def marginPerceptron_cvRate(train_matrix, train_actual_labels, rateRange,
num_folds):
av_accuracy_list = []
for rate in rateRange:
print 'rate: ', rate
accuracy_list = []
training_lab = copy.deepcopy(train_actual_labels)
training_lab = training_lab.tolist()
training_mat = copy.deepcopy(train_matrix)
training_mat = training_mat.tolist()
subset_size = len(training_mat) / num_folds
for i in range(num_folds):
#
if i == num_folds - 1:
temp_test_Matrix = training_mat[(i) * subset_size:]
temp_train_Matrix = training_mat[0:][:i * subset_size]
#
temp_test_Labels = training_lab[(i) * subset_size:]
temp_train_Labels = training_lab[0:][:i * subset_size]
else:
temp_test_Matrix = training_mat[i * subset_size:][:subset_size]
#temp_train_Matrix =np.append(training_mat[:i * subset_size],training_mat[(i + 1) * subset_size:])
temp_train_Matrix = training_mat[:i *
subset_size] + training_mat[
(i + 1) * subset_size:]
#
temp_test_Labels = training_lab[i * subset_size:][:subset_size]
temp_train_Labels = training_lab[:i *
subset_size] + training_lab[
(i + 1) * subset_size:]
# Training
w_train, numUpdates_simpleP = percepFunctions.marginPerceptron(
temp_train_Matrix, temp_train_Labels, 'zeros', 1, rate)
#print 'weight %s is %f' %(i,max(w_train))
predicted_labels = projectFunctions.prediction_Perceptron(
temp_test_Matrix, temp_test_Labels, w_train)
# Metric
#test_accuracy, test_f1_score = projectFunctions.accuracyMETRIC(temp_test_Labels, predicted_labels)
test_accuracy = projectFunctions.accuracyMETRIC(
temp_test_Labels, predicted_labels)
accuracy_list.append(test_accuracy)
av_accuracy = np.mean(accuracy_list)
av_accuracy_list.append(av_accuracy)
print av_accuracy
best_accuracy = max(av_accuracy_list)
max_index = av_accuracy_list.index(best_accuracy)
best_rate = rateRange[max_index]
#
line, = plt.plot(rateRange, av_accuracy_list)
plt.xlabel('Learning Rate')
plt.ylabel('Accuracy')
plt.title('Accuracy v/s Learning Rate, for Simple perceptron')
plt.grid(True)
plt.show()
return best_accuracy, best_rate, av_accuracy_list
def eliminateBADfeaturesWITHperceptron(finalFeatureCASE, labelCASE,
thresholdBIAS):
featureCASE = 'w'
tempList = []
delLIST = []
train_matrix, train_labels, test_matrix, test_labels, eval_matrix, eval_labels = eliminateZEROfetures(
featureCASE, labelCASE)
for c in range(len(train_matrix[0])):
temp_train_matrix = copy.deepcopy(train_matrix)
temp_train_matrix = np.delete(temp_train_matrix, c, axis=1)
w, dummy = percepFunctions.weightedPerceptron(temp_train_matrix,
train_labels, 'zeros',
0.675)
train_pred_labels = projectFunctions.prediction_Perceptron(
temp_train_matrix, train_labels, w)
acc_train = projectFunctions.accuracyMETRIC(train_labels,
train_pred_labels)
dummyP, dummyR, f1_train = projectFunctions.f1METRIC(
train_labels, train_pred_labels)
tempList.append([c, acc_train, f1_train])
#print 'c:',c,' acc:',acc_train,' f1:',f1_train
columnSUM = np.sum(tempList, axis=0)
meanF1 = columnSUM[2] / len(tempList)
for c in range(len(train_matrix[0])):
if tempList[c][2] < meanF1 + thresholdBIAS:
delLIST.append(c)
train_matrix = np.delete(train_matrix, delLIST, axis=1)
test_matrix = np.delete(test_matrix, delLIST, axis=1)
eval_matrix = np.delete(eval_matrix, delLIST, axis=1)
#
if finalFeatureCASE == '01':
