def classify(data, labels):
method = sys.argv[1].upper()
method_name = {'SVM': 'SVM', 'RF': 'Random Forest'}
test_size_arr = [0.6, 0.5, 0.4]
print '方法：%s' % method_name[method]
for test_size in test_size_arr:
if (method == 'SVM'):
train_score, test_score = svm.run(data, labels, test_size)
elif (method == 'RF'):
train_score, test_score = rf.run(data, labels, test_size)
else:
train_score, test_score = svm.run(data, labels, test_size)
print '============================================='
print '训练集 %s | 测试集 %s' % (1 - test_size, test_size)
print '训练集正确率：%s' % train_score
print '测试集正确率：%s' % test_score
def runSVM(self):
# print("--------TRAINING--------")
if self.fileName != "":
self.splitSize = int(self.split_lineEdit.text())
self.kernelType = self.kernel_cb.currentText()
self.degree = int(self.degree_lineEdit.text())
self.regParam = float(self.regParam_lineEdit.text())
self.tol = float(self.tol_lineEdit.text())
if self.splitSize <=40:
# print("test size =",self.splitSize,"%")
# print("kernel:",self.kernelType)
# print("degree:",self.degree)
# print("tolerance value:",self.tol)
# print("regularization parameter:",self.regParam)
self.results = svm.run(self.fileName,self.splitSize,self.kernelType,self.degree,self.tol,self.regParam)
else:
pass# print("cannot train on such small dataset")
else:
pass# print("incorrect file name!")
# print("--------SUCCESSFUL--------")
QMessageBox.about(self,"Results:", self.results)
def run(k, X, y, gamma):
# Your code goes here return z
n = np.shape(X)[0]
d = np.shape(X)[1]
y_pred = np.zeros(shape=(n, 1))
for i in range(k):
# boundaries for each fold
lower = math.floor(n * i / k)
upper = math.floor((n * (i + 1) / k) - 1)
# testing set
T = list(range(lower, upper + 1))
# training set
S = [*range(lower), *range(upper + 1, n)]
X_train = np.zeros(shape=(len(S), d))
Y_train = np.zeros(shape=(len(S), 1))
# fill up X_train and Y_train according to training set
for j in range(len(S)):
row = S[j]
X_train[j] = X[row]
Y_train[j] = y[row]
# training
clf = svm.run(X_train, Y_train.reshape((len(S))), gamma)
# predicting labels
for t in T:
test_point = X[t].reshape((1, d))
y_pred[t] = svmpredict.run(clf, test_point)
error = np.mean(y != y_pred)
return error
def result():
sns.set(palette="muted", color_codes=True)
# 3a
print '============================================='
print 'HW1-3a'
print '============================================='
# fittingGaussian('average_age')
normalTest(df['average_age'].values, name='Col[7]')
# 3b
print '============================================='
print 'HW1-3b'
print '============================================='
arr_7 = grouping('average_age')
for i in range(0, 5):
normalTest(arr_7[i], name='Col[7] in Group %s' % (i + 1))
print '---------------------------------------------'
leveneTest(arr_7, name='Col[7]')
# 3c
print '============================================='
print 'HW1-3c'
print '============================================='
oneway(arr_7, name='Col[7]')
# 4
print '============================================='
print 'HW1-4'
print '============================================='
selectCol = ['message_number', 'variance_age', 'conversation_number']
for key in selectCol:
normalTest(df[key].values, name='Col[%s]' % key)
print '---------------------------------------------'
for key in selectCol:
normalTest(np.log(df[key].values), name='log Col[%s]' % key)
# 5b
print '============================================='
print 'HW1-5'
print '============================================='
# no box-cox
for key in selectCol:
arr = grouping(key)
oneway(arr, 'Col[%s]' % key)
print '---------------------------------------------'
# box-cox
for key in selectCol:
arr = grouping(key, mode=True)
oneway(arr, 'box-cox Col[%s]' % key)
# 6
print '============================================='
print 'HW1-5'
print '============================================='
testSizeArr = [0.6, 0.5, 0.4]
print "SVM ['average_age', 'variance_age']"
for size in testSizeArr:
train_score, test_score = svm.run(df[['average_age', 'variance_age']],
df['group_category'], size)
