def run(cross=True, verbose=False, xml="../xml/RTE2_dev.xml", pre_processec_xml="../xml/RTE2_dev.preprocessed.xml"):
learning_data="learningdata.tab" # the data features. extracted from an earlier run of features.
filename = "results_part3.txt"
clean_file(filename)
if cross: features.run(xml, pre_processec_xml) # extracts the features
data = orange.ExampleTable(learning_data)
l = orange.BayesLearner(data)
if cross:
if verbose:
print "result: ", validation(data)
for item in data:
if item.getclass() != l(item):
print '\033[1;41m'
print item, l(item),
print '\033[1;m'
print
else:
print item, l(item)
else:
print "result: ", validation(data)
else:
file = open(filename, "a")
file.write("ranked: no\n")
if file:
for item in data:
s = str(item['id']) +" "+ str(l(item))
file.write(s+"\n")
else:
print "Error opening file"
file.close()
print "finished writing to results_part3"
#run(True, False, "../xml/blind-test-data.xml") # runs the learning
#run(False) # runs the writing to results file.
if cfg.quick == True:
g.run(cfg.name, quick=True)
else:
g.run(cfg.name)
if cfg.roc == True:
import roccurves as rc
if cfg.quick == True:
rc.run(cfg.name + str(cfg.maxdepth), quick=True)
else:
rc.run(cfg.name + str(cfg.maxdepth))
if cfg.features == True:
import features as f
if cfg.quick == True:
f.run(cfg.name + str(cfg.maxdepth), quick=True)
else:
f.run(cfg.name + str(cfg.maxdepth))
if cfg.checksignal == True:
import checksignal as cs
if cfg.quick == True:
cs.run(cfg.name + str(cfg.maxdepth), quick=True)
else:
cs.run(cfg.name + str(cfg.maxdepth))
if cfg.crossvalidation == True:
import crossvalidation as cv
if cfg.quick == True:
cv.run(cfg.name + str(cfg.maxdepth), quick=True)
else:
def single_run(data,
split_rate,
threshold=0,
first_N=300,
window_size=50,
with_pca=False,
n_components=20):
ds = data['ds']
us = data['us']
random.shuffle(ds)
random.shuffle(us)
data['ds'] = ds
data['us'] = us
total_set = []
total_label = []
flag_f1 = False
flag_f2 = False
for i in range(1000):
try:
if len(data['ds'][i][1]) > threshold:
if len(data['ds'][i][3]) > threshold:
total_set.append([
features.run("all", data['ds'][i][1], window_size,
first_N),
features.run("all", data['ds'][i][3], window_size,
first_N)
])
total_label.append(1)
except:
flag_f1 = True
try:
if len(data['us'][i][1]) > threshold:
if len(data['us'][i][3]) > threshold:
total_set.append([
features.run("all", data['us'][i][1], window_size,
first_N),
features.run("all", data['us'][i][3], window_size,
first_N)
])
total_label.append(0)
except:
flag_f2 = True
if flag_f1 and flag_f2:
break
#train and test split
train = total_set[:int(split_rate * len(total_set))]
train_label = total_label[:int(split_rate * len(total_set))]
test = total_set[int(split_rate * len(total_set)):]
test_label = total_label[int(split_rate * len(total_set)):]
# flattens data automaticcaly for all data point
train = np.reshape(
np.asarray(train),
(len(train), len(train[0]) * len(train[0][0]) * len(train[0][0][0])))
test = np.reshape(
np.asarray(test),
(len(test), len(test[0]) * len(test[0][0]) * len(test[0][0][0])))
if with_pca:
pca = PCA(n_components=n_components, svd_solver="arpack")
pca.fit(train)
train = pca.transform(train)
test = pca.transform(test)
print("SVM training started...")
svm = SVC(kernel='linear')
svm.fit(np.asarray(train), np.asarray(train_label))
print("SVM prediction started...")