for i in range(len(train_matrix)):
train_matrix[i][train_matrix[i] > 0] = 1
for i in range(len(test_matrix)):
test_matrix[i][test_matrix[i] > 0] = 1
for i in range(len(eval_matrix)):
eval_matrix[i][eval_matrix[i] > 0] = 1
elif finalFeatureCASE == 'normalize':
colRangeLIST = train_matrix.max(axis=0) - train_matrix.min(axis=0)
for i in range(len(train_matrix)):
for j in range(len(train_matrix[0])):
train_matrix[i][j] = train_matrix[i][j] / colRangeLIST[j]
colRangeLIST = test_matrix.max(axis=0) - test_matrix.min(axis=0)
colRangeLIST[colRangeLIST == 0] = 1
for i in range(len(test_matrix)):
for j in range(len(test_matrix[0])):
test_matrix[i][j] = (test_matrix[i][j]) / colRangeLIST[j]
colRangeLIST = eval_matrix.max(axis=0) - eval_matrix.min(axis=0)
colRangeLIST[colRangeLIST == 0] = 1
for i in range(len(eval_matrix)):
for j in range(len(eval_matrix[0])):
eval_matrix[i][j] = (eval_matrix[i][j]) / colRangeLIST[j]
return train_matrix, train_labels, test_matrix, test_labels, eval_matrix, eval_labels
def allSampleTreeEnsembles(train_MATRIX, train_LABELS, test_MATRIX,
eval_MATRIX, N, mPercent):
lasttestCOLindex = len(test_MATRIX[0]) - 1
test_labels = test_MATRIX[:, lasttestCOLindex]
ensembleDataset_train_TRANS = []
ensembleDataset_test_TRANS = []
ensembleDataset_eval_TRANS = []
numSamples = int(math.ceil(len(train_MATRIX) * mPercent))
for numTrees in range(N):
Sdata = np.array(
[random.choice(train_MATRIX) for _ in range(numSamples)])
#Sdata = train_MATRIX
predicted_train_list = []
predicted_test_list = []
predicted_eval_list = []
num_TrainingFEATURES = len(Sdata[0]) - 1
depth_of_tree = 0
limiting_depth = 10000
to_be_visited = range(num_TrainingFEATURES)
major_trainLABEL = (Counter(train_LABELS)).most_common(1)[0][0]
tree = treeFunctions.all_feautres_ID3_tree(Sdata, num_TrainingFEATURES,
depth_of_tree,
to_be_visited,
limiting_depth)
list_of_labels = list(Counter(train_LABELS))
predicted_train_list = treeFunctions.predictFUNCTION(
tree, train_MATRIX, list_of_labels, major_trainLABEL)
predicted_test_list = treeFunctions.predictFUNCTION(
tree, test_MATRIX, list_of_labels, major_trainLABEL)
predicted_eval_list = treeFunctions.predictFUNCTION(
tree, eval_MATRIX, list_of_labels, major_trainLABEL)
ensembleDataset_train_TRANS.append(predicted_train_list)
ensembleDataset_test_TRANS.append(predicted_test_list)
ensembleDataset_eval_TRANS.append(predicted_eval_list)
testACCuracy = projectFunctions.accuracyMETRIC(test_labels,
predicted_test_list)
print 'After ', numTrees + 1, ' trees, Test Accuracy: ', testACCuracy
ensembleDataset_train = np.transpose(np.array(ensembleDataset_train_TRANS))
ensembleDataset_test = np.transpose(np.array(ensembleDataset_test_TRANS))
ensembleDataset_eval = np.transpose(np.array(ensembleDataset_eval_TRANS))
return ensembleDataset_train, ensembleDataset_test, ensembleDataset_eval
def sklearnTreeEnsembles(train_MATRIX, train_LABELS, test_MATRIX, eval_MATRIX,
N, mPercent, featureSplit):
lasttestCOLindex = len(test_MATRIX[0]) - 1
test_labels = test_MATRIX[:, lasttestCOLindex]
train_MATRIX = np.delete(train_MATRIX, -1, 1)
test_MATRIX = np.delete(test_MATRIX, -1, 1)
eval_MATRIX = np.delete(eval_MATRIX, -1, 1)
ensembleDataset_train_TRANS = []
ensembleDataset_test_TRANS = []
