def ejer5_loss(loss_fun, act_fun_last, label, nfig1, nfig2):
print(label)
reg1 = regularizador.L2(0.1)
reg2 = regularizador.L2(0.1)
proto = clasificador.Classifier(epochs=300, batch_size=50, eta=0.001)
outputfile = "ejer5_" + label + "_v3.dat"
(x_train, y_train), (x_test, y_test) = datasets.cifar10.load_data()
mean_train = x_train.mean()
n_clasifi = 10
X, Y = clasificador.flattening(x_train, y_train, n_clasifi, mean_train)
X_test, Y_test = clasificador.flattening(x_test, y_test, n_clasifi,
mean_train)
proto.fit(X,
Y,
X_test,
Y_test,
act_function1=act.ReLU(0),
reg1=reg1,
loss_function=loss_fun,
act_function2=act_fun_last,
reg2=reg2)
# plt.figure(nfig1)
# plt.ylabel("Accuracy [%]")
# plt.plot(proto.acc_vect, label="Entrenamiento", c='red', alpha=0.6, ls='--')
# plt.plot(proto.pres_vect, label="Validación", c='blue', alpha=0.6)
# plt.legend(loc=0)
# plt.savefig("ejer5_acc_"+label+".pdf")
# plt.figure(nfig2)
# plt.ylabel("Pérdida")
# plt.plot(proto.loss_vect, label="Entrenamiento", c='red', alpha=0.6, ls='--')
# plt.plot(proto.loss_test, label="Validación", c='blue', alpha=0.6)
# plt.legend(loc=0)
# plt.savefig("ejer5_loss_"+label+".pdf")
# plt.show()
#plt.close()
#plt.clf()
np.savetxt(
outputfile,
np.array([
proto.acc_vect, proto.pres_vect, proto.loss_vect, proto.loss_test
]).T)
def evaluate_classifier(model_filename):
with tf.Graph().as_default():
classfier_net = classifier.Classifier()
classification_model = classfier_net.get_model(122, 122)
classification_model.load(model_filename)
annotations = data.load_annotations()
image_list = data.create_image_list(annotations)
ok = 0
n = 0
with open(predicted_annotations_path) as data_file:
bounding_box_data = json.load(data_file)
#bounding_box_data = data.load_annotations()
for filepath in image_list:
x = int(bounding_box_data[filepath][0])
w = int(bounding_box_data[filepath][1])
y = int(bounding_box_data[filepath][2])
h = int(bounding_box_data[filepath][3])
# Extending bounding box by 10%
x -= int(0.1 * w)
w += int(0.2 * w)
y -= int(0.1 * h)
h += int(0.2 * h)
crop = cv2.imread(data.get_image_path(filepath))
crop = crop[y:y + h, x:x + w]
if crop is None:
continue
height, width, _ = crop.shape
if height == 0 or width == 0:
continue
crop = cv2.resize(crop, (122, 122))
classification = classification_model.predict([crop])[0]
if data.classes[np.argmax(classification)] == data.get_image_label(
filepath):
ok += 1
#print(data.classes[np.argmax(classification)] + " " + data.get_image_label(filepath))
n += 1
print(ok / n)
def main():
config = get_config()
random.seed(config["algorithm"]["random_seed"])
train_data, test_data = acq.extract_data(config)
if config["verbose"]["enabled"]:
acq.show_data(train_data[0], config["verbose"]["sample_timeout"])
algo = classifier.Classifier(config["algorithm"]["random_seed"])
algo.train(train_data[0], train_data[1])
estimates = algo.test(test_data[0])
algo.eval(estimates, test_data[1])
show_code_in_action(test_data, algo)
print("Finished algorithm.")
