def train(model_id,train_x,train_y,valid_x,valid_y,test_x):
train_x,train_y=shuffle(train_x,train_y)
random_state=random.randint(0, 1000000)
print('random state: {state}'.format(state=random_state))
clf = RandomForestClassifier(bootstrap=False, class_weight=None,
criterion='entropy', max_depth=29008, max_features=36,
max_leaf_nodes=None, min_samples_leaf=5, min_samples_split=3,
min_weight_fraction_leaf=0.0, n_estimators=4494, n_jobs=8,
oob_score=False, random_state=979271, verbose=0,
warm_start=False)
clf.fit(train_x, train_y)
ccv = CalibratedClassifierCV(base_estimator=clf,method="sigmoid",cv="prefit")
ccv.fit(valid_x,valid_y)
valid_predictions = ccv.predict_proba(valid_x)
test_predictions= ccv.predict_proba(test_x)
loss = test(valid_y,valid_predictions,True)
if loss<0.52:
data.saveData(valid_predictions,"../valid_results/valid_"+str(model_id)+".csv")
data.saveData(test_predictions,"../results/results_"+str(model_id)+".csv")
def train(model_id,train_x,train_y,valid_x,valid_y,test_x):
train_x,train_y=shuffle(train_x,train_y)
random_state=random.randint(0, 1000000)
rf = RandomForestClassifier(n_jobs=8)
param_dist = {
"n_estimators":sp_randint(100,300),
"criterion": ["gini"],
#"max_depth": sp_randint(3, 10000),
#"min_samples_split": sp_randint(1, 300),
#"min_samples_leaf": sp_randint(1, 300),
"max_features": sp_randint(10, 26),
"bootstrap": [True, False],
'random_state':sp_randint(1, 1000000),
}
clf = RandomizedSearchCV(rf, param_distributions=param_dist,
n_iter=50,cv=10,scoring='roc_auc')
clf.fit(train_x, train_y)
valid_predictions = clf.predict_proba(valid_x)[:, 1]
test_predictions= clf.predict_proba(test_x)[:, 1]
loss = roc_auc_score(valid_y,valid_predictions)
print('loss:')
print(loss)
print(clf.best_estimator_)
data.saveData(valid_id,valid_predictions,"./valid_results/valid_"+str(model_id)+".csv")
data.saveData(test_id,test_predictions,"./results/results_"+str(model_id)+".csv")
def createSignalFromWav(file_wav_name, file_data_name):
try:
with wave.open(file_wav_name, mode='rb') as wav:
sample_rate = wav.getframerate()
n_frame = wav.getnframes()
# Duração do áudio
duration = n_frame / sample_rate
# Buffer de Dados
dataBuffer = wav.readframes(n_frame)
# Descompacta os bytes em pacotes de 2 bytes (short)
DataGenerator = struct.unpack("%ih" % (n_frame * wav.getnchannels()), dataBuffer)
# Converte 16bits wav para float, entre -1.0 e 1.0
DataGenerator = [float(value) / pow(2, 15) for value in DataGenerator]
# Cria nosso sinal e salva em um file
dataSignal = DataSignal(sample_rate, duration, len(DataGenerator), DataGenerator)
data.saveData(file_data_name, dataSignal)
return True
except:
return False
def saveSettings(self):
settings.settings.setValue("data/MaximumURLLength", self.maximumURLLength.value())
#settings.settings.setValue("data/MaximumCacheSize", self.maximumCacheSize.value())
settings.settings.setValue("data/RememberHistory", self.rememberHistoryToggle.isChecked())
settings.settings.setValue("network/GeolocationEnabled", self.geolocationToggle.isChecked())
data.geolocation_whitelist = [self.geolocationWhitelist.item(authority).text() for authority in range(0, self.geolocationWhitelist.count())]