ensembleDataset_eval_TRANS = []
numSamples = int(math.ceil(len(train_MATRIX) * mPercent))
for numTrees in range(N):
Sdata = np.array(
[random.choice(train_MATRIX) for _ in range(numSamples)])
#Sdata = train_MATRIX
predicted_train_list = []
predicted_test_list = []
predicted_eval_list = []
#
clf = tree.DecisionTreeClassifier(max_features="sqrt")
clf = clf.fit(train_MATRIX, train_LABELS)
predicted_train_list = clf.predict(train_MATRIX)
predicted_test_list = clf.predict(test_MATRIX)
predicted_eval_list = clf.predict(eval_MATRIX)
ensembleDataset_train_TRANS.append(predicted_train_list)
ensembleDataset_test_TRANS.append(predicted_test_list)
ensembleDataset_eval_TRANS.append(predicted_eval_list)
testACCuracy = projectFunctions.accuracyMETRIC(test_labels,
predicted_test_list)
print 'After ', numTrees + 1, ' trees, Test Accuracy: ', testACCuracy
ensembleDataset_train = np.transpose(np.array(ensembleDataset_train_TRANS))
ensembleDataset_test = np.transpose(np.array(ensembleDataset_test_TRANS))
ensembleDataset_eval = np.transpose(np.array(ensembleDataset_eval_TRANS))
return ensembleDataset_train, ensembleDataset_test, ensembleDataset_eval
import timeit
import crossValidationExperiments
import numpy as np
import scipy
import operator
import featureTRANSFORM
train_matrix, train_labels, test_matrix, test_labels, eval_matrix, eval_labels = featureTRANSFORM.eliminateZEROfetures(
'01', -1)
print '\n------------------ Winnow: ------------------'
w_train, numUpdates = winnowFunctions.simpleWinnow(train_matrix, train_labels,
'ones', 2)
train_pred_labels = projectFunctions.prediction_Winnow(train_matrix,
train_labels, w_train)
train_accuracy = projectFunctions.accuracyMETRIC(train_labels,
train_pred_labels)
dummy, fummy, train_f1_score = projectFunctions.f1METRIC(
train_labels, train_pred_labels)
print 'Accuracy : train: ', train_accuracy
#Prediction on test
test_pred_labels = projectFunctions.prediction_Winnow(test_matrix, test_labels,
w_train)
test_accuracy = projectFunctions.accuracyMETRIC(test_labels, test_pred_labels)
test_f1_score = projectFunctions.f1METRIC(test_labels, test_pred_labels)
print 'Accuracy : test: ', test_accuracy
#Prediction on eval.anon
eval_pred_labels = projectFunctions.prediction_Winnow(eval_matrix, eval_labels,
w_train)
eval_accuracy = projectFunctions.accuracyMETRIC(eval_labels, eval_pred_labels)
eval_f1_score = projectFunctions.f1METRIC(eval_labels, eval_pred_labels)
print 'Accuracy : eval.anon: ', eval_accuracy
import bewNeighborFunctions as kNN
import featureTRANSFORM
import projectFunctions as pF
train_matrix, train_labels, test_matrix, test_labels, eval_matrix, eval_labels = featureTRANSFORM.eliminateZEROfetures(
'01', -1)
k = 3
p = 2
'''
pred_train_labels = kNN.kneighbors(train_matrix,train_labels,train_matrix,k,p)
train_accuracy = pF.accuracyMETRIC(train_labels, pred_train_labels)
dum,pum,train_f1 = pF.f1METRIC(train_labels,pred_train_labels)
print 'train: Accuracy:',train_accuracy,'% F1:',train_f1
'''
pred_test_labels = kNN.kneighbors(train_matrix, train_labels, test_matrix, k,
p)
test_accuracy = pF.accuracyMETRIC(test_labels, pred_test_labels)
dum, pum, test_f1 = pF.f1METRIC(test_labels, pred_test_labels)
print 'test: Accuracy:', test_accuracy, '% F1:', test_f1