print '[train:test = %s:%s] train correct=%s,test correct=%s.' % (
1 - size, size, train_score, test_score)
print '---------------------------------------------'
print "Random Forest ['average_age', 'variance_age']"
for size in testSizeArr:
train_score, test_score = rf.run(df[['average_age', 'variance_age']],
df['group_category'], size)
print '[train:test = %s:%s] train correct=%s,test correct=%s.' % (
1 - size, size, train_score, test_score)
if __name__ == "__main__":
loan_2 = 'data/loan_2.csv'
sourceData = np.loadtxt(loan_2,
dtype=float,
delimiter=',',
converters={0: loan_type},
skiprows=1)
# x是数据 y是标签
y, x = np.split(sourceData, (1, ), axis=1)
method = sys.argv[1].upper()
method_name = {'SVM': 'SVM', 'RF': 'Random Forest'}
test_size_arr = [0.6, 0.5, 0.4]
print '方法：%s' % method_name[method]
for test_size in test_size_arr:
if (method == 'SVM'):
train_score, test_score = svm.run(x, y, test_size)
elif (method == 'RF'):
train_score, test_score = rf.run(x, y, test_size)
else:
train_score, test_score = svm.run(x, y, test_size)
print '============================================='
print '训练集 %s | 测试集 %s' % (1 - test_size, test_size)
print '训练集正确率：%s' % train_score
print '测试集正确率：%s' % test_score
def run(arg_model,
arg_modelname,
arg_train,
arg_test,
arg_features,
arg_featurename,
arg_name,
arg_preprocess,
arg_labels,
arg_dev=True,
arg_hyperopt=False,
arg_dataset="mds",
arg_n_feats=398,
arg_anova="f_classif",
arg_nodes=297,
dataloc="/u/sjeblee/research/va/data/datasets",
arg_rebalance=""):
dataloc = dataloc + "/" + arg_dataset
trainname = arg_train + "_cat" # all, adult, child, or neonate
devname = arg_test + "_cat"
pre = arg_preprocess
print "pre: " + pre
labels = arg_labels
featureset = arg_featurename # Name of the feature set for feature file
features = arg_features # type, checklist, narr_bow, narr_tfidf, narr_count, narr_vec, kw_bow, kw_tfidf, symp_train
modeltype = arg_model # svm, knn, nn, lstm, nb, rf
modelname = arg_modelname
#resultsloc_name = arg_name
# Location of data files
resultsloc = "/u/sjeblee/research/va/data/" + arg_name
heideldir = "/u/sjeblee/tools/heideltime/heideltime-standalone"
scriptdir = "/u/sjeblee/research/va/git/verbal-autopsy"
# Setup
if not os.path.exists(resultsloc):
os.mkdir(resultsloc)
trainset = dataloc + "/train_" + trainname + ".xml"
devset = ""
devfeatures = ""
devresults = ""
if arg_dev:
devset = dataloc + "/dev_" + devname + ".xml"
devfeatures = dataloc + "/dev_" + devname + ".features." + featureset
devresults = resultsloc + "/dev_" + devname + ".results." + modelname + "." + featureset
else:
devset = dataloc + "/test_" + devname + ".xml"
devfeatures = dataloc + "/test_" + devname + ".features." + featureset
devresults = resultsloc + "/test_" + devname + ".results." + modelname + "." + featureset
trainfeatures = dataloc + "/train_" + trainname + ".features." + featureset
element = "narrative"
# Preprocessing
spname = "spell"
print "Preprocessing..."
if "spell" in pre:
print "Running spelling correction..."
trainsp = dataloc + "/train_" + trainname + "_" + spname + ".xml"
devsp = ""
if arg_dev:
devsp = dataloc + "/dev_" + devname + "_" + spname + ".xml"
else:
devsp = dataloc + "/test_" + devname + "_" + spname + ".xml"
if not os.path.exists(trainsp):
print "spellcorrect on train data..."
spellcorrect.run(trainset, trainsp)
if not os.path.exists(devsp):
print "spellcorrect on test data..."
spellcorrect.run(devset, devsp)
trainset = trainsp
devset = devsp
devname = devname + "_" + spname
trainname = trainname + "_" + spname
if "heidel" in pre:
print "Running Heideltime..."
with cd(heideldir):
trainh = dataloc + "/train_" + trainname + "_ht.xml"
if not os.path.exists(trainh):
heidel_tag.run(trainset, trainh)
fixtags(trainh)
trainset = trainh
devh = ""
if arg_dev:
devh = dataloc + "/dev_" + devname + "_ht.xml"
else:
devh = dataloc + "/test_" + devname + "_ht.xml"
if not os.path.exists(devh):
heidel_tag.run(devset, devh)
fixtags(devh)
devset = devh
devname = devname + "_ht"
trainname = trainname + "_ht"
if "medttk" in pre:
print "Running medttk..."