data.geolocation_blacklist = [self.geolocationBlacklist.item(authority).text() for authority in range(0, self.geolocationBlacklist.count())]
data.saveData()
def prepareQuit():
try: os.remove(settings.crash_file)
except: pass
common.downloadManager.saveSession()
saveSession()
settings.settings.hardSync()
data.saveData()
data.data.hardSync()
filtering.adblock_filter_loader.quit()
filtering.adblock_filter_loader.wait()
server_thread.httpd.shutdown()
server_thread.quit()
server_thread.wait()
def trainrf(model_id,train_x,train_y,valid_x,valid_y,test_x):
train_x,train_y=shuffle(train_x,train_y)
random_state=random.randint(0, 1000000)
print('random state: {state}'.format(state=random_state))
clf = RandomForestClassifier(n_estimators=random.randint(50,5000),
criterion='gini',
max_depth=random.randint(10,1000),
min_samples_split=random.randint(2,50),
min_samples_leaf=random.randint(1,10),
min_weight_fraction_leaf=random.uniform(0.0,0.5),
max_features=random.uniform(0.1,1.0),
max_leaf_nodes=random.randint(1,10),
bootstrap=False,
oob_score=False,
n_jobs=30,
random_state=random_state,
verbose=0,
warm_start=True,
class_weight=None
)
clf.fit(train_x, train_y)
valid_predictions1 = clf.predict_proba(valid_x)
test_predictions1= clf.predict_proba(test_x)
t1 = test(valid_y,valid_predictions1)
ccv = CalibratedClassifierCV(base_estimator=clf,method="sigmoid",cv='prefit')
ccv.fit(valid_x,valid_y)
valid_predictions2 = ccv.predict_proba(valid_x)
test_predictions2= ccv.predict_proba(test_x)
t2 = test(valid_y,valid_predictions2)
if t2<t1:
valid_predictions=valid_predictions2
test_predictions=test_predictions2
t=t2
else:
valid_predictions=valid_predictions1
test_predictions=test_predictions1
t=t1
if t < 0.450:
data.saveData(valid_predictions,"../valid_results/valid_"+str(model_id)+".csv")
data.saveData(test_predictions,"../results/results_"+str(model_id)+".csv")
def prepareQuit():
try:
os.remove(settings.crash_file)
except:
pass
common.downloadManager.saveSession()
saveSession()
settings.settings.hardSync()
data.saveData()
data.data.hardSync()
filtering.adblock_filter_loader.quit()
filtering.adblock_filter_loader.wait()
server_thread.httpd.shutdown()
server_thread.quit()
server_thread.wait()
def saveSettings(self):
settings.settings.setValue("data/MaximumURLLength",
self.maximumURLLength.value())
#settings.settings.setValue("data/MaximumCacheSize", self.maximumCacheSize.value())
settings.settings.setValue("data/RememberHistory",
self.rememberHistoryToggle.isChecked())
settings.settings.setValue("network/GeolocationEnabled",
self.geolocationToggle.isChecked())
data.geolocation_whitelist = [
self.geolocationWhitelist.item(authority).text()
for authority in range(0, self.geolocationWhitelist.count())
]
data.geolocation_blacklist = [
self.geolocationBlacklist.item(authority).text()
for authority in range(0, self.geolocationBlacklist.count())
]
data.saveData()
def train(model_id,train_x,train_y,valid_x,valid_y,test_x):
train_x,train_y=shuffle(train_x,train_y)
# normalization
scaler = StandardScaler()
train_x = scaler.fit_transform(train_x)
valid_x = scaler.transform(valid_x)
test_x = scaler.transform(test_x)
random_state=random.randint(0, 1000000)
print('random state: {state}'.format(state=random_state))
clf = LogisticRegression(penalty='l2',
dual=False,
tol=0.0001,
