def LAG_Fit(slctd_Features_Train, trainLabel):
#lag1 = LAG(encode='distance', num_clusters=[12,12,12,12,12,12,12,12,12,12], alpha=10, learner=myLLSR(onehot=False))
#lag2 = LAG(encode='distance', num_clusters=[12,12,12,12,12,12,12,12,12,12], alpha=10, learner=myLLSR(onehot=False))
#lag3 = LAG(encode='distance', num_clusters=[12,12,12,12,12,12,12,12,12,12], alpha=10, learner=myLLSR(onehot=False))
lag1 = LAG(encode='distance',
num_clusters=[5, 5, 5, 5, 5, 5, 5, 5, 5, 5],
alpha=10,
learner=myLLSR(onehot=False))
lag2 = LAG(encode='distance',
num_clusters=[5, 5, 5, 5, 5, 5, 5, 5, 5, 5],
alpha=10,
learner=myLLSR(onehot=False))
lag3 = LAG(encode='distance',
num_clusters=[5, 5, 5, 5, 5, 5, 5, 5, 5, 5],
alpha=10,
learner=myLLSR(onehot=False))
#Extract Training Features
slctd_Features_Train1 = slctd_Features_Train[0]
slctd_Features_Train2 = slctd_Features_Train[1]
slctd_Features_Train3 = slctd_Features_Train[2]
#Fit the three LAG units
lag1.fit(slctd_Features_Train1, trainLabel)
lag2.fit(slctd_Features_Train2, trainLabel)
lag3.fit(slctd_Features_Train3, trainLabel)
#Get the Transformed Training Features
features_Train_Trans1 = lag1.transform(slctd_Features_Train1)
features_Train_Trans2 = lag2.transform(slctd_Features_Train2)
features_Train_Trans3 = lag3.transform(slctd_Features_Train3)
#Cascade all the LAG features
features_Train_LAG = np.concatenate(
(features_Train_Trans1, features_Train_Trans2, features_Train_Trans3),
axis=1)
return features_Train_LAG, lag1, lag2, lag3
def LAG_Fit(slctd_Features_Train, trainLabel):
num1 = 10
num2 = 10
num3 = 10
num4 = 10
num5 = 10
numCluster1 = [num1, num1, num1, num1, num1, num1, num1, num1, num1, num1]
numCluster2 = [num2, num2, num2, num2, num2, num2, num2, num2, num2, num2]
numCluster3 = [num3, num3, num3, num3, num3, num3, num3, num3, num3, num3]
numCluster4 = [num4, num4, num4, num4, num4, num4, num4, num4, num4, num4]
numCluster5 = [num5, num5, num5, num5, num5, num5, num5, num5, num5, num5]
lag1 = LAG(encode='distance',
num_clusters=numCluster1,
alpha=10,
learner=myLLSR(onehot=False))
lag2 = LAG(encode='distance',
num_clusters=numCluster2,
alpha=10,
learner=myLLSR(onehot=False))
lag3 = LAG(encode='distance',
num_clusters=numCluster3,
alpha=10,
learner=myLLSR(onehot=False))
lag4 = LAG(encode='distance',
num_clusters=numCluster4,
alpha=10,
learner=myLLSR(onehot=False))
lag5 = LAG(encode='distance',
num_clusters=numCluster5,
alpha=10,
learner=myLLSR(onehot=False))
#Extract Training Features
slctd_Features_Train1 = slctd_Features_Train[0]
slctd_Features_Train2 = slctd_Features_Train[1]
slctd_Features_Train3 = slctd_Features_Train[2]
slctd_Features_Train4 = slctd_Features_Train[3]
slctd_Features_Train5 = slctd_Features_Train[4]
#Fit the three LAG units
lag1.fit(slctd_Features_Train1, trainLabel)
lag2.fit(slctd_Features_Train2, trainLabel)
lag3.fit(slctd_Features_Train3, trainLabel)
lag4.fit(slctd_Features_Train4, trainLabel)
lag5.fit(slctd_Features_Train5, trainLabel)
#Get the Transformed Training Features
features_Train_Trans1 = lag1.transform(slctd_Features_Train1)
features_Train_Trans2 = lag2.transform(slctd_Features_Train2)
features_Train_Trans3 = lag3.transform(slctd_Features_Train3)
features_Train_Trans4 = lag4.transform(slctd_Features_Train4)
features_Train_Trans5 = lag5.transform(slctd_Features_Train5)
#Cascade all the LAG features
features_Train_LAG = np.concatenate(
(features_Train_Trans1, features_Train_Trans2, features_Train_Trans3,
features_Train_Trans4, features_Train_Trans5),
axis=1)
return features_Train_LAG, lag1, lag2, lag3, lag4, lag5
def lag_oper(feature_ce, test_ce, train_label):
my_LAG = LAG(encode='distance', num_clusters=[5, 5, 5, 5, 5, 5, 5, 5, 5, 5],
alpha=10, learner=myLLSR(onehot=False))
my_LAG.fit(feature_ce, train_label)
X_train_trans = my_LAG.transform(feature_ce)
X_test_trans = my_LAG.transform(test_ce)
