y_test = testLabelArr
# clf = ELMClassifier(n_hidden=100, activation_func='gaussian', alpha=0.0, random_state=1)
# clf = ELMClassifier(n_hidden=100 000, activation_func='hardlim', alpha=1.0, random_state=0)
clf = ELMClassifier(n_hidden=100,
activation_func='hardlim',
alpha=1.0,
random_state=1)
for iter in range(2):
# y_target = mat(labelArr)
y_target = sign(mat(labelArr) - mat(y_train_pred))
print y_target
print y_train_pred
# print shape(y_target)
# iterate over classifiers
clf.fit(datArr, list(array(y_target).reshape(-1)))
y_train_pred = clf.predict(datArr)
y_pred = clf.predict(testDatArr)
acc = calcAccurary(sign(y_train_pred), labelArr)
print "Training accurarcy:\n"
print acc
acc = calcAccurary(sign(y_pred), y_test)
print acc
y_preds = y_preds + LRATE * y_pred
print mat(y_preds)
acc = calcAccurary(sign(y_preds), y_test)
# print y_preds
print "Final score:\n"
print acc
"""
for name, clf in zip(names, classifiers):
clf.fit(X_train, y_train)
def run(self):
rawDataset = open('../../selectedUserMore', 'r')
# Value 0 represents max number of apps => It must be set
maxValuesByFeature = [6, 4, 2, 2, 2, 0, 4, 5, 3]
datasetFeatures = []
datasetLabels = []
maxApps = self.findMaxNumberOfApps(rawDataset)
# Setting max number of apps
maxValuesByFeature[5] = maxApps
rawDataset.seek(0)
for line in rawDataset:
tuples = line.split(';')
for t in range(len(tuples)):
tup = tuples[t]
vals = tup.split(',')
label = vals[0]
features = vals[1].split('_')
datasetFeatures.append(features)
datasetLabels.append(label)
rawDataset.close()
nTuples = len(datasetLabels)
nTrain = int(0.7 * nTuples)
nMaxHidden = 3005
totalAccuracy = 0.0
trainSetFeatures = np.array(datasetFeatures[:nTrain], dtype=np.int)
trainSetLabels = np.array(datasetLabels[:nTrain], dtype=np.int)
testSetFeatures = np.array(datasetFeatures[nTrain:], dtype=np.int)
testSetLabels = np.array(datasetLabels[nTrain:], dtype=np.int)
# for x in range(200,nMaxHidden):
ffsn = ELMClassifier(n_hidden=1000, activation_func='sine')
ffsn.fit(trainSetFeatures, trainSetLabels)
results = ffsn.predict(testSetFeatures)
nPredictedCorrectly = 0
for test in range(len(testSetLabels)):
realValue = testSetLabels[test]
predictedValue = results[test]
if (int(realValue) == int(predictedValue)):
nPredictedCorrectly += 1
# print "Real: " + str(realValue) + " / Predicted: " + str(predictedValue)
totalTests = nTuples - nTrain
accuracy = float(nPredictedCorrectly) / totalTests
# print "N Hidden: " + str(x) + " / Accuracy: " + str(accuracy)
print "Accuracy: " + str(accuracy)
totalAccuracy += accuracy
train_label = train_data.map(lambda x: x[class_index]).collect()
print train_label
############### ELM 训练 #############
train_array = train_.collect()
print "train_array_num:", len(train_array)
#训练分类器
elmc.fit(train_array, train_label)
pred_class = elmc.predict_class(test_array)
#分类精度
soc = accuracy_score(pred_class, test_label)
print "test_soc:", soc
#软最大化函数
ruanzuida = lambda x: np.exp(x) / sum(np.exp(x))
#对无类标的数据集进行预测 每个样例的到一个向量 然后进行软最大化处理