def _custom_classifier(features):
clsf = classifier.Classifier(CLASSES, TRAINING_SAMPLES_LIMIT, features, 'avg')
_train_classifier_from_data(clsf)
clsf.finalize()
stats = _test_custom_classifier(clsf)
for cls in sorted(stats.keys()):
util.debug("Class {} => {}% correct ({} total)".format(cls, round(stats[cls]['accuracy'] * 100, 1), stats[cls]['total']))
# And some machine-readable output suitable for appending to a CSV for
# analysis
feature_id = ','.join([ d['name'] for d in features ])
print("decision_mode,features,class,total,accuracy")
for cls in stats:
print("{},\"{}\",{},{},{}".format(clsf.decision_mode, feature_id, cls, stats[cls]['total'], stats[cls]['accuracy']))
def main():
makeSub = True
featureImportance = False
cvfold = True
df = pd.read_csv('../data/cprobTrain15NA.csv')
X, y = np.array(pd.read_csv('../data/train.csv',
usecols=range(1, 9))), np.array(
pd.read_csv('../data/train.csv').ACTION)
X = np.hstack((X, np.array(df)))
params = {
'max_depth': 4,
'subsample': 0.5,
'verbose': 0,
'random_state': 1337,
'min_samples_split': 10,
'min_samples_leaf': 10,
'max_features': 10,
'n_estimators': 350,
'learning_rate': 0.05
}
clf = GradientBoostingClassifier(**params)
prefix = 'lib/gbm350d4m10c15'
if cvfold:
c = classifier.Classifier(X, y)
c.validate(clf, nFolds=10, out=prefix + 'Train.csv')
if makeSub:
Xt = np.array(pd.read_csv('../data/test.csv', usecols=range(1, 9)))
Xt = np.hstack(
(Xt, np.array(pd.read_csv('../data/cprobTest15NA.csv'))))
clf.fit(X, y)
y_ = clf.predict_proba(Xt)[:, 1]
out = pd.read_csv('subs/nbBaseTest.csv')
out.ACTION = y_
out.to_csv(prefix + 'Test.csv', index=False)
if featureImportance:
print "Feature ranking:"
importances = clf.feature_importances_
indices = np.argsort(importances)[::-1]
np.savetxt('indices.txt', indices, delimiter=',')
for f in xrange(df.shape[1]):
print "%d. feature (%s,%f)" % (f + 1, df.columns[indices[f]],
importances[indices[f]])
def predict():
""" PREDICT THE POTENTIAL PIECE OF EQUIPMENT BASED ON THE REQUEST """
"""
RECEIVING JSON FORMAT:
{
"body" :
"imgsource" : *base64 value*
}
"""
# Extract infomation from the JSON
bodyJson = request.get_json(force=True).get('body')
data = bodyJson.get('imgsource')
path = "./temp/test.png"
# Decode the BASE64 image and save it into the temp file
imgdata = base64.b64decode(str(data))
image = Image.open(BytesIO(imgdata))
img = cv2.cvtColor(np.array(image), cv2.COLOR_BGR2RGB)
cv2.imwrite(path, img)
r = requests.post(
'https://api.remove.bg/v1.0/removebg',
files={'image_file': open('./temp/test.png', 'rb')},
data={
'size': 'auto',
'bg_color': 'white'
},
headers={'X-Api-Key': 'iTUtE8hsnt76HMLmfjPAi2hp'},
)
if r.status_code == requests.codes.ok:
with open('./temp/test.png', 'wb') as out:
out.write(r.content)
else:
print("Error:", r.status_code, r.text)
# Call the image recognition function
predictor = classifier.Classifier()
predicted_label = predictor.image_recognition('./temp/test.png')
print(predicted_label)
max_index = np.argmax(predicted_label[0])
probability = str(float(predicted_label[0][max_index] * 100))
data_response = {MAP.get(max_index): probability}
response_json = json.dumps(data_response, indent=4)
return Response(response=response_json, status=200)
def main(argv):
help = 'main.py [-t -v -f --test]\n'
help += '-t or --twitter : runs classifier within twitter GUI\n'
help += '-v or --verbose : prints extra information\n'
help += '-f [folds] or --folds=[folds] : set number of folds '
help += 'for validation\n'
help += '--test : runs test code\n'
help += '--timing : runs timing code\n'
try:
opts, args = getopt.getopt(
argv, "htcvf:",
['twitter', 'classifier', 'verbose', 'test', 'timing', 'folds='])
except getopt.GetoptError:
print(help)