# X_train_pred_prob = my_LAG.predict_proba(feature_ce)
# print(" --> train acc: %s" % str(my_LAG.score(feature_ce, train_label)))
return X_train_trans, X_test_trans
def lag_module(params, use_filters):
all_train_labels = load_data(
os.path.join(params.save_data, "all_train_labels"))
for ratio in params.ratios:
time_start = time.time()
if use_filters is True:
filters = load_data(
os.path.join(params.save_data, "filters_" + str(ratio)))
subset_label = all_train_labels[:int(all_train_labels.shape[0] *
ratio)].copy()
for i in range(2, params.num_layers): ## ATTENTION
data = []
for j in range(0, 50000, 5000):
data.append(
load_data(
os.path.join(params.save_data,
'out_{}_{}'.format(j, i))))
data = np.concatenate(data, axis=0)
subset_data = data[:int(data.shape[0] * ratio)].copy()
del data
subset_data = np.reshape(subset_data,
newshape=(subset_data.shape[0], -1))
print("subset data shape:", subset_data.shape)
if use_filters is True:
print("Number of selected channels:", np.sum(filters[i]))
subset_data = subset_data[:, filters[i]]
lag = LAG(
num_clusters=[10, 10, 10, 10, 10, 10, 10, 10, 10, 10],
alpha=10,
learner=myLLSR(onehot=False))
lag.fit(subset_data, subset_label)
subset_predprob = lag.predict_proba(subset_data)
save_data(
lag,
os.path.join(params.save_data,
'LAG_{}_{}'.format(i, ratio)))
save_data(
subset_predprob,
os.path.join(params.save_data,
'lag_predict_{}_{}'.format(i, ratio)))
print("RATIO=", ratio, " LAYER=", i, " DONE")
print("Time cost - LAG:",
time.time() - time_start, "ratio=", ratio)
save_data(
subset_label,
os.path.join(params.save_data,
'lag_labels_{}_{}'.format(i, ratio)))
#selectedFeaturesIndeces_unit1_train , reducedFeatureImgs_unit1_train, numOfNs_unit1_train = featureSelector(pooled_module1Out1_train,subYTrain4)
selectedFeaturesIndeces_unit1_train, reducedFeatureImgs_unit1_train, numOfNs_unit1_train = featureSelector_binMethod(
pooled_module1Out1_train, subYTrain4)
reducedFeatureImgs_unit1_test = featureSelectorWithoutCrossEnt(
pooled_module1Out1_test, selectedFeaturesIndeces_unit1_train,
numOfNs_unit1_train)
print("feature selection is done for train and test::"
) #Elapsed time is 99.167789 seconds.
timerObj.toc()
#LAG Unit
lag_unit1 = LAG(encode='distance',
num_clusters=[7, 7, 7, 7, 7, 7, 7, 7, 7, 7],
alpha=10,
learner=myLLSR(onehot=False))
lag_unit1.fit(reducedFeatureImgs_unit1_train, subYTrain4)
timerObj.tic()
finalFeature_unit1_train = lag_unit1.transform(reducedFeatureImgs_unit1_train)
print("LAG is done for train set:") #Elapsed time is 0.321958 seconds.
timerObj.toc()
timerObj.tic()
finalFeature_unit1_test = lag_unit1.transform(reducedFeatureImgs_unit1_test)
print("LAG is done for test set:") #Elapsed time is 0.173667 seconds.
timerObj.toc()
# In[ ]:
##2nd unit(depth 2)
##Module 2
##Module 2
#Max-pooling
pooled_module1Out1_train = MaxpoolingOperator(module1Out1_train)
pooled_module1Out1_test = MaxpoolingOperator(module1Out1_test)
#Feature Selection
timerObj.tic()
#selectedFeaturesIndeces_unit1_train , reducedFeatureImgs_unit1_train, numOfNs_unit1_train = featureSelector(pooled_module1Out1_train,subYTrain4)
selectedFeaturesIndeces_unit1_train , reducedFeatureImgs_unit1_train, numOfNs_unit1_train = featureSelector_binMethod(pooled_module1Out1_train,subYTrain4)
reducedFeatureImgs_unit1_test = featureSelectorWithoutCrossEnt(pooled_module1Out1_test, selectedFeaturesIndeces_unit1_train, numOfNs_unit1_train)
print("feature selection is done for train and test::")#Elapsed time is 309.562113 seconds.
timerObj.toc()
#LAG Unit
lag_unit1 = LAG(encode='distance', num_clusters=[7,7,7,7,7,7,7,7,7,7], alpha=10, learner=myLLSR(onehot=False))
lag_unit1.fit(reducedFeatureImgs_unit1_train,subYTrain4)
timerObj.tic()
finalFeature_unit1_train = lag_unit1.transform(reducedFeatureImgs_unit1_train)
print("LAG is done for train set:")#Elapsed time is 0.321958 seconds.