#with cd(heideldir):
trainh = dataloc + "/train_" + trainname + "_medttk.xml"
if not os.path.exists(trainh):
medttk_tag.run(trainset, trainh)
fixtags(trainh)
trainset = trainh
devh = ""
if arg_dev:
devh = dataloc + "/dev_" + devname + "_medttk.xml"
else:
devh = dataloc + "/test_" + devname + "_medttk.xml"
if not os.path.exists(devh):
medttk_tag.run(devset, devh)
fixtags(devh)
devset = devh
devname = devname + "_medttk"
trainname = trainname + "_medttk"
element = "narr_medttk"
if "symp" in pre:
print "Tagging symptoms..."
sympname = "symp"
trainsp = dataloc + "/train_" + trainname + "_" + sympname + ".xml"
#devsp = ""
#if arg_dev:
# devsp = dataloc + "/dev_" + devname + "_" + sympname + ".xml"
#else:
# devsp = dataloc + "/test_" + devname + "_" + sympname + ".xml"
if not os.path.exists(trainsp):
print "tag_symptoms on train data..."
tag_symptoms.run(trainset, trainsp)
#fixtags(trainsp)
#if not os.path.exists(devsp):
# print "tag_symptoms on test data..."
# tag_symptoms.run(devset, devsp)
# fixtags(devsp)
trainset = trainsp
#devset = devsp
#devname = devname + "_" + spname
trainname = trainname + "_" + spname
# Feature Extraction
if arg_dev:
devset = dataloc + "/dev_" + devname + ".xml"
devfeatures = dataloc + "/dev_" + devname + ".features." + featureset
devresults = resultsloc + "/dev_" + devname + ".results." + modelname + "." + featureset
else:
devset = dataloc + "/test_" + devname + ".xml"
devfeatures = dataloc + "/test_" + devname + ".features." + featureset
devresults = resultsloc + "/test_" + devname + ".results." + modelname + "." + featureset
trainfeatures = dataloc + "/train_" + trainname + ".features." + featureset
print "trainfeatures: " + trainfeatures
print "devfeatures: " + devfeatures
stem = False
lemma = False
if "stem" in pre:
stem = True
if "lemma" in pre:
lemma = True
print "stem: " + str(stem) + " lemma: " + str(lemma)
if not (os.path.exists(trainfeatures) and os.path.exists(devfeatures)):
print "Extracting features..."
extract_features_dirichlet.run(trainset, trainfeatures, devset,
devfeatures, features, labels, stem,
lemma, element)
# Rebalance dataset?
if arg_rebalance != "":
rebalancedfeatures = trainfeatures + "." + arg_rebalance
rebalance.run(trainfeatures, rebalancedfeatures, labels, arg_rebalance)
trainfeatures = rebalancedfeatures
# Model
if arg_hyperopt:
print "Running hyperopt..."
model_dirichlet.hyperopt(modeltype, trainfeatures, devfeatures,
devresults, resultsloc, labels)
else:
print "Creating model..."
if modeltype == "nb":
svm.run(modeltype, trainfeatures, devfeatures, devresults, labels)
#elif modeltype == "cnn":
# model_temp.run(modeltype, modelname, trainfeatures, devfeatures, devresults, resultsloc, labels, arg_n_feats, arg_anova, arg_nodes)
else:
model_dirichlet.run(modeltype, modelname, trainfeatures,
devfeatures, devresults, resultsloc, labels,
arg_n_feats, arg_anova, arg_nodes)
# Results statistics
print "Calculating scores..."
results_stats.run(devresults, devresults + ".stats")
print "Done"
dataTypes = ['los', 'heart', 'adult']
dataTypes = ['los', 'heart', 'adult']
dataTypes = ['Heart', 'Adult']
# dataTypes = ['heart']
# dataTypes = ['adult']
scores = []
for dataType in dataTypes:
print(dataType)
scores.append(runAll(dataType))
print('DT')
dt.run(dataType)
print('Neural Network')
network.run(dataType)
print('Boosted')
boosted.run(dataType)
print('SVM')
svm.run(dataType)
print('KNN')
knn.run(dataType)
endTime = time()
print(str(endTime - startTime))
print(scores)
alpha, iterr = kerperceptron.run(5, X_train, y_train)
#make prediction for each test point
index = 0
for t in X_test:
y_pred_perceptron[index] = kerpredict.run(alpha, X_train, y_train, t)
index += 1
perceptron_err = np.mean(y_test != y_pred_perceptron)
perceptron_error.append(perceptron_err)
#svm
(training_length, features) = np.shape(X_train)
#training svm
clf = svm.run(X_train, y_train.reshape((training_length)), 0.001)
#testing svm
y_pred_svm = svmpredict.run(clf, X_test)
y_pred_svm = y_pred_svm.reshape((test_length, 1))
svm_err = np.mean(y_test != y_pred_svm)
svm_error.append(svm_err)
print(svm_error)
print(perceptron_error)
# plotting data
fig, axs = plt.subplots(2)
fig.suptitle('Error Rate vs. Sample Size')
axs[0].plot(subset, svm_error)
axs[0].set_ylabel("SVM Error Rate")