predictions = svm.predict(np.asarray(test))
accuracy = accuracy_score(test_label, predictions)
print(accuracy)
print("Train split size: " + str(len(train)))
print("Test split size: " + str(len(test)))
def k_fold(data,
K,
threshold=0,
first_N=300,
window_size=50,
with_pca=False,
n_components=20):
test_split_rate = 1 / K
accuracies = []
ds = data['ds']
us = data['us']
random.shuffle(ds)
random.shuffle(us)
data['ds'] = ds
data['us'] = us
total_set = []
total_label = []
flag_f1 = False
flag_f2 = False
for i in range(
1000
): #tries to add all data from all classes, if all flags are true, that means all data collected
try:
if len(data['ds'][i][1]) > threshold:
if len(data['ds'][i][3]) > threshold:
total_set.append([
features.run("all", data['ds'][i][1], window_size,
first_N),
features.run("all", data['ds'][i][3], window_size,
first_N)
])
total_label.append(1)
except:
flag_f1 = True
try:
if len(data['us'][i][1]) > threshold:
if len(data['us'][i][3]) > threshold:
total_set.append([
features.run("all", data['us'][i][1], window_size,
first_N),
features.run("all", data['us'][i][3], window_size,
first_N)
])
total_label.append(0)
except:
flag_f2 = True
if flag_f1 and flag_f2:
break
for i in range(K):
if i + 1 == K: #when i is the last index, test split is not taking from middle, no need for 2 concatenation in train
train = total_set[:int(i * test_split_rate * len(total_set))]
train_label = total_label[:int(i * test_split_rate *
len(total_set))]
test = total_set[int(i * test_split_rate * len(total_set)):]
test_label = total_label[int(i * test_split_rate *
len(total_set)):]
else: #means that test split is taken from middle or from end of the list which creates no problem
train = total_set[:int(i * test_split_rate * len(total_set))] + \
total_set[int((i+1) * test_split_rate * len(total_set)):]
train_label = total_label[:int(i * test_split_rate * len(total_set))] + \
total_label[int((i+1) * test_split_rate * len(total_set)):]
test = total_set[int(i * test_split_rate *
len(total_set)):int((i + 1) *
test_split_rate *
len(total_set))]
test_label = total_label[int(i * test_split_rate *
len(total_set)):int((i + 1) *
test_split_rate *
len(total_set))]
#flattens data automaticcaly for all data point
train = np.reshape(np.asarray(train),
(len(train), len(train[0]) * len(train[0][0]) *
len(train[0][0][0])))
test = np.reshape(
np.asarray(test),
(len(test), len(test[0]) * len(test[0][0]) * len(test[0][0][0])))
if with_pca:
pca = PCA(n_components=n_components, svd_solver="arpack")
pca.fit(train)
train = pca.transform(train)
test = pca.transform(test)
print("SVM training started...")
svm = SVC(kernel='linear')
svm.fit(np.asarray(train), np.asarray(train_label))
print("SVM prediction started...")
predictions = svm.predict(np.asarray(test))
accuracy = accuracy_score(test_label, predictions)
accuracies.append(accuracy)
print("Accuracies: ", end='')
print(accuracies)
print("Average accuracy: ", end='')
print(sum(accuracies) / len(accuracies))
print("Train split size: " + str(len(train)))
print("Test split size: " + str(len(test)))
return accuracies
if cfg.quick == True:
g.run(cfg.name, quick = True)
else:
g.run(cfg.name)
if cfg.roc == True:
import roccurves as rc
if cfg.quick == True:
rc.run(cfg.name + str(cfg.maxdepth), quick = True)
else:
rc.run(cfg.name + str(cfg.maxdepth))
if cfg.features == True:
import features as f
if cfg.quick == True:
f.run(cfg.name + str(cfg.maxdepth), quick = True)
else:
f.run(cfg.name + str(cfg.maxdepth))
if cfg.checksignal == True:
import checksignal as cs
if cfg.quick == True:
cs.run(cfg.name + str(cfg.maxdepth), quick = True)
else:
cs.run(cfg.name + str(cfg.maxdepth))
if cfg.crossvalidation == True:
import crossvalidation as cv
if cfg.quick == True:
cv.run(cfg.name + str(cfg.maxdepth), quick = True)
else:
