcCorr(y_pred, y_train))
print('Correlation Coefficient test: %.4f' % calcCorr(y_val, y_test))
def pls_thing(scenario_data, xcols, ycols, titlestr):
#PLS Summary Stats
pls = PLSRegression(n_components=3)
pls.fit(scenario_data[xcols], scenario_data[ycols])
k = 0
transformed_x_full = pls.transform(scenario_data[xcols])
y = scenario_data[ycols]
results = pd.DataFrame(columns=('Case Label', 'Explained Variance Ratio',
'RegressionCoefs', 'Regression R^2',
'SpearmanCorr', 'SpearmanPvalue',
'Loadings', 'X Weights', 'X Loadings',
'X Scores'))
if type(titlestr) == type([]):
titlestr = ' '.join(titlestr)
#Linear fits for each individual component
for c in range(np.shape(pls.x_weights_)[1]):
x_transformed_1pc = transformed_x_full[:, k].reshape(-1, 1)
lr = linear_model.LinearRegression(fit_intercept=True, normalize=True)
lr.fit(x_transformed_1pc, y)
print('Regression Coefs', lr.coef_)
print('R^2', lr.score(x_transformed_1pc, y))
print('Spearman: ', scipy.stats.spearmanr(x_transformed_1pc, y))
print('Component: ', c)
results.loc[len(results)] = np.nan
results.loc[len(results) - 1,
'Case Label'] = titlestr + ' Component ' + str(k)
# results.loc[len(results)-1,'Explained Variance Ratio'] = pls.explained_variance_ratio_[k]
results.set_value(len(results) - 1, 'RegressionCoefs', lr.coef_)
results.loc[len(results) - 1,
'Regression R^2'] = lr.score(x_transformed_1pc, y)
results.loc[len(results) - 1, 'SpearmanCorr'] = scipy.stats.spearmanr(
x_transformed_1pc, y)[0]
results.loc[len(results) - 1,
'SpearmanPvalue'] = scipy.stats.spearmanr(
x_transformed_1pc, y)[1]
results.set_value(len(results) - 1, 'X Weights', pls.x_weights_[:, k])
results.set_value(
len(results) - 1, 'X Loadings', pls.x_loadings_[:, k])
results.set_value(len(results) - 1, 'X Scores', pls.x_scores_[:, k])
plt.plot(x_transformed_1pc, y, '*')
plt.xlabel('Component ' + str(k))
plt.ylabel('Performance')
plt.title('PLS ' + titlestr)
plt.show()
k += 1
print(results)
fig = plt.figure()
ax = fig.add_subplot(111)
ax.set_title("PLS PC0 vs PC1 vs Performance " + ' '.join(cs), fontsize=14)
ax.set_xlabel("PC0", fontsize=12)
ax.set_ylabel("PC1", fontsize=12)
ax.scatter(transformed_x_full[:, 0],
transformed_x_full[:, 1],
s=100,
c=y,
marker='*',
cmap=cm.bwr)
plt.show()
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(transformed_x_full[:, 0],
transformed_x_full[:, 1],
transformed_x_full[:, 2],
s=100,
c=y,
marker='*',
cmap=cm.bwr)
ax.set_title("PLS PC0 vs PC1 vs PC2 vs Performance " + ' '.join(cs),
fontsize=14)
ax.set_xlabel("PC0", fontsize=12)
ax.set_ylabel("PC1", fontsize=12)
ax.set_zlabel("PC2", fontsize=12)
plt.show()
print(results)
return results
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for c, i, target_name in zip("rb", target_names, target_names):
ax.scatter(X_r[y == i, 0], X_r[y == i, 1], X_r[y == i, 2], c=c)
ax.set_xlabel('X Label')
ax.set_ylabel('Y Label')
ax.set_zlabel('Z Label')
plt.axis('equal')
ax.set_xlim([-1000,4000])
ax.set_ylim([-1000,4000])
ax.set_zlim([-1000,4000])
plt.show()
# part b
PLS1 = PLS(n_components=3)
number_map = {"M": 0,"B": 1}
numeric_y = np.array(map(lambda x : number_map[x],y))
result = PLS1.fit_transform(x,numeric_y)
X_r = result[0]
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for c, i, target_name in zip("rb", target_names, target_names):
ax.scatter(X_r[y == i, 0], X_r[y == i, 1], X_r[y == i, 2], c=c)
ax.set_xlabel('X Label')
ax.set_ylabel('Y Label')
ax.set_zlabel('Z Label')
plt.axis('equal')
plt.show()
def make_plots(m,data,colors,names,groundtruth=None,waves=None,sample_size=10,ux=0,
remove_mean=False,log_x=False,ylim=(0.3,1),res_out='',title=None):
inds_sup_train = np.random.choice(data['X'].shape[0],size=sample_size)
inds_sup_valid = np.random.choice(data['X_valid'].shape[0],size=sample_size)
inds_train_x = np.random.choice(data['X_'].shape[0],size=sample_size)
inds_train_y = np.random.choice(data['_y'].shape[0],size=sample_size)
y = np.hstack([data['y'],1-data['y'].sum(axis=1,keepdims=True)])
y_valid = np.hstack([data['y_valid'],1-data['y_valid'].sum(axis=1,keepdims=True)])
y_corners = np.vstack((np.eye(data['y'].shape[1]),np.zeros(data['y'].shape[1]))).astype('float32')
simplex = []
for point in product(*([np.linspace(0,1,50)]*y.shape[1])):
if np.sum(point) == 1:
simplex += [point]
simplex = np.asarray(simplex).astype('float32')
simplex = simplex[:,:-1]
if waves is None:
waves = np.arange(data['X'].shape[1])
if remove_mean:
_ux = ux
else:
_ux = 0
if log_x:
f = lambda x: np.exp(x)
else:
f = lambda x: x
if ylim is not None:
force_ylim = True
pls_XY = PLSRegression(n_components=8,scale=False)
pls_XY.fit(data['X'],y)
pred_train_pls = pls_XY.predict(data['X'])
pred_train_pls = (pred_train_pls.T/np.sum(pred_train_pls,axis=1)).T
pred_valid_pls = pls_XY.predict(data['X_valid'])
pred_valid_pls = (pred_valid_pls.T/np.sum(pred_valid_pls,axis=1)).T
score_pred_train_pls = KL(pred_train_pls,y)
score_pred_valid_pls = KL(pred_valid_pls,y_valid)
pls_YX = PLSRegression(n_components=min(8,y.shape[1]),scale=False)
pls_YX.fit(y,data['X'])
gen_train_pls = pls_YX.predict(y)
gen_valid_pls = pls_YX.predict(y_valid)
score_gen_train_pls = L2(gen_train_pls,data['X'])
score_gen_valid_pls = L2(gen_valid_pls,data['X_valid'])
pred_train = m.predict(x=data['X'],deterministic=True)
pred_train = np.hstack([pred_train,1-pred_train.sum(axis=1,keepdims=True)])
score_pred_train = KL(pred_train,y)
pred_valid = m.predict(x=data['X_valid'],deterministic=True)
pred_valid = np.hstack([pred_valid,1-pred_valid.sum(axis=1,keepdims=True)])
score_pred_valid = KL(pred_valid,y_valid)
if m.model_type in [1,2]:
z2_train = m.getZ2(x=data['X'],y=data['y'],deterministic=True)
z2_valid = m.getZ2(x=data['X_valid'],y=data['y_valid'],deterministic=True)
z2_train_mean = z2_train.mean(axis=0)
z2_valid_mean = z2_valid.mean(axis=0)
z2_gen_train = z2_train_mean*np.ones_like(z2_train).astype('float32')
z2_gen_valid = z2_valid_mean*np.ones_like(z2_valid).astype('float32')
z2_gen_manifold = z2_valid_mean*np.ones((simplex.shape[0],z2_valid.shape[1])).astype('float32')
z2_gen_endmembers = z2_train_mean*np.ones((y_corners.shape[0],z2_train.shape[1])).astype('float32')
gen_train = f(_ux + m.generate(y=data['y'][inds_sup_train],z2=z2_gen_train[inds_sup_train],deterministic=True)) # true by default for non-variational, variational default is False
gen_valid = f(_ux + m.generate(y=data['y_valid'][inds_sup_valid],z2=z2_gen_valid[inds_sup_valid],deterministic=True))
manifold = f(_ux + m.generate(y=simplex,z2=z2_gen_manifold,deterministic=True))
endmembers = f(_ux + m.generate(y=y_corners,z2=z2_gen_endmembers,deterministic=True))
if m.variational:
endmembers_dists = []
for idx_c, c in enumerate(y_corners):
endmembers_dist = [f(_ux + m.generate(y=np.atleast_2d(c),z2=z2_gen_endmembers[idx_c:idx_c+1],deterministic=False)).squeeze() for i in range(sample_size)]
endmembers_dists += [np.asarray(endmembers_dist)]
endmembers_dists = endmembers_dists
else:
gen_train = f(_ux + m.generate(y=data['y'][inds_sup_train],deterministic=True)) # true by default for non-variational, variational default is False
gen_valid = f(_ux + m.generate(y=data['y_valid'][inds_sup_valid],deterministic=True))
manifold = f(_ux + m.generate(y=simplex,deterministic=True))
endmembers = f(_ux + m.generate(y=y_corners,deterministic=True))
if m.variational:
endmembers_dists = []
for idx_c, c in enumerate(y_corners):
endmembers_dist = [f(_ux + m.generate(y=np.atleast_2d(c),deterministic=False)).squeeze() for i in range(sample_size)]
endmembers_dists += [np.asarray(endmembers_dist)]
endmembers_dists = endmembers_dists
recon_train = f(_ux + m.generate(x=data['X_'][inds_train_x],deterministic=True))
recon_sup_valid = f(_ux + m.generate(x=data['X_valid'][inds_sup_valid],deterministic=True))
fs = 24
fs_tick = 18
# change xticks to be names
p = 100
plt.plot(p*y[inds_sup_train][0],'k',lw=2,label='Ground Truth')
ssdgm_label = 'SSDGM ({:.3f})'.format(score_pred_train)
plt.plot(p*pred_train[inds_sup_train][0],'r-.',lw=2,label=ssdgm_label)
pls_label = 'PLS ({:.3f})'.format(score_pred_train_pls)
plt.plot(p*pred_train_pls[inds_sup_train][0],'b-.',lw=2,label=pls_label)
plt.plot(p*y[inds_sup_train].T,'k',lw=2)
plt.plot(p*pred_train[inds_sup_train].T,'r-.',lw=2)
plt.plot(p*pred_train_pls[inds_sup_train].T,'b-.',lw=2)
plt.title('Predicting Composition - Training Error', fontsize=fs)
plt.ylabel('Composition (%)', fontsize=fs)
ax = plt.gca()
ax.set_ylim((0,1*p))
ax.set_xticks(np.arange(y.shape[1]))
ax.set_xticklabels(names, fontsize=fs)
ax.tick_params(axis='x',direction='out',top='off',length=10,labelsize=fs_tick)
lgd = plt.legend(loc='center left',bbox_to_anchor=(1, 0.5))
ax = plt.gca()
plt.savefig(res_out+'/comp_train.png',additional_artists=[lgd],bbox_inches='tight')
plt.close()
plt.plot(p*y_valid[inds_sup_valid][0],'k',lw=2,label='Ground Truth')
ssdgm_label = 'SSDGM ({:.3f})'.format(score_pred_valid)
plt.plot(p*pred_valid[inds_sup_valid][0],'r-.',lw=2,label=ssdgm_label)
pls_label = 'PLS ({:.3f})'.format(score_pred_valid_pls)
plt.plot(p*pred_valid_pls[inds_sup_valid][0],'b-.',lw=2,label=pls_label)
plt.plot(p*y_valid[inds_sup_valid].T,'k',lw=2)
plt.plot(p*pred_valid[inds_sup_valid].T,'r-.',lw=2)
plt.plot(p*pred_valid_pls[inds_sup_valid].T,'b-.',lw=2)
plt.title('Predicting Composition - Validation Error', fontsize=fs)
plt.ylabel('Composition (%)', fontsize=fs)
ax = plt.gca()
ax.set_ylim((0,1*p))
ax.set_xticks(np.arange(y.shape[1]))
ax.set_xticklabels(names, fontsize=fs)
ax.tick_params(axis='x',direction='out',top='off',length=10,labelsize=fs_tick)
lgd = plt.legend(loc='center left',bbox_to_anchor=(1, 0.5))
ax = plt.gca()
plt.savefig(res_out+'/comp_valid.png',additional_artists=[lgd],bbox_inches='tight')
plt.close()
plt.plot(waves,f(_ux+data['X'][inds_sup_train]).T,'k')
plt.plot(waves,gen_train.T,'r-.')