C=1.0,
fit_intercept=True,
intercept_scaling=1,
class_weight=None,
random_state=None,
solver='lbfgs',
max_iter=1000,
multi_class='ovr',
verbose=True
)
clf.fit(train_x, train_y)
valid_predictions = clf.predict_proba(valid_x)
test(valid_y,valid_predictions)
ccv = CalibratedClassifierCV(base_estimator=clf,method="sigmoid",cv='prefit')
ccv.fit(train_x,train_y)
valid_predictions = ccv.predict_proba(valid_x)
test(valid_y,valid_predictions)
test_predictions= ccv.predict_proba(test_x)
data.saveData(valid_predictions,"../valid_results/valid_"+str(model_id)+".csv")
data.saveData(test_predictions,"../results/results_"+str(model_id)+".csv")
def signalGenerator(file_name, signalGen):
try:
# List com os dados para serem salvos
DataGenerator = []
'''
FS -> SAMPLE FRAME
É necessário para gerar a amostragem do sinal
A cada sample_rate, teremos um ciclo do sinal gerado
'''
FS = 1.0 / signalGen.sample_rate
'''
A fórmula utilizada para gerar o sinal foi
Magnetude*sin(Frequence*2*PI*FS*x + Phase)
O X representa a posição da amostra, e como dito,
a cada sample_rate*FS teremos o sinal de repetindo
'''
for g in range(0, signalGen.sample_rate*signalGen.duration):
DataGenerator.append(0)
for i in range(0, len(signalGen.frequencies)):
# Magnetude é opcional, portanto, se não encontrar dados, adiciona 1 como padrão
if(i >= len(signalGen.magnetude)):
signalGen.magnetude.append(1)
# Phase é opcional, portanto, se não encontrar dados, adiciona 0º como padrão
if(i >= len(signalGen.phases)):
signalGen.phases.append(0)
DataGenerator[g] += signalGen.magnetude[i]*math.sin(signalGen.frequencies[i]*2*math.pi*g*FS + (signalGen.phases[i]*math.pi / 180));
# Gera a classe e sinal e salva o objeto no arquivo de dados
dataSignal = DataSignal(signalGen.sample_rate, signalGen.duration, len(DataGenerator), DataGenerator)
data.saveData(file_name, dataSignal)
return True
except:
return False
def train(model_id,train_x,train_y,valid_x,valid_y,test_x):
train_x,train_y=shuffle(train_x,train_y)
random_state=random.randint(0, 1000000)
print('random state: {state}'.format(state=random_state))
# build a classifier
clf = RandomForestClassifier(n_jobs=8)
# specify parameters and distributions to sample from
param_dist = {
"n_estimators":sp_randint(20,40),
"criterion": ["gini", "entropy"],
"max_depth": sp_randint(3, 10000),
"min_samples_split": sp_randint(1, 30),
"min_samples_leaf": sp_randint(1, 30),
"max_features": sp_randint(1, 93),
"bootstrap": [True, False],
'random_state':sp_randint(1, 1000000),
}
# run randomized search
random_search = RandomizedSearchCV(clf, param_distributions=param_dist,
n_iter=2,cv=9,n_jobs=3)
random_search.fit(train_x,train_y)
valid_predictions = random_search.predict_proba(valid_x)
test_predictions= random_search.predict_proba(test_x)
loss = test(valid_y,valid_predictions,True)
if loss<10.438:
output=[loss,random_search.best_estimator_]
print("model[\""+str(model_id)+"\"]="),
print(output)
data.saveData(valid_predictions,"../valid_results/valid_"+str(model_id)+".csv")
data.saveData(test_predictions,"../results/results_"+str(model_id)+".csv")
def trainxgb(model_id,train_x,train_y,valid_x,valid_y,test_x):
train_x,train_y=shuffle(train_x,train_y)
random_state=random.randint(0, 1000000)