eval_pred_labels = kNN.kneighbors(train_matrix, train_labels, eval_matrix, k,
p)
solution_filename = raw_input(
'Enter x__x in ./solutions_log/x__x_solutions.csv: ')
pF.write_solutions('perceptron', eval_pred_labels,
'./' + solution_filename + '.solutions.csv')
def algoFORepochs(numEpochs,
trainMatrix,
trainLabels,
testMatrix,
testLabels,
algo_type,
learningRate,
initialWcondition,
margin=None):
#for info ARRAY
infoList = []
temp_list = []
temp_list.append('shuffleTYPE')
temp_list.append('num of Epochs')
temp_list.append('acc_train')
temp_list.append('f1_train')
temp_list.append('acc_test')
temp_list.append('f1_test')
temp_list.append('weights')
#
infoList.append(temp_list)
tempTrainMatrix = copy.deepcopy(trainMatrix)
tempTrainLabels = copy.deepcopy(trainLabels)
best_w = np.zeros(361)
best_train_acc = 0
best_f1_train = 0
av_count = 1
#for noYesShuffle in range(2):
total_numUpdates = 0
w = 0
for e_ind in range(numEpochs):
#shuffling section start
if e_ind == 0:
shuffleTYPE = 'no shuffle'
trainingExMat = trainMatrix
trainExLabels = trainLabels
else:
shuffleTYPE = 'with shuffle'
if e_ind == 0:
trainingExMat = trainMatrix
trainExLabels = trainLabels
else:
trainingExMat, trainExLabels = shuffle(tempTrainMatrix,
tempTrainLabels,
random_state=0)
#trainingExMat, trainExLabels = shuffleFunction(tempTrainMatrix, tempTrainLabels)
tempTrainMatrix = trainingExMat
tempTrainLabels = trainExLabels
# shuffling section end
if e_ind == 0:
weightCondition = initialWcondition
else:
weightCondition = w
if algo_type == 'simple':
w, numUpdates = simplePerceptron(trainingExMat, trainExLabels,
weightCondition, learningRate)
elif algo_type == 'margin':
w, numUpdates = marginPerceptron(trainingExMat, trainExLabels,
weightCondition, learningRate,
margin)
elif algo_type == 'aggressive':
w, numUpdates = agrresiveMarginPerceptron(trainingExMat,
trainExLabels,
weightCondition,
learningRate, margin)
elif algo_type == 'weightedAggressive':
w, numUpdates = WeightedAgrresiveMarginPerceptron(
trainingExMat, trainExLabels, weightCondition, learningRate,
margin)
elif algo_type == 'weightedSimple':
w, numUpdates = weightedPerceptron(trainingExMat, trainExLabels,
weightCondition, learningRate)
elif algo_type == 'winnow':
w, numUpdates = winnowFunctions.simpleWinnow(
trainingExMat, trainExLabels, 'ones', 2)
total_numUpdates += numUpdates
if algo_type == 'winnow':
train_pred_labels = projectFunctions.prediction_Winnow(
trainMatrix, trainLabels, w)
acc_train = projectFunctions.accuracyMETRIC(
trainLabels, train_pred_labels)
test_pred_labels = projectFunctions.prediction_Winnow(
testMatrix, testLabels, w)
acc_test = projectFunctions.accuracyMETRIC(testLabels,
test_pred_labels)
dummyP, dummyR, f1_train = projectFunctions.f1METRIC(
trainLabels, train_pred_labels)
dummyP, dummyR, f1_test = projectFunctions.f1METRIC(
testLabels, test_pred_labels)
else:
train_pred_labels = projectFunctions.prediction_Perceptron(
trainMatrix, trainLabels, w)
acc_train = projectFunctions.accuracyMETRIC(
trainLabels, train_pred_labels)
test_pred_labels = projectFunctions.prediction_Perceptron(
testMatrix, testLabels, w)
acc_test = projectFunctions.accuracyMETRIC(testLabels,
test_pred_labels)
dummyP, dummyR, f1_train = projectFunctions.f1METRIC(