plt.title('Generating Spectra - Training Error', fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
if force_ylim:
plt.gca().set_ylim(ylim)
plt.savefig(res_out+'/genspectra_train.png')
plt.close()
plt.plot(waves,f(_ux+data['X_valid'][inds_sup_valid]).T,'k')
plt.plot(waves,gen_valid.T,'r-.')
plt.title('Generating Spectra - Validation Error', fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
if force_ylim:
plt.gca().set_ylim(ylim)
plt.savefig(res_out+'/genspectra_valid.png')
plt.close()
if m.variational:
for endmember, color, name in zip(endmembers,colors,names):
plt.plot(waves,endmember,color=color,lw=2,label=name)
for endmember_dist, color in zip(endmembers_dists,colors):
plt.plot(waves,endmember_dist.T,'-.',color=color,lw=1)
plt.title('Generating Endmembers with Distributions', fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
lgd = plt.legend(loc='center left',bbox_to_anchor=(1, 0.5))
ax = plt.gca()
if force_ylim:
ax.set_ylim(ylim)
plt.savefig(res_out+'/endmembers_dist.png',additional_artists=[lgd],bbox_inches='tight')
plt.close()
for endmember, color, name in zip(endmembers,colors,names):
plt.plot(waves,endmember,color=color,lw=2,label=name)
plt.title('Generating Endmembers', fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
lgd = plt.legend(loc='center left',bbox_to_anchor=(1, 0.5))
if m.variational:
plt.gca().set_ylim(ax.get_ylim())
if force_ylim:
plt.gca().set_ylim(ylim)
plt.savefig(res_out+'/endmembers_means.png',additional_artists=[lgd],bbox_inches='tight')
plt.close()
for endmember, color, name in zip(endmembers,colors,names):
plt.plot(waves,endmember,color=color,lw=2,label=name)
for endmember, color, name in zip(groundtruth,colors,names):
plt.plot(waves,endmember[:len(waves)],color=color,lw=6,alpha=0.4)
score_gen_endmembers = L2(endmembers,groundtruth[:,:len(waves)])
if title is None:
plt.title('Generating Endmembers with Ground Truth ({:.3f})'.format(score_gen_endmembers), fontsize=fs)
else:
plt.title(title+' ({:.3f})'.format(score_gen_endmembers), fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
lgd = plt.legend(loc='lower right', fontsize=fs)
# lgd = plt.legend(loc='center left',bbox_to_anchor=(1, 0.5))
if m.variational:
plt.gca().set_ylim(ax.get_ylim())
if force_ylim:
plt.gca().set_ylim(ylim)
plt.savefig(res_out+'/endmembers_means_with_groundtruth.png',additional_artists=[lgd],bbox_inches='tight')
plt.close()
plt.plot(waves,manifold.T,color='lightgray',lw=1,alpha=0.1)
for endmember, color, name in zip(groundtruth,colors,names):
plt.plot(waves,endmember[:len(waves)],color=color,lw=6,alpha=1.0)
plt.title('Spectral Manifold', fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
lgd = plt.legend(loc='center left',bbox_to_anchor=(1, 0.5))
if m.variational:
plt.gca().set_ylim(ax.get_ylim())
if force_ylim:
plt.gca().set_ylim(ylim)
plt.savefig(res_out+'/manifold.png',bbox_inches='tight')
plt.close()
plt.plot(waves,f(_ux+data['X_'][inds_train_x]).T,'k')
plt.plot(waves,recon_train.T,'r-.')
plt.title('Reconstructing Spectra - Training Error', fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
if force_ylim:
plt.gca().set_ylim(ylim)
plt.savefig(res_out+'/recon_train.png')
plt.close()
plt.plot(waves,f(_ux+data['X_valid'][inds_sup_valid]).T,'k')
plt.plot(waves,recon_sup_valid.T,'r-.')
plt.title('Reconstructing Spectra - Validation Error', fontsize=fs)
plt.xlabel('Channels', fontsize=fs)
plt.ylabel('Intensities', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
if force_ylim:
plt.gca().set_ylim(ylim)
plt.savefig(res_out+'/recon_valid.png')
plt.close()
if m.model_type in [1,2]:
# need to use vertical lines to denote edges of datasets
# write dataset i in middle of range on xlabel
for i in range(z2_train.shape[1]):
plt.plot(z2_train[:,i],'r-.')
plt.title('Nuisance Variable '+str(i)+' - Training', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
plt.savefig(res_out+'/nuisance_train_'+str(i)+'.png')
plt.close()
plt.plot(z2_valid[:,i],'r-.')
ax = plt.gca()
ylim = ax.get_ylim()
# should make this general if possible
plt.plot([1866,1866],[-5,5],'k--')
plt.plot([1866+1742,1866+1742],[-5,5],'k--')
# plt.plot([1866+1742+1746,1866+1742+1746],[-5,5],'k--')
ax.set_ylim(ylim)
plt.title('Nuisance Variable '+str(i)+' - Validation', fontsize=fs)
plt.tick_params(axis='both', which='major', labelsize=fs_tick)
plt.savefig(res_out+'/nuisance_valid_'+str(i)+'.png')
plt.close()
(Xtrain, ytrain) = loadData(xtrainpath, ytrainpath) (Xtest, ytest) = loadData(xtestpath, ytestpath) #trim off background and scale ytrain=ytrain[:,1:] #ytrain=scale(ytrain) Xtrain=standardize(Xtrain) #trim off background and scale ytest = ytest[:,1:] #ytest = scale(ytest) Xtest = standardize(Xtest) pls = PLSRegression(n_components=10) pls.fit(Xtrain, ytrain) y_pls = pls.predict(Xtest) print 1 + pls.score(Xtest, ytest) pls_rmse=[] pls_rmse.append(sqrt(mean_squared_error(ytest[:,0], y_pls[:,0]))) pls_rmse.append(sqrt(mean_squared_error(ytest[:,1], y_pls[:,1]))) pls_rmse.append(sqrt(mean_squared_error(ytest[:,2], y_pls[:,2]))) pls_rmse.append(sqrt(mean_squared_error(ytest[:,3], y_pls[:,3]))) fig = plt.figure(figsize=(20,10)) ax1 = fig.add_subplot(241) ax1.plot(y_pls[:,0], c='r', label='PLS Fit')
import os
import numpy as np
from matplotlib import pyplot as plt
from sklearn.cross_decomposition import PLSRegression
from sklearn import metrics
os.chdir("D:/01. CLASS/Machine Learning/")
pls = PLSRegression(n_components=6, scale=False, max_iter=50000, copy=True)
init_lamda_PLS = 1
lamda_PLS = 1
Tgt = np.array([0, 50])
A_p1 = np.array([[0.5, -0.2], [0.25, 0.15]])
d_p1 = np.array([[0.1, 0], [0.05, 0]])
C_p1 = np.transpose(
np.array([[0, 0.5, 0.05, 0, 0.15, 0], [0.085, 0, 0.025, 0.2, 0, 0]]))
sample_init_EP = []
sample_vm_EP = []
sample_init_VP = []
sample_vm_VP = []
np.random.seed(1000000) # 4
I = np.identity(2)
# L1_SC = 0.55
# L2_SC = 0.75
L1_SC = 0.45
L2_SC = 0.35
def __init__(self, dataset):
pitchRoll = dataset[:, [2, 3]]
motorPos = dataset[:, [0, 1]]
self.polyModelLeft = PLSRegression(n_components=2)
self.polyModelLeft.fit(pitchRoll, motorPos)
dms = pd.get_dummies(data[['League', 'Division', 'NewLeague']])
# 准备数据
y = data['Salary']
x_ = data.drop(['Salary', 'League', 'Division', 'NewLeague'],
axis=1).astype('float64')
x = pd.concat([x_, dms[['League_N', 'Division_W', 'NewLeague_N']]], axis=1)
# 训练集、测试集划分
x_train, x_test, y_train, y_test = train_test_split(x,
y,
test_size=0.25,
random_state=42)
# 回归模型、参数
pls_model_setup = PLSRegression(scale=True, max_iter=5000, n_components=5)
param_grid = {'n_components': range(1, 20)}
# GridSearchCV优化参数、训练模型
gsearch = GridSearchCV(pls_model_setup, param_grid)
pls_model = gsearch.fit(x_train, y_train)
# 打印 coef
print('Partial Least Squares Regression coefficients:',
pls_model.best_estimator_.coef_)
# 对测试集做预测
pls_prediction = pls_model.predict(x_test)
# 计算R2，均方差
pls_r2 = r2_score(y_test, pls_prediction)
def VIP(X, Y, H, NumDes):
from sklearn.cross_decomposition import PLSRegression
import numpy as np
from sklearn.cross_validation import KFold
import PCM_workflow as PW
print('############## VIP is being processed ###############')
M = list(X.keys())
H_VIP, X_VIP, Y_VIP, HArray = {}, {}, {}, {}
NumDesVIP = np.zeros((13, 6), dtype=int)
for kk in M:
Xtrain, Ytrain = X[kk], Y
kf = KFold(len(Ytrain), 10, indices=True, shuffle=True, random_state=1)
HH = H[kk]
nrow, ncol = np.shape(Xtrain)
ArrayYpredCV, Q2, RMSE_CV, OptimalPC = PW.CV_Processing(
Xtrain, Ytrain, kf)
plsmodel = PLSRegression(n_components=OptimalPC)
plsmodel.fit(Xtrain, Ytrain)
x_scores = plsmodel.x_scores_
x_weighted = plsmodel.x_weights_
m, p = nrow, ncol
m, h = np.shape(x_scores)
p, h = np.shape(x_weighted)
X_S, X_W = x_scores, x_weighted
co = []
for i in range(h):
corr = np.corrcoef(np.squeeze(Ytrain), X_S[:, i])