print('random state: {state}'.format(state=random_state))
xgb = XGBoostClassifier(base_estimator='gbtree',
objective='multi:softprob',
metric='mlogloss',
num_classes=9,
learning_rate=random.uniform(0.01,0.05),
max_depth=random.randint(10,20),
max_samples=random.uniform(0.0,1.0),
max_features=random.uniform(0.0,1.0),
max_delta_step=random.randint(1,10),
min_child_weight=random.randint(1,10),
min_loss_reduction=1,
l1_weight=0.0,
l2_weight=0.0,
l2_on_bias=False,
gamma=0.02,
inital_bias=random.uniform(0.0,1.0),
random_state=random_state,
watchlist=[[valid_x,valid_y]],
n_jobs=30,
n_iter=3000,
)
xgb.fit(train_x, train_y)
valid_predictions = xgb.predict_proba(valid_x)
if test(valid_y,valid_predictions) <0.450:
test_predictions= xgb.predict_proba(test_x)
data.saveData(valid_predictions,"../valid_results/valid_"+str(model_id)+".csv")
data.saveData(test_predictions,"../results/results_"+str(model_id)+".csv")
from time import sleep
# for triggering the modules
import data as dm
import motor as mm
import webcam as wm
import joystick as jm
maxThrottle = 0.25
motor = mm.Motor(2, 3, 4, 17, 22, 27)
record = 0
while True:
joyval = jm.getJS()
steering = joyval['axis1']
throttle = joyval['o'] * maxThrottle
if joyval['share'] == 1:
if record ==0:
print('Recording Started ...')
record +=1
sleep(0.300)
if record == 1:
img = wm.getImg(True, size=[240,120])
dm.saveData(img, steering)
elif record == 2:
dm.saveLog()
record = 0
motor.move(throttle, -steering)
cv2.waitKey(1)
from sklearn.decomposition import NMF
from data import loadData, saveData
import numpy as np
# 最重要的参数是n_components、alpha、l1_ratio、solver
nmf = NMF(
n_components=128, # k value,默认会保留全部特征
# W H 的初始化方法，'random' | 'nndsvd'(默认) | 'nndsvda' | 'nndsvdar' | 'custom',
init=None,
solver='cd', # 'cd' | 'mu'
# {'frobenius', 'kullback-leibler', 'itakura-saito'},
beta_loss='frobenius',
tol=1e-10, # 停止迭代的极限条件
max_iter=200, # 最大迭代次数
random_state=None,
alpha=0., # 正则化参数
l1_ratio=0., # 正则化参数
verbose=0, # 冗长模式
shuffle=False # 针对"cd solver"
)
trius = loadData(filename='trius.npy')
X = np.abs(trius)
nmf.fit(X)
W = nmf.fit_transform(X)
H = nmf.components_
print('reconstruction_err_', nmf.reconstruction_err_) # 损失函数值
print('n_iter_', nmf.n_iter_) # 实际迭代次数
saveData(W, filename='nmf.npy')
saveData(H, filename='basis.npy')
i * batch_size:len(x_train)]
# Run
_, c = sess.run([optimizer, loss], feed_dict={X: batch_xs})
if epoch % display_step == 0:
print("Iteration: %04d " % (epoch), "loss=", "{:.9f}".format(c))
print('Triang CLF')
for epoch in range(training_epoch):
# Loop over all batches
for i in range(total_batch):
batch_xs = x_train[i * batch_size:(i + 1) *
batch_size] if i < total_batch else x_train[
i * batch_size:len(x_train)]
batch_ys = y_train[i * batch_size:(i + 1) *
batch_size] if i < total_batch else y_train[
i * batch_size:len(y_train)]
# Run
_, c = sess.run([optimizer_clf, loss_clf],
feed_dict={
X: batch_xs,
Y: batch_ys
})
if epoch % display_step == 0:
a = sess.run(acc, feed_dict={X: x_test, Y: y_test})
mse = sess.run(loss, feed_dict={X: x_test})
print("Iteration: %04d " % (epoch), "loss=",
"{:.9f} acc {:.9f} decode loss {:.9f}".format(c, a, mse))
code = sess.run(encoder_op, feed_dict={X: F})
saveData(code, filename='code.npy')
def main():