ruanhua_result = unlabel_.flatMap(lambda x: elmc.predict(x)).map(
lambda x: max(ruanzuida(x)))
print ruanhua_result.collect()
#对软最大化处理后的数据进行过滤 为了使排序的样例变少 所以先进行初步过滤
filter_result = ruanhua_result.filter(
lambda x: x < frist_filter).collect()
#对过滤后的样例排序 得到第10小的数值作为阈值
try:
yuzhi = sorted(filter_result)[select_num]
except IndexError:
print "select_num", (select_num / 2)
yuzhi = sorted(filter_result)[int(select_num // 2)]
continue
finally:
pass
# Shuffle target and data shfl_data = loaded_data.data[rand_perm] shfl_target = loaded_data.target[rand_perm] # divide train data and test data offset = np.floor(data_size*0.8) # 80 percent data as training train_data = shfl_data[:offset,:] train_target = shfl_target[:offset] test_data = shfl_data[offset+1:,:] test_target = shfl_target[offset+1:] # Build ELM Classifier using default value elm_classifier = ELMClassifier() # Train ELM elm_classifier.fit(train_data,train_target) predicted_class = elm_classifier.predict(test_data) error_elm = 1.0 - elm_classifier.score(test_data, test_target) print 'Predicted Class\t\t', predicted_class print 'Original Class\t\t', test_target print 'Error Score\t\t\t', error_elm print '' #print '##############################################################\nSINGLE GEN-ELM RESULT\n##############################################################' from elm import GenELMClassifier ## Build GEN_ELM Classifier using default value #genelm_classifier = GenELMClassifier() ## Train the ELM #genelm_classifier.fit(train_data,train_target) # #predicted_class = genelm_classifier.predict(test_data) #error_elm = 1.0 - genelm_classifier.score(test_data, test_target)
print "\nTime: %.3f secs" % (time() - start_time)
print "Test Min: %.3f Mean: %.3f Max: %.3f SD: %.3f" % (
min(test_res), np.mean(test_res), max(test_res), np.std(test_res))
print "Train Min: %.3f Mean: %.3f Max: %.3f SD: %.3f" % (
min(train_res), np.mean(train_res), max(train_res), np.std(train_res))
print
return (train_res, test_res)
stdsc = StandardScaler()
iris = load_iris()
irx, iry = stdsc.fit_transform(iris.data), iris.target
irx_train, irx_test, iry_train, iry_test = train_test_split(irx,
iry,
test_size=0.2)
srh = SimpleRandomHiddenLayer(activation_args='sigmoid', n_hidden=500)
elmc = ELMClassifier(hidden_layer=srh)
# elmc = ELMClassifier(SimpleRandomHiddenLayer(activation_func='sigmoid'))
# print "SimpleRandomHiddenLayer(activation_func='sigmoid')"
# tr,ts = res_dist(irx, iry, elmc, n_runs=100, random_state=0)
# plt.hist(tr),plt.hist(tr)
# plt.show()
elmc.fit(irx_train, iry_train)
r = elmc.predict(irx_test)
print r
res = elmc.score(irx_test, iry_test)
print res
def simulate(trn, tst):
start = time.time()
b_tp = b_fp = b_tn = b_fn = 0
s_tp = s_fp = s_tn = s_fn = 0
b_min = s_min = 1000000
b_max = s_max = 0
b_money = s_money = 0
b_money_vec = [0]
s_money_vec = [0]
b_gain = s_gain = 0
b_loss = s_loss = 0
b_draw = s_draw = 0
b_gain_vec = []
s_gain_vec = []
b_loss_vec = []
s_loss_vec = []
b_max_drawdown = s_max_drawdown = 0
b_pos = s_pos = False