co.append(corr[0][1]**2)
s = sum(co)
vip = []
for j in range(p):
d = []
for k in range(h):
d.append(co[k] * X_W[j, k]**2)
q = sum(d)
vip.append(np.sqrt(p * q / s))
idx_keep = [idx for idx, val in enumerate(vip) if vip[idx] >= 1]
idxDes = NumDes[int(kk[6:]) - 1, :]
L, P, LxP, LxL, PxP = [], [], [], [], []
for idx in idx_keep:
if idx >= 0 and idx < np.sum(idxDes[0:1]):
L.append(idx)
elif idx >= np.sum(idxDes[0:1]) and idx < np.sum(idxDes[0:2]):
P.append(idx)
elif idx >= np.sum(idxDes[0:2]) and idx < np.sum(idxDes[0:3]):
LxP.append(idx)
elif idx >= np.sum(idxDes[0:3]) and idx < np.sum(idxDes[0:4]):
LxL.append(idx)
elif idx >= np.sum(idxDes[0:4]) and idx < np.sum(idxDes):
PxP.append(idx)
NVIP = np.array(
[len(L),
len(P),
len(LxP),
len(LxL),
len(PxP),
len(idx_keep)])
NumDesVIP[int(kk[6:]) - 1, :] = NumDesVIP[int(kk[6:]) - 1, :] + NVIP
hvip = np.array(HH)[idx_keep]
vvip = np.array(vip)[idx_keep]
H_VIP[kk] = hvip
X_VIP[kk] = Xtrain[:, idx_keep]
Y_VIP = Ytrain
hvip = np.reshape(hvip, (len(hvip), 1))
vvip = np.reshape(vvip, (len(vvip), 1))
HArray[kk] = np.append(hvip, vvip, axis=1)
return X_VIP, Y_VIP, H_VIP, HArray, NumDesVIP
#%% PCA, SVD, PLS
from sklearn.cross_decomposition import PLSRegression
from sklearn.decomposition import PCA, TruncatedSVD
pca = PCA(n_components=8)
pca_feats = [3, 5, 10, 14, 18, 19, 22, 23, 25, 26, 27]
train_pca_df = pd.DataFrame([])
test_pca_df = pd.DataFrame([])
for feat in pca_feats:
feat_label = "F" + str(feat)
train_pca_df[feat_label] = train_features[feat_label]
test_pca_df[feat_label] = test_features[feat_label]
pls = PLSRegression(n_components=8) # This works good for the log reg model
pls.fit(train_pca_df, train_y)
train_feats_pls = pd.DataFrame(pls.transform(train_pca_df),
index=train_features.index)
test_feats_pls = pd.DataFrame(pls.transform(test_pca_df),
index=test_features.index)
#%% Replace pca feats with new feats
for feat in pca_feats:
feat_label = "F" + str(feat)
train_features = train_features.drop([feat_label], axis=1)
test_features = test_features.drop([feat_label], axis=1)
train_features = pd.concat([train_features, train_feats_pls], axis=1)
test_features = pd.concat([test_features, test_feats_pls], axis=1)
#%% Logistic Regression on the initial features
#correct not accurate
from sklearn.cross_validation import train_test_split
from sklearn.neighbors import KNeighborsClassifier
from sklearn import metrics
from sklearn.svm import SVC
import numpy as np
import pandas as pd
from sklearn.cross_decomposition import PLSRegression
from sklearn.cross_decomposition import PLSCanonical
df=pd.read_csv('newdata.csv')
x=df.drop(['tag'],axis=1)
y=df.drop(['kx','ky','kz','wa','wb','wc','wd','we','wf'],axis=1)
X_train , X_test , Y_train , Y_test = train_test_split(x,y , random_state=5)
plsr=PLSRegression()
plsr.fit(X_train,Y_train)
plsc=PLSCanonical()
plsc.fit(X_train,Y_train)
print (plsr.score(X_test,Y_test))
print (plsc.score(X_test,Y_test))
def generate(self, input=None):
dso = input
_experiment_test = self.config.get('experiment_test')
_experiment_control = self.config.get('experiment_control')
data = dso.data
plsr = PLSRegression(n_components=self.config.get('number_of_components'), scale=self.config.get('autoscale')) #, algorithm=self.config.get('algorithm'))
Y = np.array([0 if c == _experiment_control else 1 for c in dso.classes[0] ])
plsr.fit(data, Y) # Transpose it, as vars need to along the top
# Build scores into a dso no_of_samples x no_of_principal_components
scored = DataSet(size=(len(plsr.x_scores_),len(plsr.x_scores_[0])))
scored.labels[0] = input.labels[0]
scored.classes[0] = input.classes[0]
for n,s in enumerate(plsr.x_scores_.T):
scored.data[:,n] = s
scored.labels[1][n] = 'Latent Variable %d' % (n+1) #, plsr.y_weights_[0][n])
# PLS-DA regions; mean +- 95% confidence in each axis for each cluster
cw_x = defaultdict(list)
cw_y = defaultdict(list)
for c in list(cw_x.keys()):
# Calculate mean point
cx = np.mean( cw_x[c] )
cy = np.mean( cw_y[c] )
# Calculate 95% CI
rx = np.std( cw_x[c] ) *2 # 2sd = 95% #1.95 * ( / srn) # 1.95 * SEM => 95% confidence
ry = np.std( cw_y[c] ) *2 #1.95 * ( / srn)
figure_regions.append(
(c, cx, cy, rx, ry)
)
# Label up the top 50 (the values are retained; just for clarity)
wmx = np.amax( np.absolute( plsr.x_weights_), axis=1 )
dso_z = list(zip( dso.scales[1], dso.entities[1], dso.labels[1] ))
dso_z = sorted( zip( dso_z, wmx ), key=lambda x: x[1])[-50:] # Top 50
dso_z = [x for x, wmx in dso_z ]
weightsd = DataSet(size=plsr.x_weights_.T.shape)
weightsd.data = plsr.x_weights_.T
weightsd.scales[1] = input.scales[1]
dso_lv = {}
for n in range(0, plsr.x_weights_.shape[1] ):
lvd = DataSet( size=(1, input.shape[1] ) )
lvd.entities[1] = input.entities[1]
lvd.labels[1] = input.labels[1]
lvd.scales[1] = input.scales[1]
lvd.data = plsr.x_weights_[:,n:n+1].T
dso_lv['lv%s' % (n+1)] = lvd
weightsd.labels[0][n] = "Weights on LV %s" % (n+1)
weightsd.classes[0][n] = "LV %s" % (n+1)
return dict(list({
'dso': dso,
'scores':scored,
'weights':weightsd,
#'figure_data': figure_data,
#'figure_regions': figure_regions,
'y_weights': plsr.y_weights_,
'x_weights': plsr.x_weights_,
}.items()) + list(dso_lv.items()) )
'param': {
'n_estimators': range_t
}
},
'SVR': {
'name': 'SVR',
'model': SVR(),
'param': {
'gamma': range_g,
'C': range_c,
'epsilon': range_e
}
},
'PLS': {
'name': 'PLS',
'model': PLSRegression(),
'param': {
'n_components': range_p
}
},
'GPR': {
'name': 'GPR',
'model': GaussianProcessRegressor(kernel=kernel),
'param': {
'n_restarts_optimizer': range_o
}
},
}
key = 'RF' # 'RR' 'EN', 'LASSO', 'kNN', 'RF', 'GB', 'SVR', 'PLS', 'GPR'
name = model_param[key]['name']
def run_full_caltarget_test(cal_spectra,
cal_labels,
cal_names,
transformed_test,
pls_comps=[10]):
samples, comps = ct.load_data(norm=3, masked=True)
org_samples = np.copy(samples)
org_comps = np.copy(comps)
elements = [
'SiO2', 'TiO2', 'Al2O3', 'FeOT', 'MnO', 'MgO', 'CaO', 'Na2O', 'K2O'
]
for n_comps in pls_comps:
mars_preds = []
ct_preds = []
gt_preds = []
for e, elem in enumerate(elements):
if verbose:
print '-----------------------'
print elem
for transformer in transformed_test.keys():
print transformer
targets = transformed_test[transformer]
for t, target in enumerate(targets):
# Remove caltargets
if (comps['Name'] == cal_names[t]).any():
ind = np.argwhere(comps['Name'] == cal_names[t])[0, 0]
comps = np.delete(org_comps, ind, 0)
samples = np.delete(org_samples, ind, 0)
model = PLSRegression(n_components=n_comps, scale=False)
model.fit(samples, comps[elem])
lab_pred = model.predict(cal_spectra[0][t][None])
mars_pred = model.predict(cal_spectra[1][t][None])
trans_pred = model.predict(target)
gt = cal_labels[t, e]
score = (norm(mars_pred - gt, ord=1) -
norm(trans_pred - gt, ord=1))
mars_preds.append(mars_pred[0][0])
ct_preds.append(trans_pred[0][0])
gt_preds.append(gt)
if verbose:
print cal_names[t]
print 'Ground truth: %.4f' % gt
print 'Lab target: %.4f' % lab_pred
print 'Mars target: %.4f' % mars_pred
print 'Transformed Mars: %.4f' % trans_pred
print 'Score: %.4f' % score
print
pred_shape = (len(elements), len(targets))
ct_preds = np.array(ct_preds).reshape(pred_shape)
gt_preds = np.array(gt_preds).reshape(pred_shape)
mars_preds = np.array(mars_preds).reshape(pred_shape)
print '-----------------------'
print "Element\tMars\t\tCalTran\t\t%Gain/lost"
for i, e in enumerate(elements):
mars_rmsep = rmse(gt_preds[i, :], mars_preds[i, :])
ct_rmsep = rmse(gt_preds[i, :], ct_preds[i, :])
print e,
print "\t%f" % round(mars_rmsep, 4),
print "\t%f" % round(ct_rmsep, 4),
print "\t%f" % round((mars_rmsep - ct_rmsep) * 100 / mars_rmsep, 4)
print '-----------------------'
print "Sample\tMars\t\tCalTran\t\t%Gain/lost"
for i, n in enumerate(names):
mars_rmsep = rmse(gt_preds[:, i], mars_preds[:, i])
ct_rmsep = rmse(gt_preds[:, i], ct_preds[:, i])
print n,
print "\t%f" % round(mars_rmsep, 4),
print "\t%f" % round(ct_rmsep, 4),
print "\t%f" % round((mars_rmsep - ct_rmsep) * 100 / mars_rmsep, 4)
def plsvip (X, Y, V, lat_var):
attributes = len(X[0])
if not lat_var:
latent_variables = attributes
else:
latent_variables = lat_var
num_instances = len(X)
attributes_gone = []
min_att = -1
#start_time = time.time()
#attr_time = time.time()
#time_counter = 0
while attributes>0:
#if (attributes +9) %10 ==0:
# print "total time: ", time.time() - start_time
# print "attr time: ", time.time() - attr_time
# attr_time = time.time()
if (latent_variables == 0) or (latent_variables > attributes):
latent_variables = attributes
lv_best = best_latent_variable(X, Y, latent_variables, num_instances)
#print "current best lv: ", lv_best, "num. attr. ", attributes ####
#fin_pls = PLSCanonical(n_components = lv_best)
fin_pls = PLSRegression(n_components = lv_best)
fin_pls.fit(X, Y)
currentR2 = fin_pls.score(X, Y)
#######################################w
# alternative r2
"""
meanY4r2 = numpy.mean(Y)
predY = fin_pls.predict(X)
RSS = 0
for i in range (len(Y)):
RSS += numpy.power (Y[i] - predY[i], 2)
TSS = 0
for i in range (len(Y)):
TSS += numpy.power (Y[i] - meanY4r2, 2)
alterR2 = 1 - (RSS/TSS)
#print currentR2, "vs", alterR2
"""
#######################################w
min_vip = 1000
if min_att ==-1:
attributes_gone.append(["None", currentR2, attributes, lv_best])
##########################################r
#threaded version
"""
myThreads = []
VIPcurrent = []
for i in range (0,attributes):
myThreads.append(enthread( target = get_vip, args = (fin_pls, lv_best, i, attributes_gone, attributes )) )
for i in range (0,attributes):
VIPcurrent.append(myThreads[i].get())
min_vip = min(VIPcurrent)
min_att = VIPcurrent.index(min_vip)
"""
# Working version
#"""
for i in range (0,attributes):
VIPcurrent = get_vip (fin_pls, lv_best, i, attributes_gone, attributes )
if VIPcurrent< min_vip:
min_vip = VIPcurrent
min_att = i
#"""
##########################################r
if min_att >-1:
attributes_gone.append([V[min_att], currentR2, attributes, lv_best]) ####### CURRENT : to BE popped, NOT already popped
V.pop(min_att)
for i in range (num_instances):
X[i].pop(min_att)
attributes -= 1
#print attributes_gone ####
#time_counter +=1
return attributes_gone
n_components = 75 #X_train[0].shape[1]
for vid,Xt,yt in zip(subjId_val, X_val, y_val):
levelOneTest = []
levelOneTrain = []
X_levelOne = []
y_levelOne = []
level0Classifier = []
for tid,Xp,yp in zip(subjId_train,X_train,y_train):
print "Predicting subject ", vid, "from subject ", tid
y0 = np.zeros(yp.shape)
y1 = np.ones(Xt.shape[0])
X = np.vstack([Xp,Xt])
yd = np.concatenate([y0,y1])
pls = PLSRegression(n_components)
Xp_t, Xp_v, yp_t, yp_v = tts(Xp.copy(),yp.copy(),train_size=0.9)
yp_t = yp_t.astype(bool)
yp_t_not = np.vstack((yp_t,~yp_t)).T
#print "yp_t_not ", yp_t_not.shape
pls.fit(Xp_t,yp_t_not.astype(int))
yp_new = pls.predict(Xp_t, copy=True)
yp_pred = (yp_new[:,0] > yp_new[:,1]).astype(int)
yp_t = yp_t.astype(int)
#print y_new,y_pred, y_t
error = ((yp_t - yp_pred) ** 2).sum()
print "PLS Training error " , float(error)/yp_t.shape[0]
yp_new = pls.predict(Xp_v, copy=True)
yp_pred = (yp_new[:,0] > yp_new[:,1]).astype(int)
#print y_new, y_pred, y_v
#print ((y_v - y_pred) ** 2).sum(), y_v.shape[0]
plt.xticks(())
plt.yticks(())
plt.show()
# #############################################################################
# PLS regression, with multivariate response, a.k.a. PLS2
n = 1000
q = 3
p = 10
X = np.random.normal(size=n * p).reshape((n, p))
B = np.array([[1, 2] + [0] * (p - 2)] * q).T
# each Yj = 1*X1 + 2*X2 + noize
Y = np.dot(X, B) + np.random.normal(size=n * q).reshape((n, q)) + 5
pls2 = PLSRegression(n_components=3)
pls2.fit(X, Y)
print("True B (such that: Y = XB + Err)")
print(B)
# compare pls2.coef_ with B
print("Estimated B")
print(np.round(pls2.coef_, 1))
pls2.predict(X)
# PLS regression, with univariate response, a.k.a. PLS1
n = 1000
p = 10
X = np.random.normal(size=n * p).reshape((n, p))
y = X[:, 0] + 2 * X[:, 1] + np.random.normal(size=n * 1) + 5
pls1 = PLSRegression(n_components=3)
def main():
args = parser.parse_args()
# 模型选择及输入参数
model_name = {
1: 'PLSR',
2: 'LS-SVR',
3: 'GPR',
4: 'FCN',
5: 'LSTM',
6: 'GCN',
7: 'MC-GCN',
8: 'GC-LSTM'
}
print(model_name)
model_select = list(input('Select models:'))
# 初始化结果
results = {
'adj': [],
'r2': [],
'rmse': [],
'loss_hist': [],
'prediction': []
}
# os.mkdir('Results')
f = open('Results/params.txt', 'w+')
f.write('Parameters setting:\n{}\n\n'.format(args.__dict__))
# 导入数据
data = pd.read_excel('8号机磨煤机C_正常.xlsx',
index_col=0,
header=1,
nrows=args.length + 5001)
data = data.iloc[5001:, :]
# 数据划分
predict_variable = [3, 12, 15, 20, 23]
y = data.iloc[:, predict_variable]
X = data.drop(columns=y.columns)
X_train, y_train = X.iloc[:int(args.length * args.train_size
)], y.iloc[:int(args.length *
args.train_size)]
X_test, y_test = X.iloc[int(args.length * args.train_size
):], y.iloc[int(args.length *
args.train_size):]
# 导出数据
# X_train.to_csv('Results/X_train.csv', header=False, index=False)
# X_test.to_csv('Results/X_test.csv', header=False, index=False)
# y_train.to_csv('Results/y_train.csv', header=False, index=False)
# y_test.to_csv('Results/y_test.csv', header=False, index=False)
# 设定种子
np.random.seed(args.seed)
torch.manual_seed(args.seed)
# 多次实验
for exp in range(args.n_exp):
print('=====Experiment({}/{})====='.format(exp + 1, args.n_exp))
f.write('=====Experiment({}/{})=====\n'.format(exp + 1, args.n_exp))
results['adj'].append({})
results['r2'].append({})
results['rmse'].append({})
results['loss_hist'].append({})
results['prediction'].append({})
# PLSR
if '1' in model_select:
flag = 1
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
reg = PLSRegression(args.n_components).fit(X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# LS-SVR
if '2' in model_select:
flag = 2
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
reg = LssvrModel(args.c, args.sigma).fit(X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# GPR
if '3' in model_select:
flag = 3
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
kernel = DotProduct() * RBF(args.length_scale,
(args.length_scale, args.length_scale))
reg = GaussianProcessRegressor(kernel=kernel,
alpha=args.alpha).fit(
X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# FCN
if '4' in model_select:
flag = 4
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
reg = FcnModel(X_train.shape[1], y_train.shape[1],
(1024, 256, 256, 256), args.n_epoch,
args.batch_size, args.lr, args.weight_decay,
args.step_size, args.gamma).fit(X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
results['loss_hist'][-1].update({model_name[flag]: reg.loss_hist})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# LSTM
if '5' in model_select:
flag = 5
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
reg = LstmModel(X_train.shape[1], y_train.shape[1], (1024, ),
(256, 256, 256), args.seq_len, args.n_epoch,
args.batch_size, args.lr, args.weight_decay,
args.step_size, args.gamma).fit(X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
results['loss_hist'][-1].update({model_name[flag]: reg.loss_hist})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# GCN
if '6' in model_select:
flag = 6
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
reg = GcnModel(X_train.shape[1], y_train.shape[1], (1024, ),
(256, 256, 256), args.graph_reg, args.self_con,
args.n_epoch, args.batch_size, args.lr,
args.weight_decay, args.step_size,
args.gamma).fit(X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
results['loss_hist'][-1].update({model_name[flag]: reg.loss_hist})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# MC-GCN