path = "../Data/google_trace_timeseries/data_resource_usage_5Minutes_6176858948.csv"
aspects = ["meanCPUUsage", "canonical memory usage"]
predicted_aspect = "meanCPUUsage"
num_epochs = 1000
learning_rate = 0.005
n_slidings_encoder = [26, 28]
n_slidings_decoder = [2, 4, 6]
batch_sizes = [16, 32]
size_models = [[16], [32], [8, 4], [16, 8]]
activations = ["tanh", "sigmoid"]
input_keep_probs = [0.95, 0.9]
output_keep_probs = [0.9]
state_keep_probs = [0.95, 0.9]
# n_slidings_encoder = [16]
# n_slidings_decoder = [2]
# batch_sizes = [16]
# size_models = [[4, 2]]
# activations = ["tanh"]
# input_keep_probs = [0.95]
# output_keep_probs = [0.9]
# state_keep_probs = [0.95]
rate = 5
result_file_path = 'result_encoder_decoder.csv'
loss_file_path = 'loss_encoder_decoder.csv'
## GET COMBINATIONS ##
combinations = []
for n_sliding_encoder in n_slidings_encoder:
for n_sliding_decoder in n_slidings_decoder:
for batch_size in batch_sizes:
for size_model in size_models:
for activation in activations:
for input_keep_prob in input_keep_probs:
for output_keep_prob in output_keep_probs:
for state_keep_prob in state_keep_probs:
combination_i = [n_sliding_encoder, n_sliding_decoder,
batch_size, size_model, activation,
input_keep_prob, output_keep_prob,
state_keep_prob]
combinations.append(combination_i)
for combination in combinations:
tf.reset_default_graph()
n_sliding_encoder = combination[0]
n_sliding_decoder = combination[1]
batch_size = combination[2]
size_model = combination[3]
activation = combination[4]
input_keep_prob = combination[5]
output_keep_prob = combination[6]
state_keep_prob = combination[7]
### GET DATA : TRAINING SET, TEST SET, VALIDATION SET ###
nor_data, amax, amin = data.get_goodletrace_data(path, aspects)
x_train_encoder, y_train, x_test_encoder, y_test = data.get_data_samples(nor_data, n_sliding_encoder,
predicted_aspect, rate)
x_train_decoder, x_test_decoder = data.get_data_decoder(x_train_encoder, x_test_encoder, n_sliding_decoder)
x_train_encoder, x_train_decoder, y_train, x_val_encoder, x_val_decoder, y_val = \
data.getValidationSet(x_train_encoder, x_train_decoder, y_train, 5)
# print(x_train_encoder.shape, x_train_decoder.shape, y_train.shape, x_val_encoder.shape, x_val_decoder.shape,
# y_val.shape)
# return 0
loss_train_value = []
loss_valid_value = []
n_train = y_train.shape[0]
num_batches = int(x_train_encoder.shape[0] / batch_size)
timestep_encoder = n_sliding_encoder
timestep_decoder = n_sliding_decoder
input_dim = len(aspects)
X_encoder = tf.placeholder(tf.float32, [None, timestep_encoder, input_dim], name='X_encoder')
X_decoder = tf.placeholder(tf.float32, [None, timestep_decoder, input_dim], name='X_decoder')
y = tf.placeholder(tf.float32, [None, 1], name='output')
output, outputs_encoder, outputs_decoder = encoder_decoder(X_encoder, X_decoder, size_model, activation,
input_keep_prob, output_keep_prob, state_keep_prob)
outputs_encoder = tf.identity(outputs_encoder, name='outputs_encoder')
loss = tf.reduce_mean(tf.squared_difference(output, y))
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss)
init_op = tf.global_variables_initializer()
with tf.Session() as sess:
t = datetime.now().time()
start_time = (t.hour * 60 + t.minute) * 60 + t.second
sess.run(init_op)