trainLabels, train_pred_labels)
dummyP, dummyR, f1_test = projectFunctions.f1METRIC(
testLabels, test_pred_labels)
#for info ARRAY
temp_list = []
temp_list.append(shuffleTYPE)
temp_list.append(e_ind + 1)
temp_list.append(acc_train)
temp_list.append(f1_train)
temp_list.append(acc_test)
temp_list.append(f1_test)
temp_list.append(w)
#
infoList.append(temp_list)
'''
if acc_train >= best_train_acc:
best_train_acc = acc_train
w_ATbestTRAINacc = w
train_acc_ATbestTRAINacc = acc_train
test_acc_ATbestTRAINacc = acc_test
eval_acc_ATbestTRAINacc = acc_eval
train_f1_ATbestTRAINacc = f1_train
test_f1_ATbestTRAINacc = f1_test
eval_f1_ATbestTRAINacc = f1_eval
'''
if f1_train >= best_f1_train:
best_shuffle_type = shuffleTYPE
best_epoch_index = e_ind
best_acc_train = acc_train
best_f1_train = f1_train
best_acc_test = acc_test
best_f1_test = f1_test
best_w = w
best_LIST = [
best_shuffle_type, best_epoch_index, best_acc_train, best_f1_train,
best_acc_test, best_f1_test, best_w
]
return best_LIST
print tree
depth_list = treeFunctions.find_depths_of_tree(tree)
max_depth = max(depth_list)
list_of_labels = list(Counter(train_actual_labels))
predicted_train_list = treeFunctions.predictFUNCTION(tree, train_matrix,
list_of_labels,
major_trainLABEL)
predicted_test_list = treeFunctions.predictFUNCTION(tree, test_matrix,
list_of_labels,
major_trainLABEL)
predicted_eval_list = treeFunctions.predictFUNCTION(tree, eval_matrix,
list_of_labels,
major_trainLABEL)
#
train_accuracy = projectFunctions.accuracyMETRIC(train_actual_labels,
predicted_train_list)
train_f1_score = projectFunctions.f1METRIC(train_actual_labels,
predicted_train_list)
test_accuracy = projectFunctions.accuracyMETRIC(test_actual_labels,
predicted_test_list)
test_f1_score = projectFunctions.f1METRIC(test_actual_labels,
predicted_test_list)
eval_accuracy = projectFunctions.accuracyMETRIC(eval_actual_labels,
predicted_eval_list)
eval_f1_score = projectFunctions.f1METRIC(eval_actual_labels,
predicted_eval_list)
solution_filename = raw_input(
'Enter x__x in ./solutions_log/x__x_solutions.csv: ')
projectFunctions.write_solutions(
'decision tree', predicted_eval_list,
def simplePerceptron_cvWeight(train_matrix, train_actual_labels, weightRange,
num_folds):
av_accuracy_list = []
for weight in weightRange:
print 'weight: ', weight
'''
###############
kf = KFold(n_splits=5)
temp_train_Matrix = copy.deepcopy(train_matrix)
temp_train_Labels = copy.deepcopy(train_actual_labels)
count = 0
accuracy_list = []
for train_indices, test_indices in kf.split(temp_train_Matrix):
temp_test_Matrix = []
temp_test_Labels = []
for test_indx in test_indices:
count +=1
#print 'count: ', count
temp_test_Matrix.append(temp_train_Matrix[test_indx-count])
temp_test_Labels.append(temp_train_Labels[test_indx-count])
np.delete(temp_train_Matrix, test_indx-count, 0)
np.delete(temp_train_Labels, test_indx-count, 0)
################
'''
accuracy_list = []
training_lab = copy.deepcopy(train_actual_labels)
training_lab = training_lab.tolist()
training_mat = copy.deepcopy(train_matrix)
training_mat = training_mat.tolist()
subset_size = len(training_mat) / num_folds
for i in range(num_folds):
#
if i == num_folds - 1:
temp_test_Matrix = training_mat[(i) * subset_size:]