time_vec = []
aux_ii = len(tst) - 1
for t, val in enumerate(tst):
start_i = time.time()
if t == 201:
continue
if t == 0:
tst[0, 5] = id.log_return(tst[0, 0], tst[0, 3], trn[-1, 0], trn[-1,
3])
else:
tst[t, 5] = id.log_return(tst[t, 0], tst[t, 3], trn[t - 1, 0],
trn[t - 1, 3])
tst[t, 6] = mnm.get(val[5])
tst[t, 7] = obv.get_obv(val[3], val[4])
aux = bbs.sma(val[3])
if aux is not None:
tst[t, 8], tst[t, 9] = aux
aux_9 = m_9.ema(val[3])
aux12 = m12.ema(val[3])
aux26 = m26.ema(val[3])
tst[t, 10] = aux12 - aux26
tst[t, 11] = tst[t, 10] - aux_9
aux = trn[-1000:]
aux_i = [(i[1] - mn[i[0]]) * mx[i[0]] for i in enumerate(tst[t, :12])]
# aux_j = trn[-1000:, :]
b_elm = ELMClassifier(random_state=0,
n_hidden=200,
activation_func='sigmoid',
alpha=0.0)
b_elm.fit(aux[:, :12], aux[:, 12])
b_res = b_elm.predict([aux_i[:12]])
s_elm = ELMClassifier(random_state=0,
n_hidden=200,
activation_func='sigmoid',
alpha=0.0)
s_elm.fit(aux[:, :12], aux[:, 13])
s_res = s_elm.predict([aux_i[:12]])
if b_res == 1.0:
if val[12] == 1.0:
b_tp += 1
else:
b_fp += 1
if not b_pos:
# Entra
b_money -= val[3]
b_pos = True
else:
if val[12] == 0.0:
b_tn += 1
else:
b_fn += 1
if b_pos:
# Sai
b_money += val[3]
b_pos = False
if b_money < b_money_vec[-1]:
b_loss += 1
b_loss_vec.append(b_money_vec[-1] - b_money)
elif b_money > b_money_vec[-1]:
b_gain += 1
b_gain_vec.append(b_money - b_money_vec[-1])
else:
b_draw += 1
if val[14] == 1.0:
# Sai
b_money += val[3]
b_pos = False
if b_money < b_money_vec[-1]:
b_loss += 1
b_loss_vec.append(b_money_vec[-1] - b_money)
elif b_money > b_money_vec[-1]:
b_gain += 1
b_gain_vec.append(b_money - b_money_vec[-1])
else:
b_draw += 1
if b_pos:
b_money_vec.append(b_money_vec[-1])
else:
b_money_vec.append(b_money)
if b_money > b_max:
b_max = b_money
if b_money < b_min:
b_min = b_money
if s_res == 1.0:
if val[13] == 1.0:
s_tp += 1
else:
s_fp += 1
if not s_pos:
# Entra
s_money += val[3]
s_pos = True
else:
if val[13] == 0.0:
s_tn += 1
else:
s_fn += 1
if s_pos:
# Sai
s_money -= val[3]
s_pos = False
if s_money < s_money_vec[-1]:
s_loss += 1
s_loss_vec.append(s_money_vec[-1] - s_money)
elif s_money > s_money_vec[-1]:
s_gain += 1
s_gain_vec.append(s_money - s_money_vec[-1])
else:
s_draw += 1
if val[14] == 1.0:
# Sai
s_money -= val[3]
s_pos = False
if s_money < s_money_vec[-1]:
s_loss += 1
s_loss_vec.append(s_money_vec[-1] - s_money)
elif s_money > s_money_vec[-1]:
s_gain += 1
s_gain_vec.append(s_money - s_money_vec[-1])
else:
s_draw += 1
if s_pos:
s_money_vec.append(s_money_vec[-1])
else:
s_money_vec.append(s_money)
if s_money > s_max:
s_max = s_money
if s_money < s_min:
s_min = s_money
# print(aux_i + list(tst[t, 12:]))
trn = np.append(trn, [aux_i + list(tst[t, 12:])], axis=0)
time_vec.append(time.time() - start_i)
sys.stdout.write('\r' + '%6d / %d' % (t, aux_ii) + '\033[K')
sys.stdout.write('\r' + '>> %6.2f: Simulation Done!\n\n' %
(time.time() - start) + '\033[K')
print('#### ' + sys.argv[1] + ' ####')
print('Tempo médio: %f' % np.mean(time_vec))