## EARLY STOPPING ##
pre_loss_valid = 100
x = 0
early_stopping_val = 5
### START TO TRAIN ###
for i in range(num_epochs):
num_epochs_i = i + 1
for j in range(num_batches + 1):
a = batch_size * j
b = a + batch_size
if b > n_train:
b = n_train
x_batch_encoder = x_train_encoder[a:b, :, :]
x_batch_decoder = x_train_decoder[a:b, :, :]
y_batch = y_train[a:b, :]
# print(x_batch.shape, y_batch.shape)
loss_j, _ = sess.run([loss, optimizer], feed_dict={X_encoder: x_batch_encoder,
X_decoder: x_batch_decoder,
y: y_batch})
loss_train_i = sess.run(loss, feed_dict={X_encoder: x_train_encoder,
X_decoder: x_train_decoder,
y: y_train})
loss_valid_i = sess.run(loss, feed_dict={X_encoder: x_val_encoder,
X_decoder: x_val_decoder,
y: y_val})
# print(num_epochs_i, loss_train_i, loss_valid_i)
loss_train_value.append(loss_train_i)
loss_valid_value.append(loss_valid_i)
if loss_valid_i > pre_loss_valid:
x = x+1
if x == early_stopping_val:
break
else:
x = 0
pre_loss_valid = loss_valid_i
### OUTPUT ###
output_test = sess.run(output, feed_dict={X_encoder: x_test_encoder,
X_decoder: x_test_decoder,
y: y_test})
output_test = output_test * (amax[0] - amin[0]) + amin[0]
y_test_act = y_test * (amax[0] - amin[0]) + amin[0]
loss_test_act = np.mean(np.abs(output_test - y_test_act))
# print(loss_test_act)
t = datetime.now().time()
end_time = (t.hour * 60 + t.minute) * 60 + t.second
training_encoder_time = (end_time - start_time)
### SAVE DATA ###
name = data.saveData(combination, loss_test_act, num_epochs_i, result_file_path, training_encoder_time)
# print(name)
# outputs_encoder = sess.run(outputs_encoder, feed_dict={X_encoder: x_train_encoder,
# X_decoder: x_train_decoder,
# y: y_train})
# print(outputs_encoder[:, -1, :].shape)
# print(time)
### SAVE MODEL ###
# print('\nSaving...')
cwd = os.getcwd()
saved_path = 'model/model'
saved_path += str(combination)
saved_path += '.ckpt'
saved_path = os.path.join(cwd, saved_path)
# print(saved_path)
shutil.rmtree(saved_path, ignore_errors=True)
saver = tf.train.Saver()
saver.save(sess=sess, save_path=saved_path)
# print("ok")
sess.close()
def main():
path = "../Data/google_trace_timeseries/data_resource_usage_5Minutes_6176858948.csv"
aspects = ["meanCPUUsage", "canonical memory usage"]
predicted_aspect = "meanCPUUsage"
n_slidings = [3, 4, 5, 6]
batch_sizes = [16, 32]
learning_rate = 0.005
num_epochs = 2
rnn_cellsizes = [[4], [8], [16], [32], [4, 2], [8, 4], [16, 4], [16, 8],
[32, 4], [32, 8], [32, 16]]
activations = ["tanh", "sigmoid"]
rate = 5
result_file_path = 'result_multi.csv'
combination = []
for n_sliding in n_slidings:
for batch_size in batch_sizes:
for rnn_cellsize in rnn_cellsizes:
for activation in activations:
combination_i = [
n_sliding, batch_size, rnn_cellsize, activation
]
combination.append(combination_i)
for combination_i in combination:
tf.reset_default_graph()
n_sliding = combination_i[0]
batch_size = combination_i[1]
rnn_unit = combination_i[2]
activation = combination_i[3]
nor_data, amax, amin = data.get_goodletrace_data(path, aspects)
x_train, y_train, x_test, y_test = data.get_data_samples(
nor_data, n_sliding, predicted_aspect, rate)
# x_train, y_train, x_valid, y_valid = data.getValidationSet(x_train, y_train, 5)
loss_train_value = []
loss_valid_value = []
n_train = x_train.shape[0]
num_batches = int(x_train.shape[0] / batch_size)
timestep = n_sliding
input_dim = len(aspects)
X = tf.placeholder(tf.float32, [None, timestep, input_dim])
y = tf.placeholder(tf.float32, [None, 1])
output = model_rnn(X, rnn_unit, activation)
loss = tf.reduce_mean(tf.squared_difference(output, y))
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss)
init_op = tf.global_variables_initializer()
with tf.Session() as sess:
t = datetime.now().time()
start_time = (t.hour * 60 + t.minute) * 60 + t.second
sess.run(init_op)
# pre_loss_valid = 100
# x = 0
# early_stopping_val = 5
epoch_i = 0
for i in range(num_epochs):
for j in range(num_batches + 1):
a = batch_size * j
b = a + batch_size
if b > n_train:
b = n_train
x_batch = x_train[a:b, :, :]
y_batch = y_train[a:b, :]
# print(x_batch.shape, y_batch.shape)
loss_j, _ = sess.run([loss, optimizer],
feed_dict={
X: x_batch,
y: y_batch
})
loss_train_i = sess.run(loss,
feed_dict={
X: x_train,
y: y_train
})
# loss_valid_i = sess.run(loss, feed_dict={X: x_valid,
# y: y_valid})
loss_train_value.append(loss_train_i)
# loss_valid_value.append(loss_valid_i)
# if loss_valid_i > pre_loss_valid:
# x = x+1
# if x == early_stopping_val:
# break
# else:
# x = 0
# pre_loss_valid = loss_valid_i
epoch_i += 1
t = datetime.now().time()
train_time = (t.hour * 60 + t.minute) * 60 + t.second
training_time = train_time - start_time
output_test = sess.run(output, feed_dict={X: x_test, y: y_test})
output_test = output_test * (amax[0] - amin[0]) + amin[0]
y_test_act = y_test * (amax[0] - amin[0]) + amin[0]
t = datetime.now().time()
test_time = (t.hour * 60 + t.minute) * 60 + t.second
testing_time = test_time - train_time
system_time = test_time - start_time
loss_test_act = np.mean(np.abs(output_test - y_test_act))
explained_variance_score = sk.explained_variance_score(
y_test_act, output_test)
mean_absolute_error = sk.mean_absolute_error(
y_test_act, output_test)
mean_squared_error = sk.mean_squared_error(y_test_act, output_test)
median_absolute_error = sk.median_absolute_error(
y_test_act, output_test)
r2_score = sk.r2_score(y_test_act, output_test)
# t = datetime.now().time()
# end_time = (t.hour * 60 + t.minute) * 60 + t.second
#
# training_time = (end_time - start_time)
name = data.saveData(combination_i, loss_test_act,
loss_valid_value, loss_train_value, epoch_i,
result_file_path, output_test, y_test_act,
explained_variance_score, mean_absolute_error,
mean_squared_error, median_absolute_error,
r2_score, training_time, testing_time,
system_time)
print(name)
opt = {
'architecture' : [784,784,784,30,10],
'learningRate' : 1.5,
'error' : 0.001,
'epochs' : 15,
'batch' : 100
}
nn = ann.ann(opt)
# #combain
# rbm = files.loadData('rbm-1000.db')
# nn.rbm(rbm)
#train
nn.train(train_data, train_result)
files.saveData(nn, 'nn.db')
_results = nn.sim(test_data)
_results = _results.transpose()
accuracy = 0
for i in range(len(test_result)):
if i < 20:
print _results[i].argmax(), " : ", test_result[i].argmax()
if _results[i].argmax() == test_result[i].argmax():
accuracy += 1.00
print accuracy, " / ", len(test_result)
accuracy = accuracy/len(test_result)
from sklearn.utils import shuffle
from sklearn.svm import SVC