temp_train_Matrix = training_mat[0:][:i * subset_size]
#
temp_test_Labels = training_lab[(i) * subset_size:]
temp_train_Labels = training_lab[0:][:i * subset_size]
continue
temp_test_Matrix = training_mat[i * subset_size:][:subset_size]
#temp_train_Matrix =np.append(training_mat[:i * subset_size],training_mat[(i + 1) * subset_size:])
temp_train_Matrix = training_mat[:i * subset_size] + training_mat[
(i + 1) * subset_size:]
#
temp_test_Labels = training_lab[i * subset_size:][:subset_size]
temp_train_Labels = training_lab[:i * subset_size] + training_lab[
(i + 1) * subset_size:]
#
# Training
w_train, numUpdates_simpleP = percepFunctions.simplePerceptron(
temp_train_Matrix, temp_train_Labels, weight, 1)
# Prediction on eval.anon
predicted_labels = projectFunctions.prediction_Perceptron(
temp_test_Matrix, temp_test_Labels, w_train)
# Metric
#test_accuracy, test_f1_score = projectFunctions.accuracyMETRIC(temp_test_Labels, predicted_labels)
test_accuracy = projectFunctions.accuracyMETRIC(
temp_test_Labels, predicted_labels)
accuracy_list.append(test_accuracy)
av_accuracy = np.mean(accuracy_list)
av_accuracy_list.append(av_accuracy)
print av_accuracy
best_accuracy = max(av_accuracy_list)
max_index = av_accuracy_list.index(best_accuracy)
best_weight = weightRange[max_index]
#
line, = plt.plot(weightRange, av_accuracy_list)
plt.xlabel('initialization weight')
plt.ylabel('Accuracy')
plt.title('Accuracy v/s Learning weight, for Simple perceptron')
plt.grid(True)
plt.show()
return best_accuracy, weight, av_accuracy_list
bestEpochs = epochsCV ; bestGamma = gamma_; bestC = C_; bestAccuracy = accuracy
wCV = svmFunctions.SVM(train_matrix, train_labels, epochsCV, bestC, bestGamma)
pred_train_list = svmFunctions.prediction(train_matrix, train_labels, wCV)
pred_test_list = svmFunctions.prediction(test_matrix, test_labels, wCV)
acc_train = svmFunctions.accuracyMETRIC(train_labels, pred_train_list)
acc_test = svmFunctions.accuracyMETRIC(test_labels, pred_test_list)
dummyP, dummyR, f1_train = projectFunctions.f1METRIC(train_labels, pred_train_list)
dummyP, dummyR, f1_test = projectFunctions.f1METRIC(test_labels, pred_test_list)
print 'Best gamma(0) : ', bestGamma,' Best C: ', bestC,' Best epochs: ', bestEpochs,' Train Acc:', acc_train,' Train f1:', f1_train,' Test Acc', acc_test,'% Test f1:', f1_test
endTime = timeit.default_timer(); print 'Total run time: ', endTime - startTime,'secs'
'''
print 'Final Support Vector Machines:'
epochsFINAL = 20
CFINAL = 11000
gammaFINAL = 1e-06
wFINAL = svmFunctions.SVM(train_matrix, train_labels, epochsFINAL, CFINAL,
gammaFINAL)
pred_train_list = projectFunctions.prediction_Perceptron(
train_matrix, train_labels, wFINAL)
pred_test_list = projectFunctions.prediction_Perceptron(
test_matrix, test_labels, wFINAL)
acc_train = projectFunctions.accuracyMETRIC(train_labels, pred_train_list)
acc_test = projectFunctions.accuracyMETRIC(test_labels, pred_test_list)
dummyP, dummyR, f1_train = projectFunctions.f1METRIC(train_labels,
pred_train_list)
dummyP, dummyR, f1_test = projectFunctions.f1METRIC(test_labels,
pred_test_list)
print '\tBest gamma(0) : ', gammaFINAL, ' Best C: ', CFINAL, ' Train Acc:', acc_train, ' Train f1:', f1_train, ' Test Acc', acc_test, '% Test f1:', f1_test
1e-06