timerObj.toc()
timerObj.tic()
finalFeature_unit1_test = lag_unit1.transform(reducedFeatureImgs_unit1_test)
print("LAG is done for test set:")#Elapsed time is 0.173667 seconds.
timerObj.toc()
# In[ ]:
def LAG_Transform(slctd_Features_Train, slctd_Features_Test, trainNum, testNum):
lag1 = LAG(encode='distance', num_clusters=[12,12,12,12,12,12,12,12,12,12], alpha=10, learner=myLLSR(onehot=False))
lag2 = LAG(encode='distance', num_clusters=[12,12,12,12,12,12,12,12,12,12], alpha=10, learner=myLLSR(onehot=False))
lag3 = LAG(encode='distance', num_clusters=[12,12,12,12,12,12,12,12,12,12], alpha=10, learner=myLLSR(onehot=False))
#lag1 = LAG(encode='distance', num_clusters=[5,5,5,5,5,5,5,5,5,5], alpha=10, learner=myLLSR(onehot=False))
#lag2 = LAG(encode='distance', num_clusters=[5,5,5,5,5,5,5,5,5,5], alpha=10, learner=myLLSR(onehot=False))
#lag3 = LAG(encode='distance', num_clusters=[5,5,5,5,5,5,5,5,5,5], alpha=10, learner=myLLSR(onehot=False))
#Extract Training Features
slctd_Features_Train1 = slctd_Features_Train[0].reshape(trainNum, -1)
slctd_Features_Train2 = slctd_Features_Train[1].reshape(trainNum, -1)
slctd_Features_Train3 = slctd_Features_Train[2].reshape(trainNum, -1)
#Extract Testing Features
slctd_Features_Test1 = slctd_Features_Test[0].reshape(testNum, -1)
slctd_Features_Test2 = slctd_Features_Test[1].reshape(testNum, -1)
slctd_Features_Test3 = slctd_Features_Test[2].reshape(testNum, -1)
#Fit the three LAG units
lag1.fit(slctd_Features_Train1, trainLabel)
lag2.fit(slctd_Features_Train2, trainLabel)
lag3.fit(slctd_Features_Train3, trainLabel)
#Get the Transformed Training Features
features_Train_Trans1 = lag1.transform(slctd_Features_Train1)
features_Train_Trans2 = lag2.transform(slctd_Features_Train2)
features_Train_Trans3 = lag3.transform(slctd_Features_Train3)
#Get the Transformed Testing Features
features_Test_Trans1 = lag1.transform(slctd_Features_Test1)
features_Test_Trans2 = lag2.transform(slctd_Features_Test2)
features_Test_Trans3 = lag3.transform(slctd_Features_Test3)
#Cascade all the LAG features
features_Test_LAG = np.concatenate((features_Test_Trans1,features_Test_Trans2,features_Test_Trans3), axis=1)
features_Train_LAG = np.concatenate((features_Train_Trans1,features_Train_Trans2,features_Train_Trans3), axis=1)
return features_Test_LAG, features_Train_LAG
fd0_ns = fd0_sorted[0:round(len(fd0_sorted)/2)]
fd1_ns = fd1_sorted[0:round(len(fd1_sorted)/2)]
fd2_ns = fd2_sorted[0:round(len(fd2_sorted)/2)]
f_number0 = [i[0] for i in fd0_ns]
f_number1 = [i[0] for i in fd1_ns]
f_number2 = [i[0] for i in fd2_ns]
from lag import LAG
from llsr import LLSR as myLLSR
lag = LAG(encode='distance', num_clusters=[5,5,5,5,5,5,5,5,5,5], alpha=10, learner=myLLSR(onehot=False))
lag.fit(X_reshape0[:,:,f_number0].reshape(X_reshape0.shape[0],-1), y_train[0:nti])
lag1 = LAG(encode='distance', num_clusters=[5,5,5,5,5,5,5,5,5,5], alpha=10, learner=myLLSR(onehot=False))
lag1.fit(X_reshape1[:,:,f_number1].reshape(X_reshape1.shape[0],-1), y_train[0:nti])
lag2 = LAG(encode='distance', num_clusters=[5,5,5,5,5,5,5,5,5,5], alpha=10, learner=myLLSR(onehot=False))
lag2.fit(X_reshape2[:,:,f_number2].reshape(X_reshape2.shape[0],-1), y_train[0:nti])
end = time.time()
print('Module 2:',(end-start))
"""Module 3"""
def LAG_train_unit(x_1, x_2, x_3, Y):
input_list = [x_1, x_2, x_3]
the_list = []
lag_list = []
for i in range(3):
print("Currently doing LAG for hop {}".format(i+1))
lag = LAG(encode='distance', num_clusters=[2,2,2,2,2,2,2,2,2,2], alpha=10, learner=myLLSR(onehot=False))
lag.fit(input_list[i], Y)
X_train_trans = lag.transform(input_list[i])
the_list.append(X_train_trans)
lag_list.append(lag)
return the_list, lag_list