print('Final : %5.5f | %5.5f' % (b_money, s_money))
# print('Final : %5.5f | %5.5f' % (b_money_vec[-1], s_money_vec[-1]))
print('Minimo : %5.5f | %5.5f' % (b_min, s_min))
print('Maximo : %5.5f | %5.5f' % (b_max, s_max))
print('Ganho qtd : %10d | %10d' % (b_gain, s_gain))
print('Perda qtd : %10d | %10d' % (b_loss, s_loss))
print('Empate qtd : %10d | %10d' % (b_draw, s_draw))
print('Ganho medio: %5.5f | %5.5f' %
(np.mean(b_gain_vec), np.mean(s_gain_vec)))
print('Perda media: %5.5f | %5.5f' %
(np.mean(b_loss_vec), np.mean(s_loss_vec)))
print('TP : %10d | %10d' % (b_tp, s_tp))
print('FP : %10d | %10d' % (b_fp, s_fp))
print('TN : %10d | %10d' % (b_tn, s_tn))
print('FN : %10d | %10d' % (b_fn, s_fn))
plot(b_money_vec, s_money_vec, sys.argv[1], tst[:, 3])
reg = reg.fit(X, y)
cla = cla.fit(X, y)
#Read testing data from testing.csv
testData = pd.read_csv('testing.csv')
pdt = np.zeros(len(testData) - 3)
pdt2 = np.zeros(len(testData) - 3)
for i in range(len(testData) - 3):
ftr1 = float(testData['close'][i]) - float(testData['open'][i])
ftr2 = float(testData['close'][i+1]) - float(testData['open'][i+1])
ftr3 = float(testData['close'][i+2]) - float(testData['open'][i+2])
ftr4 = float(testData['highest'][i] - float(testData['lowest'][i]))
ftr5 = float(testData['highest'][i+1] - float(testData['lowest'][i+1]))
ftr6 = float(testData['highest'][i+2] - float(testData['lowest'][i+2]))
pdt[i] = reg.predict([ftr1, ftr2, ftr3, ftr4, ftr5, ftr6])
pdt2[i] = cla.predict([ftr1, ftr2, ftr3, ftr4, ftr5, ftr6])
#print pdt
print pdt2
#Verify the testing data
#The 0.01 is the threshold to determine rise or full
correct = 0
for i in range(len(testData) - 3):
if testData['changerange'][i] >=0 and float(pdt[i]) >= 0.09:
correct += 1
elif testData['changerange'][i] < 0 and float(pdt[i]) < 0.09:
correct += 1
#calculate the result
#print 'The accuracy is', float(correct / float(len(testData))) * 100, 'percent'
# testLabelArr = array(testLabelMat) # print targets[:10] y_train_pred = zeros(shape(labelArr)); y_preds = zeros(shape(testLabelArr)) X_test = testDatArr; y_test = testLabelArr # clf = ELMClassifier(n_hidden=100, activation_func='gaussian', alpha=0.0, random_state=1) # clf = ELMClassifier(n_hidden=100 000, activation_func='hardlim', alpha=1.0, random_state=0) clf = ELMClassifier(n_hidden=100, activation_func='hardlim', alpha=1.0, random_state=1) for iter in range(2): # y_target = mat(labelArr) y_target = sign(mat(labelArr) - mat(y_train_pred)); print y_target print y_train_pred # print shape(y_target) # iterate over classifiers clf.fit(datArr, list(array(y_target).reshape(-1))) y_train_pred = clf.predict(datArr) y_pred = clf.predict(testDatArr) acc = calcAccurary(sign(y_train_pred), labelArr) print "Training accurarcy:\n" print acc acc = calcAccurary(sign(y_pred), y_test) print acc y_preds = y_preds + LRATE * y_pred print mat(y_preds) acc = calcAccurary(sign(y_preds), y_test) # print y_preds print "Final score:\n" print acc """ for name, clf in zip(names, classifiers): clf.fit(X_train, y_train)