if '7' in model_select:
flag = 7
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
reg = McgcnModel(X_train.shape[1], (1024, ), (256, ), (256, 256),
y_train.shape[1], args.graph_reg, args.self_con,
args.n_epoch, args.batch_size, args.lr,
args.weight_decay, args.step_size,
args.gamma).fit(X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['adj'][-1].update({model_name[flag]: reg.adj})
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
results['loss_hist'][-1].update({model_name[flag]: reg.loss_hist})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# GC-LSTM
if '8' in model_select:
flag = 8
print('====={}====='.format(model_name[flag]))
f.write('====={}=====\n'.format(model_name[flag]))
# 训练&测试
t1 = time.time()
reg = GclstmModel(X_train.shape[1], (1024, ), (256, ), (256, 256),
y_train.shape[1], args.seq_len, args.graph_reg,
args.self_con, args.n_epoch, args.batch_size,
args.lr, args.weight_decay, args.step_size,
args.gamma).fit(X_train, y_train)
t2 = time.time()
y_pred = reg.predict(X_test)
t3 = time.time()
y_fit = reg.predict(X_train)
print(reg.get_params())
print('Time:\nFit: {:.3f}s Pred: {:.3f}s'.format(t2 - t1, t3 - t2))
print('R2:\nFit: {} Pred: {}'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 写入文件
f.write(str(reg.get_params()) + '\n')
f.write('Time:\nFit: {:.3f}s Pred: {:.3f}s\n'.format(
t2 - t1, t3 - t2))
f.write('R2:\nFit: {} Pred: {}\n'.format(
r2_score(y_train, y_fit, multioutput='raw_values'),
r2_score(y_test, y_pred, multioutput='raw_values')))
# 存储结果和模型
index = r2_rmse(y_test, y_pred, y.columns, f)
results['adj'][-1].update({model_name[flag]: reg.adj})
results['r2'][-1].update({model_name[flag]: index[0]})
results['rmse'][-1].update({model_name[flag]: index[1]})
results['prediction'][-1].update({model_name[flag]: y_pred})
results['loss_hist'][-1].update({model_name[flag]: reg.loss_hist})
joblib.dump(
reg, 'Results/{}-{}.model'.format(model_name[flag], exp + 1))
# 存储结果
np.save('Results/results.npy', results)
f.close()
if method_name[0:3] == 'jit':
nn_model = NearestNeighbors(metric='euclidean') # サンプル選択用の k-NN モデルの宣言
# オートスケーリング
autoscaled_x_train = (x_train - x_train.mean()) / x_train.std()
autoscaled_y_train = (y_train - y_train.mean()) / y_train.std()
autoscaled_x_test = (x_test - x_train.mean()) / x_train.std()
# ハイパーパラメータの最適化やモデリング
if method_name == 'pls' or method_name == 'mwpls':
# CV による成分数の最適化
components = [] # 空の list の変数を作成して、成分数をこの変数に追加していきます同じく成分数をこの変数に追加
r2_in_cv_all = [] # 空の list の変数を作成して、成分数ごとのクロスバリデーション後の r2 をこの変数に追加
for component in range(1, min(np.linalg.matrix_rank(autoscaled_x_train), max_number_of_principal_components) + 1):
# PLS
model = PLSRegression(n_components=component) # PLS モデルの宣言
estimated_y_in_cv = pd.DataFrame(cross_val_predict(model, autoscaled_x_train, autoscaled_y_train,
cv=fold_number)) # クロスバリデーション推定値の計算し、DataFrame型に変換
estimated_y_in_cv = estimated_y_in_cv * y_train.std() + y_train.mean() # スケールをもとに戻す
r2_in_cv = metrics.r2_score(y_train, estimated_y_in_cv) # r2 を計算
print(component, r2_in_cv) # 成分数と r2 を表示
r2_in_cv_all.append(r2_in_cv) # r2 を追加
components.append(component) # 成分数を追加
optimal_component_number = components[r2_in_cv_all.index(max(r2_in_cv_all))] # 最適成分数
# PLS
model = PLSRegression(n_components=optimal_component_number) # モデルの宣言
model.fit(autoscaled_x_train, autoscaled_y_train) # モデルの構築
elif method_name == 'svr' or method_name == 'mwsvr' or method_name == 'jitsvr':
# グラム行列の分散を最大化することによる γ の最適化
variance_of_gram_matrix = list()
for svr_gamma in svr_gammas:
mean_absolute_error(y_test, y_pred_test)
]
output.append(temp)
output = pd.DataFrame(output,
columns=[
'alpha', 'Train_R2', 'Train_MSE', 'Train_MAE',
'Test_R2', 'Test_MSE', 'Test_MAE'
])
output.to_csv('ElasticNet.csv', index=False)
plots(y_test, y_pred_test)
#model 2:PLS
from sklearn.cross_decomposition import PLSRegression
output = []
for i in range(1, 12, 2):
pls = PLSRegression(n_components=i, max_iter=10000)
pls.fit(X_train_std, y_train)
y_pred_train = pls.predict(X_train_std)
y_pred_test = pls.predict(X_test_std)
temp = [
i,
pls.score(X_train_std, y_train),
mean_squared_error(y_train, y_pred_train),
mean_absolute_error(y_train, y_pred_train),
pls.score(X_test_std, y_test),
mean_squared_error(y_test, y_pred_test),
mean_absolute_error(y_test, y_pred_test)
]
output.append(temp)
output = pd.DataFrame(output,
columns=[
def pls_ds(A, B, n_components=1):
model = PLSRegression(n_components=n_components, scale=False).fit(B, A)
return model.coefs, model.predict(B)
for i in range (5):
plt.plot(nComponents,plsCanScores[i,:],lw=3)
plt.xlim(1,np.amax(nComponents))
plt.title('PLS Cannonical accuracy')
plt.xlabel('Number of components')
plt.ylabel('accuracy')
plt.legend (['LR','LDA','GNB','Linear SVM','rbf SVM'],loc='lower right')
plt.grid(True)
if (0):
#%% PLS Regression
nComponents = np.arange(1,nClasses+1)
plsRegScores = np.zeros((5,np.alen(nComponents)))
for i,n in enumerate(nComponents):
plsReg = PLSRegression(n_components=n)
plsReg.fit(Xtrain,Ytrain)
XtrainT = plsReg.transform(Xtrain)
XtestT = plsReg.transform(Xtest)
plsRegScores[:,i] = util.classify(XtrainT,XtestT,labelsTrain,labelsTest)
plsReg = PLSRegression(n_components=2)
plsReg.fit(Xtrain,Ytrain)
xt = plsReg.transform(Xtrain)
fig = plt.figure()
util.plotData(fig,xt,labelsTrain,classColors)
plt.title('First 2 components of projected data')
#%% Plot accuracies for PLSSVD
from sklearn.cross_decomposition import PLSRegression
import pandas as pd
import numpy as np
_experiment_test = config['experiment_test']
_experiment_control = config['experiment_control']
plsr = PLSRegression(n_components=config['number_of_components'], scale=config['autoscale']) #, algorithm=self.config.get('algorithm'))
# We need classes to do the classification; should check and raise an error
class_idx = input_data.index.names.index('Class')
classes = list( input_data.index.levels[ class_idx ] )
Y = input_data.index.labels[ class_idx ]
plsr.fit(input_data.values, Y)
# Build scores into a dso no_of_samples x no_of_principal_components
scores = pd.DataFrame(plsr.x_scores_)
scores.index = input_data.index
scoresl =[]
for n,s in enumerate(plsr.x_scores_.T):
scoresl.append( 'Latent Variable %d' % (n+1) ) #, plsr.y_weights_[0][n])
scores.columns = scoresl
weights = pd.DataFrame( plsr.x_weights_.T )
weights.columns = input_data.columns
pcr_opt.fit(college_train_x, college_train_y)
reduced_college_test_x = pcr_opt.transform(college_test_x)
reduced_college_train_x = pcr_opt.transform(college_train_x)
lrm = LinearRegression()
lrm.fit(reduced_college_train_x, college_train_y)
print "\nPCR RMSE (M = " + str(opt_m) + ")"
print rmse(lrm, reduced_college_test_x, college_test_y)
#%% PLS
from sklearn.cross_decomposition import PLSRegression
pls_components = range(1, 18)
cv_pls = np.array([])
for m in pls_components:
pls = PLSRegression(n_components=m)
foo = np.transpose(college_train_x.get_values())
transformed_college_train_x = pls.fit_transform(college_train_x,
college_train_y)[0]
lrm = LinearRegression()
pls_this_rmse = rmse_cv(LinearRegression(), transformed_college_train_x,
college_train_y).mean()
cv_pls = np.append(cv_pls, pls_this_rmse)
min_m = pls_components[np.argmin(cv_pls)]
cv_pls = pd.Series(cv_pls, index=pls_components)
cv_pls.plot(title="PLSRegression Cross Validation")
plt.xlabel("Number of Components (M)")
plt.ylabel("Root Mean Square Error")
if show_plots_flag:
plt.show()
train = pd.read_csv('train.csv', index_col='id')
targets = pd.get_dummies(train.target)
train.drop('target', axis=1, inplace=True)
train = train.apply(np.log1p)
test = pd.read_csv('test.csv', index_col='id')
test = test.apply(np.log1p)
Xt, Xv, yt, yv = train_test_split(train, targets, test_size=0.2, random_state=27)
best = 10.
for n in range(5,16):
clf = PLSRegression(n_components=n)
clf.fit(Xt,yt)
y_pred = clf.predict(Xv)
loss = multiclass_log_loss(np.argmax(y_pred,axis=1),y_pred)
if loss < best:
n_best = n
best = loss
postfix = '(*)'
else:
postfix = ''
print ('comps: {:02d}\tLoss:{:5.4f} {}'.format(n,loss,postfix))
clf = PLSRegression(n_components=n_best)
clf.fit(train,targets)
y_pred = clf.predict(test)
class PLS_DA(object):
'''
'''
def __init__(self, n_comps=3, yIsDummyMatrix=False, scaleData=False):
'''
data contains n_samples, n_features
Y - response
'''
self.comps = n_comps
self.yIsDummyMatrix = yIsDummyMatrix
self.plsr = PLSRegression(n_components=n_comps, scale=scaleData)
def fit(self, X, Y):
if self.yIsDummyMatrix:
self.Ym = Y
else:
self.Ym = self.create_dummy_y(Y)
if self.evaluate_data(X, self.Ym):
self.plsr.fit(X, self.Ym)
def get_scores(self, block='x'):
'''
'''
if block == 'x':
return self.plsr.x_scores_
elif block == 'y':
return self.plsr.y_scores_
def get_weights(self, block='x'):
'''
'''
if block == 'x':
return self.plsr.x_weights_
def get_loadings(self, block='x'):
if block == 'x':
return self.plsr.x_loadings_
def get_squared_r(self, X, Y):
'''
'''
return self.plsr.score(X, Y)
def get_classes(self):
'''
'''
def get_dummy_Y(self):
'''
'''
return self.Ym
def evaluate_data(self, X, Y):
'''
'''
if X.shape[0] != Y.shape[0]:
print("Number of rows in X does not equal number of rows in Y")
return False
else:
return True
def create_dummy_y(self, Y):
'''
'''
uniqueVals = np.unique(Y)
nClasses = uniqueVals.size
Ydummy = np.zeros((Y.shape[0], nClasses))
for n, target in enumerate(Y):
col = np.where(uniqueVals == target)
Ydummy[n, col] = 1
self.classOrder = uniqueVals.tolist()
return Ydummy
# sfs_plot(sfs.get_metric_dict(), kind='std_dev')
return sfs_fit
def find_maxfit(data):
max_score = 100
print(data["training scores"])
return max(data["test scores"]), max(data["training scores"])
if __name__ == "__main__":
x, y = get_data.get_data("mango", "as7262", int_time=150,
position=2,
led_current="25 mA")
pls = PLSRegression(n_components=6)
y = y['Total Chlorophyll (µg/cm2)']
# x, y = get_data.get_data("mango", "as7262", int_time=150,
# position=2, led="b'White'",
# led_current="25 mA")
# print(x.shape)
# # pls_screen_as726x(x, y, n_comps=10)
# print(type(x))
poly = PolynomialFeatures(degree=1)
x_trans = poly.fit_transform(x)
# pls.fit(x_trans, y)
# y_predict = pls.predict(x_trans)
# print(mean_absolute_error(y, y_predict))
# ham
# n_comps = 6
# regr = PLSRegression(n_components=n_comps)
class Local_FWC_Simulator:
def __init__(self, A, d, C, seed):
self.pls = PLSRegression(n_components=6,
scale=False,
max_iter=50000,
copy=True)
np.random.seed(seed)
self.A = A
self.d = d
self.C = C
def sampling_up(self):
u1 = np.random.normal(0.4, np.sqrt(0.2))
u2 = np.random.normal(0.6, np.sqrt(0.2))
u = np.array([u1, u2])
return u
def sampling_vp(self):
v1 = np.random.normal(1, np.sqrt(0.2))
v2 = 2 * v1
v3 = np.random.uniform(0.2, 1.2)
v4 = 3 * v3
v5 = np.random.uniform(0, 0.4)
v6 = np.random.normal(-0.6, np.sqrt(0.2))
v = np.array([v1, v2, v3, v4, v5, v6])
return v
def sampling_ep(self):
e1 = np.random.normal(0, np.sqrt(0.1))
e2 = np.random.normal(0, np.sqrt(0.2))
e = np.array([e1, e2])
return e
def sampling(self,
k,
uk=np.array([0, 0]),
vp=np.array([0, 0, 0, 0, 0, 0]),
ep=np.array([0, 0]),
isInit=True):
u1 = uk[0]
u2 = uk[1]
u = uk
v1 = vp[0]
v2 = vp[1]
v3 = vp[2]
v4 = vp[3]
v5 = vp[4]
v6 = vp[5]
v = vp
e = ep
if isInit == True:
k1 = k % 100
k2 = k % 200
e = np.array([0, 0])
else:
k1 = k % 100 # n = 100 일 때 #1 entity maintenance event
k2 = k % 200 # n = 200 일 때 #1 entity maintenance event
eta_k = np.array([[k1], [k2]])
psi = np.array([u1, u2, v1, v2, v3, v4, v5, v6, k1, k2])
y = u.dot(self.A) + v.dot(self.C) + np.sum(eta_k * self.d, axis=0) + e
rows = np.r_[psi, y]
idx_end = len(rows)
idx_start = idx_end - 2
return idx_start, idx_end, rows
def pls_update(self, V, Y):
self.pls.fit(V, Y)
return self.pls
def setDoE_Mean(self, DoE_Mean):
self.DoE_Mean = DoE_Mean
def getDoE_Mean(self):
return self.DoE_Mean
def setPlsWindow(self, PlsWindow):
self.PlsWindow = PlsWindow
def getPlsWindow(self):
return self.PlsWindow
def plt_show1(self, n, y_act, y_prd):
plt.plot(np.arange(n), y_act, 'rx--', y_prd, 'bx--', lw=2, ms=5, mew=2)
plt.xticks(np.arange(0, n + 1, 50))
plt.xlabel('Run No.')
plt.ylabel('Actual and Predicted Response (y1)')
def plt_show2(self, n, y1, y2):
plt.figure()
plt.plot(np.arange(n), y1, 'bx-', y2, 'gx--', lw=2, ms=5, mew=2)
plt.xticks(np.arange(0, n + 1, 5))
plt.yticks(np.arange(-1.2, 1.3, 0.2))
plt.xlabel('Metrology Run No.(z)')
plt.ylabel('e(z)')
def DoE_Run(self, Z, M):
N = Z * M
DoE_Queue = []
for k in range(1, N + 1): # range(101) = [0, 1, 2, ..., 100])
idx_start, idx_end, result = self.sampling(k, self.sampling_up(),
self.sampling_vp(),
self.sampling_ep(),
True)
DoE_Queue.append(result)
initplsWindow = DoE_Queue.copy()
npPlsWindow = np.array(initplsWindow)
plsWindow = []
for i in range(len(npPlsWindow)):
plsWindow.append(npPlsWindow[i])
npDoE_Queue = np.array(plsWindow)
DoE_Mean = np.mean(npDoE_Queue, axis=0)
plsModelData = npDoE_Queue - DoE_Mean
V0 = plsModelData[:, 0:idx_start]
Y0 = plsModelData[:, idx_start:idx_end]
pls = self.pls_update(V0, Y0)
print('Init VM Coefficients: \n', pls.coef_)
y_prd = pls.predict(V0) + DoE_Mean[idx_start:idx_end]
y_act = npDoE_Queue[:, idx_start:idx_end]
print("Init DoE VM Mean squared error: %.3f" %
metrics.mean_squared_error(y_act[:, 0:1], y_prd[:, 0:1]))
print("Init DoE VM r2 score: %.3f" %
metrics.r2_score(y_act[:, 0:1], y_prd[:, 0:1]))
self.setDoE_Mean(DoE_Mean)
self.setPlsWindow(plsWindow)
# self.plt_show1(N, y_act[:,0:1], y_prd[:,0:1])
def VM_Run(self, lamda_PLS, Z, M):
N = Z * M
## V0, Y0 Mean Center
DoE_Mean = self.getDoE_Mean()
idx_end = len(DoE_Mean)
idx_start = idx_end - 2
meanVz = DoE_Mean[0:idx_start]
meanYz = DoE_Mean[idx_start:idx_end]
M_Queue = []
ez_Queue = []
ez_Queue.append([0, 0])
y_act = []
y_prd = []
plsWindow = self.getPlsWindow()
for z in np.arange(0, Z):
for k in np.arange(z * M + 1, ((z + 1) * M) + 1):
idx_start, idx_end, result = self.sampling(
k, self.sampling_up(), self.sampling_vp(),
self.sampling_ep(), False)
psiK = result[0:idx_start]
psiKStar = psiK - meanVz
y_predK = self.pls.predict(psiKStar.reshape(
1, idx_start)) + meanYz
rows = np.r_[result, y_predK.reshape(2, )]
M_Queue.append(rows)
y_prd.append(rows[idx_end:idx_end + 2])
y_act.append(rows[idx_start:idx_end])
del plsWindow[0:M]
ez = M_Queue[M - 1][idx_start:idx_end] - M_Queue[
M - 1][idx_end:idx_end + 2]
print("ez : ", ez)
ez_Queue.append(ez)
if z == 0:
ez = np.array([0, 0])
npM_Queue = np.array(M_Queue)
npM_Queue[0:M - 1,
0:idx_start] = lamda_PLS * npM_Queue[0:M - 1,
0:idx_start]
npM_Queue[0:M - 1, idx_start:idx_end] = lamda_PLS * (
npM_Queue[0:M - 1, idx_end:idx_end + 2] + 0.5 * ez)
npM_Queue = npM_Queue[:, 0:idx_end]
for i in range(M):
plsWindow.append(npM_Queue[i])
M_Mean = np.mean(plsWindow, axis=0)
meanVz = M_Mean[0:idx_start]
meanYz = M_Mean[idx_start:idx_end]
plsModelData = plsWindow - M_Mean
V = plsModelData[:, 0:idx_start]
Y = plsModelData[:, idx_start:idx_end]
self.pls_update(V, Y)
del M_Queue[0:M]
y_act = np.array(y_act)
y_prd = np.array(y_prd)
self.plt_show1(N, y_act[:, 0:1], y_prd[:, 0:1])
print("VM Mean squared error: %.3f" %
metrics.mean_squared_error(y_act[:, 0:1], y_prd[:, 0:1]))
print("VM r2 score: %.3f" %
metrics.r2_score(y_act[:, 0:1], y_prd[:, 0:1]))
ez_run = np.array(ez_Queue)
self.plt_show2(Z + 1, ez_run[:, 0:1], ez_run[:, 1:2])
plt.bar(np.arange(np.shape(X_train_prepro)[1]), pca_wild_b.components_[i])
if i == 0:
plt.ylabel('1st component')
elif i == 1:
plt.ylabel('2nd component')
else:
plt.ylabel('3rd component')
axis_c = plt.gca()
axis_c.set_xticklabels(wild_boar_ddbb['header'][3:],fontsize = 7)
axis_c.set_xticks(axis_c.get_xticks() + 0.5)
print "dentro del bucleeeeeeeeeee"
#Select the number of components using CV
#%%
##PLSR
pls_wild_b = PLSRegression(n_components = 3)
pls_wild_b.fit(X_train_prepro,Y_train)
X_train_pls_proj = pls_wild_b.transform(X_train_prepro)
print("loadings")
for i in range(pls_wild_b.n_components):
plt.figure()
plt.bar(np.arange(np.shape(X_train_prepro)[1]), pls_wild_b.x_loadings_[:,i])
if i == 0:
plt.ylabel('PLS 1st component')
elif i == 1:
plt.ylabel('PLS2nd component')
else:
plt.ylabel('PLS 3rd component')
axis_c = plt.gca()
axis_c.set_xticklabels(wild_boar_ddbb['header'][3:],fontsize = 7)
class Global_FWC_P3_Simulator:
def __init__(self, Tgt, A, d, C, F, seed):
self.pls = PLSRegression(n_components=6, scale=False, max_iter=50000, copy=True)
np.random.seed(seed)
self.Tgt = Tgt
self.A = A
self.d = d
self.C = C
self.F = F
def sampling_vp(self):
v1 = np.random.normal(-0.4, np.sqrt(0.2))
v2 = 2 * v1
v3 = np.random.uniform(0.2, 0.6)
v4 = 3 * v3
v5 = np.random.uniform(0, 0.4)
v = np.array([v1, v2, v3, v4, v5])
return v
def sampling_ep(self):
e1 = np.random.normal(0, np.sqrt(0.05))
e2 = np.random.normal(0, np.sqrt(0.1))
e = np.array([e1, e2])
return e
def sampling(self, k, uk=np.array([0, 0]), vp=np.array([0, 0, 0, 0, 0]), ep=np.array([0, 0]), fp=np.array([0, 0]), isInit=True):
u1 = uk[0]
u2 = uk[1]
u = uk
v1 = vp[0]
v2 = vp[1]
v3 = vp[2]
v4 = vp[3]
v5 = vp[4]
v = vp
e = ep
k1 = k
k2 = k
eta_k = np.array([[k1], [k2]])
psi = np.array([u1, u2, v1, v2, v3, v4, v5, k1, k2])
if fp is not None:
psi = np.r_[psi, fp]
f = fp
y = u.dot(self.A) + v.dot(self.C) + np.sum(eta_k * self.d, axis=0) + f.dot(self.F) + e
else:
y = u.dot(self.A) + v.dot(self.C) + np.sum(eta_k * self.d, axis=0) + e
rows = np.r_[psi, y]
idx_end = len(rows)
idx_start = idx_end - 2
return idx_start, idx_end, rows #y값의 시작과 끝 정보, 전체 값 정보
def pls_update(self, V, Y):
self.pls.fit(V, Y)
return self.pls
def setDoE_Mean(self, DoE_Mean):
self.DoE_Mean = DoE_Mean
def getDoE_Mean(self):
return self.DoE_Mean
def setPlsWindow(self, PlsWindow):
self.PlsWindow = PlsWindow
def getPlsWindow(self):
return self.PlsWindow
def plt_show1(self, n, y_act):
plt.figure()
plt.plot(np.arange(1, n + 1), y_act, 'bx--', lw=2, ms=10, mew=2)
plt.xticks(np.arange(0, n + 1, 20))
plt.xlabel('Run No.')
plt.ylabel('Actual Response (y2)')
def plt_show2(self, n, y_act):
plt.plot(np.arange(1, n + 1), y_act, 'ro-', lw=2, ms=5, mew=2)
plt.xticks(np.arange(0, n + 1, 20))
plt.xlabel('Run No.')
plt.ylabel('Actual Response (y2)')
def DoE_Run(self, lamda_PLS, dEWMA_Wgt1, dEWMA_Wgt2, Z, M, f, isR2R):
N = Z * M
I = np.identity(2)
dEWMA_Wgt1 = dEWMA_Wgt1 * I
dEWMA_Wgt2 = dEWMA_Wgt2 * I
DoE_Queue = []
sample_init_VP = []
sample_init_EP = []
for k in range(0, N + 1):
sample_init_VP.append(self.sampling_vp())
sample_init_EP.append(self.sampling_ep())
vp_next = sample_init_VP[0]
ep_next = sample_init_EP[0]
ep_next = np.array([0, 0])
for k in range(1, N + 1): # range(101) = [0, 1, 2, ..., 100])
if f is not None:
fp = f[k - 1,0:2]
p1_lamda_PLS = f[k - 1, 2:3]
fp = p1_lamda_PLS * fp
if k == 1:
uk_next = np.array([-51, -102]) # 계산 공식에 의해
Dk_prev = np.array([-0.24, 26.4])
Kd_prev = np.array([0.024, 0.07])
#Dk_prev = np.array([-0.2, 20])
#Kd_prev = np.array([-0.02, 1])
else:
fp = None
if k == 1:
uk_next = np.array([0, 0]) # 계산 공식에 의해
Dk_prev = np.array([0, 0])
Kd_prev = np.array([0, 0])
idx_start, idx_end, result = self.sampling(k, uk_next, vp_next, ep_next, fp, True)
npResult = np.array(result)
#================================== initVM-R2R Control =====================================
uk = npResult[0:2]
yk = npResult[idx_start:idx_end]
Dk = (yk - uk.dot(self.A)).dot(dEWMA_Wgt1) + Dk_prev.dot(I - dEWMA_Wgt1)
Kd = (yk - uk.dot(self.A) - Dk_prev).dot(dEWMA_Wgt2) + Kd_prev.dot(I - dEWMA_Wgt2)
# Dk = (yk - uk.dot(self.A) - fp.dot(self.F)).dot(dEWMA_Wgt1) + Dk_prev.dot(I - dEWMA_Wgt1)
# Kd = (yk - uk.dot(self.A) - fp.dot(self.F) - Dk_prev).dot(dEWMA_Wgt2) + Kd_prev.dot(I - dEWMA_Wgt2)
Kd_prev = Kd
Dk_prev = Dk
if isR2R == True:
uk_next = (self.Tgt - Dk - Kd).dot(np.linalg.inv(self.A))
vp_next = sample_init_VP[k]
else:
if k % M == 0:
uk_next = (self.Tgt - Dk - Kd).dot(np.linalg.inv(self.A))
vp_next = sample_init_VP[k]
ep_next = sample_init_EP[k]
ep_next = np.array([0, 0])
DoE_Queue.append(result)
initplsWindow = DoE_Queue.copy()
npPlsWindow = np.array(initplsWindow)
plsWindow = []
#np.savetxt("output/npPlsWindow1.csv", npPlsWindow, delimiter=",", fmt="%s")
if f is not None:
for k in range(0, N): # range(101) = [0, 1, 2, ..., 100])
p1_lamda_PLS = f[k, 2:3]
if (k + 1) % M != 0:
npPlsWindow[k, idx_start - 2:idx_start] = p1_lamda_PLS * npPlsWindow[k, idx_start - 2:idx_start]
for z in np.arange(0, Z):
npPlsWindow[z * M:(z + 1) * M - 1, 0:idx_start] = lamda_PLS * npPlsWindow[z * M:(z + 1) * M - 1, 0:idx_start]
npPlsWindow[z * M:(z + 1) * M - 1, idx_start:idx_end] = lamda_PLS * (npPlsWindow[z * M:(z + 1) * M - 1, idx_start:idx_end])
for i in range(len(npPlsWindow)):
plsWindow.append(npPlsWindow[i])
#np.savetxt("output/npPlsWindow2.csv", npPlsWindow, delimiter=",", fmt="%s")
npDoE_Queue = np.array(plsWindow)
DoE_Mean = np.mean(npDoE_Queue, axis=0)
plsModelData = npDoE_Queue - DoE_Mean
V0 = plsModelData[:, 0:idx_start]
Y0 = plsModelData[:, idx_start:idx_end]
pls = self.pls_update(V0, Y0)
# print('Init VM Coefficients: \n', pls.coef_)
y_pred = pls.predict(V0) + DoE_Mean[idx_start:idx_end]
y_act = npDoE_Queue[:, idx_start:idx_end]
self.setDoE_Mean(DoE_Mean)
self.setPlsWindow(plsWindow)
# self.plt_show2(N, y_act[:, 1:2])
# self.plt_show1(N, y_pred[:, 1:2])
def VM_Run(self, lamda_PLS, dEWMA_Wgt1, dEWMA_Wgt2, Z, M, f, isR2R):
N = Z * M
I = np.identity(2)
dEWMA_Wgt1 = dEWMA_Wgt1 * I
dEWMA_Wgt2 = dEWMA_Wgt2 * I
## V0, Y0 Mean Center
DoE_Mean = self.getDoE_Mean()
idx_end = len(DoE_Mean)
idx_start = idx_end - 2
meanVz = DoE_Mean[0:idx_start]
meanYz = DoE_Mean[idx_start:idx_end]
yk = np.array([0, 0])
Dk_prev = np.array([-0.24, 26.4]) #10번째 run시 값
Kd_prev = np.array([0.024, 0.07]) #10번째 run시 값
# Dk = np.array([0, 0])
# Kd = np.array([0, 0])
uk_next = np.array([-51, -102]) #계산 공식에 의해
M_Queue = []
ez_Queue = []
ez_Queue.append([0, 0])
y_act = []
y_pred = []
VM_Output = []
plsWindow = self.getPlsWindow()
sample_vm_VP = []
sample_vm_EP = []
for k in range(0, N + 1):
sample_vm_VP.append(self.sampling_vp())
sample_vm_EP.append(self.sampling_ep())
vp_next = sample_vm_VP[0]
ep_next = sample_vm_EP[0]
for z in np.arange(0, Z):
for k in np.arange(z * M + 1, ((z + 1) * M) + 1):
if f is not None:
fp = f[k - 1, 0:2]
else:
fp = None
if k == 1:
uk_next = np.array([0, 0]) # 계산 공식에 의해
Dk_prev = np.array([0, 0])
Kd_prev = np.array([0, 0])
# y값의 시작과 끝 정보, 전체 값 정보
idx_start, idx_end, result = self.sampling(k, uk_next, vp_next, ep_next, fp, False)
psiK = result[0:idx_start] # 파라미터 값들
psiKStar = psiK - meanVz # 파라미터 값들 평균 마이너스
y_predK = self.pls.predict(psiKStar.reshape(1, idx_start)) + meanYz # 예측값 + 평균
rows = np.r_[result, y_predK.reshape(2, )] #실제값 + 2개 예측값을 rows로, run 10일때가 actual, vm 차이 비교
y_pred.append(rows[idx_end:idx_end + 2]) #예측 값 ==> 10개의 VM 값인데..
y_act.append(rows[idx_start:idx_end]) #실제 값 ==> 시뮬레이션의 실제 값 인데..
# ================================== VM + R2R Control =====================================
if k % M != 0: #예측 값
yk = rows[idx_end:idx_end + 2]
else:
yk = rows[idx_start:idx_end] #실제 값
e1 = np.absolute(rows[idx_start + 1:idx_end] - rows[idx_end + 1:idx_end + 2])
uk = psiK[0:2]
Dk = (yk - uk.dot(self.A)).dot(dEWMA_Wgt1) + Dk_prev.dot(I - dEWMA_Wgt1)
Kd = (yk - uk.dot(self.A) - Dk_prev).dot(dEWMA_Wgt2) + Kd_prev.dot(I - dEWMA_Wgt2)
Kd_prev = Kd
Dk_prev = Dk
if isR2R == True:
uk_next = (self.Tgt - Dk - Kd).dot(np.linalg.inv(self.A))
vp_next = sample_vm_VP[k]
uk_next = uk_next.reshape(2, )
ep_next = sample_vm_EP[k]
M_Queue.append(rows) # M_Queue에 rows의 정보
del plsWindow[0:M] #Queue의 가장 처음 Run 10이 없어진다.
if isR2R == False:
uk_next = (self.Tgt - Dk - Kd).dot(np.linalg.inv(self.A))
vp_next = sample_vm_VP[k]
# 여기서 부터는 모델 업데이트를 위한 과정이다. 이미 VM은 rows 정보에 있지만, 가중치를 반영해 준다.
if z == 0:
ez = 0
npM_Queue = np.array(M_Queue) #parameter + 실제값 + 2개 예측값을 rows로, run 10일 때가 actual, vm 차이 비교
# M은 Run 주기이며, 10, M-1은 run = 10을 제외한 VM들이겠지
# idx_start 까지는 파라미터 값들로 lamda_PLS 0.1을 반영하겠다는 의미지..
for i in range(M): #VM_Output 구한다. lamda_pls 가중치를 반영하지 않는다.
if i == M - 1:
temp = npM_Queue[i:i + 1, idx_start:idx_end]
else:
temp = npM_Queue[i:i + 1, idx_end:idx_end + 2]
VM_Output.append(np.array([temp[0, 0], temp[0, 1]]))
# emax = 5
# lamda_PLS = 1 - e1/emax
# if lamda_PLS <= 0:
# lamda_PLS = 0.1
#
# print("e1 : ", e1, "P2 lamda_PLS : ", lamda_PLS)
if f is not None:
p1_lamda_PLS = f[k - 1, 2:3]
npM_Queue[0:M - 1, idx_start - 2:idx_start] = p1_lamda_PLS * npM_Queue[0:M - 1, idx_start - 2:idx_start]
#np.savetxt("output/npM_Queue2.csv", npM_Queue, delimiter=",", fmt="%s")
npM_Queue[0:M - 1, 0:idx_start] = lamda_PLS * npM_Queue[0:M - 1, 0:idx_start]
# idx_start:idx_end는 실제 값에 VM 값들의 조정을 통해 모델을 위해 VM의 정보를 업데이트 한다.
npM_Queue[0:M - 1, idx_start:idx_end] = lamda_PLS * (npM_Queue[0:M - 1, idx_end:idx_end + 2] + 0.5 * ez)
#npM_Queue[0:M - 1, idx_start:idx_end] = lamda_PLS * (npM_Queue[0:M - 1, idx_end:idx_end + 2]) # 0.5 * ez 반영안할시
npM_Queue = npM_Queue[:, 0:idx_end] #여기에는 VM + Actual 실제값들이 저장되어 있다.
# for i in range(M): #VM_Output 구한다. lamda_pls 가중치를 반영하지 않는다.
# temp = npM_Queue[i:i + 1, idx_start:idx_end]
# VM_Output.append(np.array([temp[0, 0], temp[0, 1]]))
for i in range(M):
plsWindow.append(npM_Queue[i]) #전체 Queue에 넣는다.
M_Mean = np.mean(plsWindow, axis=0) #Queue의 평균을 구한다.
meanVz = M_Mean[0:idx_start] #파라미터 평균
meanYz = M_Mean[idx_start:idx_end] #y값 run시마다 vm 9개(lamda_pla 0.1) 실제 1개(lamda_pls 1) 평균
plsModelData = plsWindow - M_Mean #Queue의 평균 제외
V = plsModelData[:, 0:idx_start] #모델을 위한 파라미터
Y = plsModelData[:, idx_start:idx_end] #모델을 위한 y값
self.pls_update(V, Y)
ez = M_Queue[M - 1][idx_start:idx_end] - M_Queue[M - 1][idx_end:idx_end + 2]
ez_Queue.append(ez)
# print("ez : ", ez)
del M_Queue[0:M]
y_act = np.array(y_act)
y_pred = np.array(y_pred)
# print("VM Mean squared error: %.3f" % metrics.mean_squared_error(y_act[:,1:2], y_pred[:,1:2]))
# print("VM r2 score: %.3f" % metrics.r2_score(y_act[:,1:2], y_pred[:,1:2]))
return y_act[:, 1:2]
X = dataset["data"]
y = dataset["target"]
# Center each feature and scale the variance to be unitary
X = preprocessing.scale(X)
# Compute the variance for each column
print(numpy.var(X, 0).sum())
# Now use PCA using 3 components
pca = PCA(3)
X2 = pca.fit_transform(X)
print(numpy.var(X2, 0).sum())
pls = PLSRegression(3)
pls.fit(X, y)
X2 = pls.transform(X)
print(numpy.var(X2, 0).sum())
# Make predictions using an SVM with PCA and PLS
pca_error = 0
pls_error = 0
n_folds = 10
svc = LinearSVC()
for train_inds, test_inds in KFold(X.shape[0], n_folds=n_folds):
X_train, X_test = X[train_inds], X[test_inds]
y_train, y_test = y[train_inds], y[test_inds]
def get_best_estimator():
"""Hyperparameter optimization"""
df = load_data()
Y = df['Fitness']
X = df[['Variants']]
features = FeatureUnion([
#('one_hot_encoder', OneHotEncoder()),
#('one_hot_pair_encoder', OneHotPairEncoder()),
#('pybiomed_encoder', PyBioMedEncoder()),
('aaindex_encoder', AAIndexEncoder())
])
print('*' * 40)
print('Extracting features...')
print('*' * 40)
start = timer()
X = features.transform(X)
end = timer()
print('Finished in: {}'.format(end - start))
num_rows, num_cols = X.shape
assert num_rows == len(df)
print('Got {} features'.format(num_cols))
# TODO include this in pipeilne
imp = SimpleImputer(missing_values=np.nan, strategy='mean')
imp.fit(X)
X = imp.transform(X)
assert not pd.DataFrame(X).isna().any().any()
X = FFTEncoder().fit_transform(X)
import ipdb
ipdb.set_trace()
n_features_options = [int(num_cols * ratio) for ratio in N_FEATURES_RATIOS]
print('n_features_options:', n_features_options)
feature_reduction_grid = [
{
'reduce': [
#PCA(),
NMF()
],
'reduce__n_components': n_features_options,
},
#{
# 'reduce': [SelectKBest()],
# 'reduce__score_func': [
# f_regression,
# mutual_info_regression
# ],
# 'reduce__k': n_features_options,
#},
]
# Random forest features
# Number of trees
n_estimators = [int(x) for x in np.linspace(start=200, stop=2000, num=10)]
# Number of features to consider at every split
max_features = ['auto', 'sqrt']
# Maximum number of levels in tree
max_depth = [int(x) for x in np.linspace(10, 110, num=11)]
max_depth.append(None)
# Minimum number of samples required to split a node
min_samples_split = [2, 5, 10]
# Minimum number of samples required at each leaf node
min_samples_leaf = [1, 2, 4]
# Method of selecting samples for training each tree
bootstrap = [True, False]
# TODO: search over more params
regression_grid = [
#{
# 'regress': [
# #KNeighborsRegressor(),
# #linear_model.ARDRegression(),
# #linear_model.BayesianRidge(),
# #linear_model.ElasticNet(),
# #linear_model.LassoLars(),
# #linear_model.LinearRegression(),
# #linear_model.Ridge(),
# #linear_model.SGDRegressor(),
# #tree.DecisionTreeRegressor(),
# #ensemble.AdaBoostRegressor(),
# #ensemble.BaggingRegressor(),
# #ensemble.GradientBoostingRegressor(),
# ]
#},
#{
# 'regress': [ensemble.RandomForestRegressor()],
# #'regress__n_estimators': n_estimators,
# #'regress__max_features': max_features,
# #'regress__max_depth': max_depth,
# #'regress__min_samples_split': min_samples_split,
# #'regress__min_samples_leaf': min_samples_leaf,
# #'regress__bootstrap': bootstrap
#},
{
'regress': [PLSRegression()]
}
#{
# 'regress': [svm.NuSVR()],
# 'regress__C': [1, 10, 100, 1000],
# 'regress__kernel': ['rbf', 'linear', 'poly'],
#},
#{
# 'regress': [svm.LinearSVR()],
# 'regress__C': [1, 10, 100, 1000]
#}
#{
# 'regress': [neural_network.MLPRegressor()],
# 'regress__hidden_layer_sizes': [(100,)]
#},
]
pipeline = Pipeline(
[
#('fft', FFTEncoder()),
#('reduce', DummyEstimator()),
('regress', DummyEstimator())
],
#memory=memory
)
grid_steps = [
#feature_reduction_grid,
regression_grid
]
combined_grids = get_combined_grids(grid_steps)
print('combined_grids:')
pprint(combined_grids)
kfold = KFold(n_splits=NUM_FOLDS or num_rows, random_state=0)
search = GridSearchCV(pipeline,
combined_grids,
error_score=np.nan,
verbose=5,
n_jobs=-1,
cv=kfold)
print('*' * 40)
print('Searching')
print('*' * 40)
start = timer()
search.fit(X, Y)
end = timer()
print('Finished in: {}'.format(end - start))
best_estimator = search.best_estimator_
best_params = search.best_params_
best_score = search.best_score_
best_index = search.best_index_
best_std = search.cv_results_['std_test_score'][best_index]
print('best_estimator:', best_estimator)
print('best_params:', best_params)
print('best_score:', best_score)
print('best_std:', best_std)
return Pipeline([('features', features), ('estimator', best_estimator)])