from sklearn.grid_search import GridSearchCV
from test import test
train_x,train_y,valid_x,valid_y,test_x=data.loadData()
train_x,train_y=shuffle(train_x,train_y)
param_grid = [
{'C': [1, 10, 100, 1000], 'kernel': ['linear']},
{'C': [1, 10, 100, 1000], 'gamma': [0.001, 0.0001], 'kernel': ['rbf']},
]
clf = SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,
gamma=0.0, kernel='rbf', max_iter=-1, probability=True,
random_state=None, shrinking=True, tol=0.001, verbose=True)
clf.fit(train_x, train_y)
#gs = GridSearchCV(svc,param_grid,n_jobs=8,verbose=2)
#gs.fit(train_x, train_y)
valid_predictions = clf.predict_proba(valid_x)
test_predictions= clf.predict_proba(test_x)
test(valid_y,valid_predictions)
data.saveData(valid_predictions,"../valid_results/valid_215.csv")
data.saveData(test_predictions,"../results/results_215.csv")
def train(model_id,train_x,train_y,valid_x,valid_y,test_x):
train_x,train_y=shuffle(train_x,train_y)
maximum_auc=0.0
random_state=random.randint(0, 1000000)
for i in tqdm(range(1000)):
config={
'base_estimator':'gbtree',
'objective':'multi:softprob',
'metric':'mlogloss',
'num_classes':2,
'learning_rate':random.uniform(0.01,0.15),
'max_depth':20+random.randint(0,10),
'max_samples':random.uniform(0.3,1.0),
'max_features':random.uniform(0.3,1.0),
'max_delta_step':random.randint(1,10),
'min_child_weight':random.randint(1,8),
'min_loss_reduction':1,
'l1_weight':random.uniform(0.0,10.0),
'l2_weight':random.uniform(0.0,10.0),
'l2_on_bias':False,
'gamma':random.uniform(0.0,0.1),
'inital_bias':0.5,
'random_state':random_state,
}
clf = XGBoostClassifier(
config['base_estimator'],
config['objective'],
config['metric'],
config['num_classes'],
config['learning_rate'],
config['max_depth'],
config['max_samples'],
config['max_features'],
config['max_delta_step'],
config['min_child_weight'],
config['min_loss_reduction'],
config['l1_weight'],
config['l2_weight'],
config['l2_on_bias'],
config['gamma'],
config['inital_bias'],
config['random_state'],
watchlist=[[valid_x,valid_y]],
n_jobs=8,
n_iter=30000,
)
clf.fit(train_x, train_y)
valid_predictions = clf.predict_proba(valid_x)[:, 1]
test_predictions= clf.predict_proba(test_x)[:, 1]
auc = roc_auc_score(valid_y,valid_predictions)
if auc>maximum_auc:
maximum_auc=auc
best_config=config
print('new auc:')
print(auc)
data.saveData(valid_id,valid_predictions,"./valid_results/valid_"+str(model_id)+".csv")
data.saveData(test_id,test_predictions,"./results/results_"+str(model_id)+".csv")
print('maximum_auc:')
print(maximum_auc)
print(config)
# plt.show()
def load_save_label(filename):
binfile = open(filename , 'rb')
buf = binfile.read()
labelNum = len(buf)
labelNum -= 8
# print labelNum
index = 0
labels = struct.unpack_from('>'+str(labelNum)+'B' , buf , index);
print 'success: '+filename
return labels
train_data = load_save_function('train-images-idx3-ubyte');
train_result = load_save_label('train-labels-idx1-ubyte');
test_data = load_save_function('t10k-images-idx3-ubyte');
test_result = load_save_label('t10k-labels-idx1-ubyte');
print train_data.shape
train_data.tofile("mnist_train_data.db")
data.saveData(train_result, "mnist_train_result.db")
test_data.tofile("mnist_test_data.db")
data.saveData(test_result, "mnist_test_result.db")