sys.exit(2)
twitter = False
verbose = False
test = False
timing = False
folds = 5
for opt, arg in opts:
if opt == '-h':
print(help)
sys.exit()
elif opt in ('-t', '--twitter'):
twitter = True
elif opt in ('-v', '--verbose'):
verbose = True
elif opt in ('--test'):
test = True
elif opt in ('-f', '--folds'):
folds = arg
elif opt in ('--timing'):
timing = True
if test:
subprocess.call(['pytest', '..\\Test\\test_net.py'])
else:
clf = classifier.Classifier(folds=folds, timing=timing)
if verbose:
clf.run(verb=True)
else:
clf.run()
if twitter:
win = twitterGUI.TwitWindow(clf)
win.CreateWindow()
def init_classifier(self):
'''
SVM initialization, stores the list of SVM objects in a list
at instance level
'''
svms = []
print '\nSVM initialization:'
startt = time.time()
t_lab, trn, v_lab, val = self.init_ds()
for n in self.names:
tmp = classifier.Classifier(self.get_lab(n, t_lab), trn, \
self.get_lab(n, v_lab), val, n)
print '\tSVM for %s initialized' % n
svms.append(tmp)
print 'Classifier initialized in %s sec.\n' % \
(time.time() - startt)
self.svms = svms
def __init__(self, classifier_folder):
self.class_name = classifier_folder
self.image_size = (227, 227)
net_model = classifier_folder + "/net.caffemodel"
net_proto = classifier_folder + "/deploy.prototxt"
net_label = classifier_folder + "/labels.txt"
self.classifier = classifier.Classifier(image_size=self.image_size,
net=net_proto,
weight=net_model,
class_label=net_label)
labels = self.classifier.get_label_list()
super(UI, self).__init__()
self.setGeometry(300, 300, 720, 720)
self.setWindowTitle(self.class_name)
self.w = dragarea()
self.w.dropEvent = self.load_images_dd
self.root_vbox = QtGui.QVBoxLayout()
self.w.setLayout(self.root_vbox)
self.setCentralWidget(self.w)
self.detail_btn = QtGui.QPushButton("Details")
self.detail_btn.clicked.connect(self.click_detail)
self.detail_w = detail_panel(labels)
self.detail_page = QtGui.QWidget()
self.controller_hbox = QtGui.QHBoxLayout()
self.detail_page.setLayout(self.controller_hbox)
self.prev_image = QtGui.QPushButton(" < prev")
self.prev_image.clicked.connect(self.prev_detail)
self.next_image = QtGui.QPushButton(" next >")
self.next_image.clicked.connect(self.next_detail)
self.return_table = QtGui.QPushButton(" score table ")
self.return_table.clicked.connect(self.display_table)
self.controller_hbox.addWidget(self.prev_image)
self.controller_hbox.addWidget(self.return_table)
self.controller_hbox.addWidget(self.next_image)
self.image_added = False
self.image_acceptable = True
self.detail_view = False
self.detail_index = -1
self.table = None
self.images = []
self.scores = []
self.show()
def trainClassifer (third_dataset, learning_rate=0.0001, decay_rate=0.99,
batch_size=10, training_epochs=10, display_step=1, n_samples=1000, noise=1):
cl = classifier.Classifier(learning_rate=learning_rate)
for epoch in range(training_epochs):
avg_cost = 0.
total_batch = int(n_samples / batch_size)
for i in range(total_batch):
batch_xs, batch_ys = third_dataset[i*batch_size: (i+1)*batch_size], train_labels[i*batch_size: (i+1)*batch_size]
cl.train_step.run({cl.x: batch_xs, cl.y_: batch_ys})
# Display logs per epoch step
if epoch % 10 == 0:
print("Epoch:", '%04d' % (epoch+1), \
"cross_entropy=", "{:.9f}".format(cl.cross_entropy.eval({cl.x: third_dataset, cl.y_: train_labels[0:n_samples]})))
cl.learning_rate *= decay_rate
# Test trained model
print(cl.accuracy.eval({cl.x: third_dataset, cl.y_: train_labels[0:n_samples]}))
def HyperSearch():
# Courtesy of Miroslaw Horbal
base = [127, 96, 53, 3, 103, 71, 151, 1, 65, 152]
f = fileio.Preprocessed('../data/quads10Threshold.csv')
f.encode(base)
train, truth = f.transformTrain(base)
print "Performing hyperparameter selection..."
clf = LogisticRegression(C=2.3, class_weight='auto')
# Hyperparameter selection loop
score_hist = []
Cvals = np.linspace(1, 4, 32)
eval_ = classifier.Classifier(train, truth)
for C in Cvals:
clf.C = C
score = eval_.holdout(clf, nFolds=10, fraction=0.2)
score_hist.append((score, C))
print "C: %f Mean AUC: %f" % (C, score)
bestC = sorted(score_hist)[-1][1]
print "Best C value: %f" % (bestC)
def classify_posts(posts):
with classifier.Classifier() as c:
for post in posts:
timestamp = post['taken_at_timestamp']
picture_url = post['display_url']
print(f"checking url {picture_url}")
if queries.image_exists(picture_url):
print(f"\talready processed. skipping {picture_url}")
continue
print(f"\timage not loaded, adding to db {picture_url}")
image = Image(url=picture_url, post_date=timestamp)
print(f"\tclassifying image {picture_url}")
c.classify(image)
queries.add(image)
queries.commit()
def classification_test(verbose=True):
rdr = reader.Reader()
clssfr = classifier.Classifier()
test_data_tables = rdr.get_classifier_test_data_tables()
correct_count = 0
total_count = 0
for data_table in test_data_tables:
col_idx = 0
while True:
col = data_table.get_col(col_idx)
if col == None: # out of columns to read
break
(header, records) = col
classified_type = clssfr.classify(col_idx, header, records)
correct_type = data_table.get_type(col_idx)
if verbose or classified_type != correct_type:
print("Column :", data_table.csv_file,
"(Column " + str(col_idx) + ")")
print("Classified:", classified_type)
print("Correct :", correct_type)
if verbose and classified_type == correct_type:
print()
if classified_type == correct_type:
correct_count += 1
else:
print("Header :", repr(header))
print("Records :", records)
print()
total_count += 1
col_idx += 1
print("===================================================")
print("Overall result:",
str(correct_count) + "/" + str(total_count),
"(" + str(round(correct_count / total_count * 100, 2)) + "%)",
"correct classifications")
def fleetclassify_status(status):
#pp.pprint(status)
image_url = get_image_url_from_status(status)
print("image: " + str(image_url))
if not image_url:
return None
image_filepath = download_image_file(image_url)
print("image filepath: " + str(image_filepath))
image_paths = [image_filepath]
result_txt = ''
classifier_ = classifier.Classifier(LEARNED_MODEL_FILEPATH)
predictions = classifier_.classify(image_paths)
for image_path, prediction in zip(image_paths, predictions):
if prediction < 0.5:
result_txt += "艦これ"
else:
result_txt += "アズールレーン"
result_txt += ' (%.3f)' % (prediction)
return result_txt
"--yolo",
default='make_model_classifier/yolo-coco',
help="base path to YOLO directory")
ap.add_argument("-c",
"--confidence",
type=float,
default=0.5,
help="minimum probability to filter weak detections")
ap.add_argument("-t",
"--threshold",
type=float,
default=0.3,
help="threshold when applying non-maxima suppression")
args = vars(ap.parse_args())
car_color_classifier = classifier.Classifier()
# load the COCO class labels our YOLO model was trained on
labelsPath = os.path.sep.join([args["yolo"], "coco.names"])
LABELS = open(labelsPath).read().strip().split("\n")
# initialize a list of colors to represent each possible class label
np.random.seed(42)
COLORS = np.random.randint(0, 255, size=(len(LABELS), 3), dtype="uint8")
# derive the paths to the YOLO weights and model configuration
weightsPath = os.path.sep.join([args["yolo"], "yolov3.weights"])
configPath = os.path.sep.join([args["yolo"], "yolov3.cfg"])
# load our YOLO object detector trained on COCO dataset (80 classes)
# print("[INFO] loading YOLO from disk...")
pcad_index = int(list(sys.argv)[1])
window_sizes = [None, 5, 10, 50, 100, 200]
window_size = window_sizes[pcad_index % 6]
if pcad_index > 5:
baselined = True
output_txt_dir = os.path.join(
out_dir,
'increasing_ordered_PCA_comps_ws_%s_baselined.txt' % str(window_size))
else:
baselined = False
output_txt_dir = os.path.join(
out_dir, 'increasing_ordered_PCA_comps_ws_%s.txt' % str(window_size))
test_class = cl.Classifier()
test_class.recordings = recordings
test_class.pre_trial_window = 2
test_class.post_trial_window = 2
test_class.make_unit_response(['20Hz_cor_AB', '20Hz_acor_BA', '20Hz_acor_AB'],
baseline=baselined)
test_class.test_size = 0.2
n_components = 599
pcad_response, y_var = test_class.make_pcad_response(
n_components, ['20Hz_cor_AB', '20Hz_acor_AB', '20Hz_acor_BA'],
reassign_y_var=[['20Hz_acor_AB', '20Hz_acor_BA']],
window_size=window_size)
with open(output_txt_dir, 'a') as f:
all_accs = []
def predict_aspect_file(self, inputFilePath, outputFilePath, mindmap,
language):
config_path = 'sentiment_config.ini'
import pickle
classifier = pickle.load(open("english_classifier.pickle", 'rb'),
fix_imports=True,
encoding="latin1")
print(1)
try:
clf = classifier.Classifier(config_path=config_path,
language=language)
except:
self.messageBoxSignal.emit("Error",
"Failed to load the classifier file")
self.clear()
return
#analyzer = aspect_detector.AspectDetector()
asa = aspect_sentiment_analyzer.AspectSentimentAnalyzer()
try:
data, fieldName = utils.read_excel_dict(inputFilePath)
#data.columns = map(str.lower, data.columns)
except:
self.messageBoxSignal.emit("Error",
"Failed to read the input file")
self.clear()
return
full_df = pd.DataFrame()
'''
try:
verbatim = str(data[0]['verbatim'])
except:
self.messageBoxSignal.emit("Error","There is no 'verbatim' column in the input file")
self.clear()
return
try:
verbatim = str(data[0]['date'])
except:
self.messageBoxSignal.emit("Error","There is no 'date' column in the input file")
self.clear()
return
'''
progress_step = int(len(data) / 80)
c = 0
step = 20
for row in data:
verbatim = str(row['verbatim'])
cleaned_verbatim = clf.clean_verbatim(verbatim)
# return dictionary consist of key as sentences, and value as a tuple consist of category and subcategory
predicted_categories, words_in_mindmap_per_sentence = asa.predict_verbatim_aspect(
verbatim.lower(), mindmap)
rest_data_frame_columns = pd.DataFrame(row, index=[0])
rest_data_frame_columns = rest_data_frame_columns.loc[:,
rest_data_frame_columns
.columns !=
'verbatim']
words_in_mindmap_per_verbatim = utils.change_listoflist_to_list(
words_in_mindmap_per_sentence)
words_in_mindmap_per_verbatim = ','.join(
words_in_mindmap_per_verbatim)
#words_in_mindmap_per_sentence=utils.change_listoflist_to_listofstrings(words_in_mindmap_per_sentence)
# get month from 'Response Date' Column
if '/' in str(row['date']):
date = str(row['date']).split("/")
elif '-' in str(row['date']):
date = str(row['date']).split("-")
else:
self.messageBoxSignal.emit(
"Error",
"Please check date format in the file , it should have '/' or '-' separator"
)
self.clear()
return
month = calendar.month_name[int(date[1])]
#mindmap_index=0
try:
sentiment, cleaned_sentence = clf.predict_sentence(
cleaned_verbatim)
except:
self.messageBoxSignal.emit("Error",
"Failed to predict sentiment")
self.clear()
return
if len(predicted_categories) > 0:
for sentence in predicted_categories:
category_subcategory = predicted_categories[sentence]
# check if after cleaning there are still any word in the sentence
rest_data_frame_columns['month'] = month
rest_data_frame_columns['original verbatim'] = verbatim
rest_data_frame_columns[
'cleaned verbatim'] = cleaned_verbatim
rest_data_frame_columns['sentiment'] = sentiment
for category, subcategory in category_subcategory:
rest_data_frame_columns['category'] = category
rest_data_frame_columns['subcategory'] = subcategory
#rest_data_frame_columns['words in mindmap'] = words_in_mindmap_per_verbatim + '| ' + words_in_mindmap_per_sentence[mindmap_index]
rest_data_frame_columns[
'words in mindmap'] = words_in_mindmap_per_verbatim
full_df = pd.concat([full_df, rest_data_frame_columns],
sort=False,
axis=0)
elif cleaned_verbatim != '':
rest_data_frame_columns['month'] = month
rest_data_frame_columns['original verbatim'] = verbatim
rest_data_frame_columns['cleaned verbatim'] = cleaned_verbatim
rest_data_frame_columns['sentiment'] = sentiment
rest_data_frame_columns['category'] = "Not Defined"
rest_data_frame_columns['subcategory'] = "Not Defined"
rest_data_frame_columns['words in mindmap'] = "No Words"
full_df = pd.concat([full_df, rest_data_frame_columns],
sort=False,
axis=0)
#mindmap_index+=1
c += 1
if c % progress_step == 0:
step += 1
if step <= 99:
self.countChanged.emit(step)
try:
self.countChanged.emit(100)
full_df.drop_duplicates(inplace=True)
full_df.to_csv(outputFilePath, encoding='utf-8', index=False)
self.messageBoxSignal.emit("Done",
"The file was written successfully")
self.clear()
except:
self.messageBoxSignal.emit("Error",
"Failed to write the file correctly")
self.clear()
return
if args.study:
# To conduct a study with n number of trials as parameter and the type of the model
study = paramstudy.conduct_study(args.n_trials, args.model)
if args.plot == True:
# Functionality for loading a study dump is dropped in the last build since it was not a necessity
# and only used during development, but it can be used by removing the comment sign on the next line.
# study = joblib.load("Results/Study/convpool_study_6.pkl")
paramstudy.generate_graphs_from_study(study)
exit()
#Load the cfg for the selected model
if (args.model == "capsnet"):
cfg = variables.capsnet_cfg
elif (args.model == "convpool"):
cfg = variables.convpool_cfg
# Initializate the model
TClassifier = classifier.Classifier(cfg)
# If selected, load the weights from file
if args.load_weights is not None:
TClassifier.load_weights(variables.saved_weights_path + args.load_weights)
print("Weights loaded.")
#If train is selected, train the model
if args.train:
TClassifier.load_images()
TClassifier.train(cfg["epochs"], cfg["mini_batch_size"], cfg["test_batch_size"])
#Plot the results of the training if required
if args.plot:
TClassifier.plot_loss()
TClassifier.plot_accuracy()
TClassifier.plot_test_accuracy()
# Evaluate the model
n_steps = 100 # number of steps in game
C = 1.0 # classifier parameter
norm_trsh = 0.001
train_dataset, train_labels, valid_dataset, valid_labels, test_dataset, test_labels = dp.read_data(
)
m = len(train_dataset[0])
print('data read complete')
steps = np.arange(n_initial, len(train_labels),
int(np.ceil((len(train_labels) - n_initial) / n_steps)))
svm1 = svm.SVC(kernel='linear', C=C).fit(train_dataset, train_labels)
err_best = 1 - accuracy_score(test_labels, svm1.predict(test_dataset))
print('best possible performance on test dataset is ', 1 - err_best)
classifiers = []
classifier1 = cl.Classifier(train_dataset[:n_initial, :],
train_labels[:n_initial], valid_dataset,
valid_labels, C, 8.2)
err_test1 = 1 - accuracy_score(test_labels, classifier1.predict(test_dataset))
print('classifier1: initiated, error on test dataset is ', err_test1)
classifiers.append(classifier1)
classifier2 = cl.Classifier(train_dataset[:n_initial, :],
train_labels[:n_initial], valid_dataset,
valid_labels, C, 8, 'asc')
err_test2 = 1 - accuracy_score(test_labels, classifier2.predict(test_dataset))
print('classifier2: initiated, error on test dataset is ', err_test2)
classifiers.append(classifier2)
classifier3 = cl.Classifier(train_dataset[:n_initial, :],
train_labels[:n_initial], valid_dataset,
valid_labels, C, 12, 'desc')
import base64
import json
import os
import tempfile
import classifier
import picto_matcher
import web
urls = ('/post_capture', 'PostCapture')
db_file = './../synsets/database/synset.sqlite3'
db = web.database(dbn='sqlite', db=db_file)
cl = classifier.Classifier()
app = web.application(urls, globals())
class PostCapture(object):
def POST(self):
data = web.input()
wnid = data["wnid"]
imagestring = data["captured_picto"]
fd, path = tempfile.mkstemp(suffix=".jpeg", prefix='pc_')
#-------Samples of the training set to reducte computational time-------# # n_train_s = 80000 # X_train_s, y_train_s = sampling(X_train, y_train, n_train_s, balanced=False) # text+= '\n-sampled train = True (%s samples)'%n_train_s #---------------- #print 'classRatio = ', float(len(y_train[y_train == 1]))/len(y_train[y_train == 0]) ####################################################################################################################### # Train and predict # ####################################################################################################################### print '--Training \n' clf = classifier.Classifier(verbose=True) start = time.time() #clf.fit(X_train_s, y_train_s) clf.fit(X_train, y_train) #weight = train_mrf(y_pred_svm_train, img_train, nb_class, max_map_iter, [alpha, beta, sigma_blur], threshold_learning, y_train, threshold_sensitivity, threshold_error) training_time = time.clock() - start print '\n--Prediction \n' start = time.time() y_pred = parallized_pred(X_test, clf) y_pred_train = parallized_pred(X_train, clf) prediction_time = time.time() - start weight = train_mrf(y_pred_train, img_train, nb_class, max_map_iter,
'learning_rate': 0.05,
'n_estimators': 100,
'max_depth': 4,
'subsample': 0.5,
'n_jobs': 4,
'min_child_weight': 15
}
train_params = {'early_stopping_rounds': 10, 'verbose': 0}
xgb = XGBClassifier(**params)
xgb.set_params(**train_params)
xgb_kin = clf.Classifier(model=xgb,
cv=skf,
variables=var_kin,
model_name='XGBoost',
var_name='kinetic',
fig_name='xgb',
train_params=train_params)
xgb_kin.fit(train)
xgb_kin.check_ks_and_cvm(train,
check_agreement=check_agreement,
check_correlation=check_correlation)
xgb_kin.predict(data=test)
params = {
'learning_rate': 0.05,
'n_estimators': 200,
'max_depth': 4,
'subsample': 0.5,
'n_jobs': 4,
'min_child_weight': 15
def classify_context(self, raw_text=None, type="string"):
cl_o = cl.Classifier()
return cl_o.classifier_handler(raw_text)
else:
########## generate image features and classification evalution #########
generator = Generator(latent_dim,class_embed_dim,feature_dim)
if torch.cuda.is_available():
generator = generator.cuda()
checkpoint = torch.load(Model_GAN_path)
print("===> Loading Wasserstein_GAN Model... Start Epoch:{}".format(checkpoint['epoch']))
generator.load_state_dict(checkpoint['G_state_dict'])
if validation:
# for fake data generation
unique_attributes_val, unique_labels_val = get_unique_vector(data.attributes_val,data.labels_val)
gen_features,gen_labels = generate_img_feature(generator,unique_attributes_val,unique_labels_val)
gen_labels = map_label(gen_labels.astype(int))
cls = classifier.Classifier(gen_features,gen_labels,data,zsl_classifier_path,lr=0.0001,batch_size=64,epoch=100,validation=True,generalized=False)
unseen_acc = cls.unseen_acc
else:
if generalized:
#unique_attributes_trainval, unique_labels_trainval = get_unique_vector(data.attributes_train, data.labels_train)
#gen_features_trainval,gen_labels_trainval = generate_img_feature(generator,unique_attributes_trainval,unique_labels_trainval)
unique_attributes_test_unseen, unique_labels_test_unseen = get_unique_vector(data.attributes_test_unseen, data.labels_test_unseen)
gen_features_test_unseen,gen_labels_test_unseen = generate_img_feature(generator,unique_attributes_test_unseen,unique_labels_test_unseen)
features_train = np.concatenate((data.features_train, gen_features_test_unseen), axis=0)
labels_train = np.concatenate((data.labels_train,gen_labels_test_unseen),axis=0)
labels_train = np.where(labels_train.astype(int)==1)[1]
cls = classifier.Classifier(features_train,labels_train,data,gzsl_classifier_path,lr=0.0001,batch_size=64,epoch=100,validation=False,generalized=True)
best_H, seen_acc, unseen_acc = cls.best_H, cls.seen_acc, cls.unseen_acc
else:
unique_attributes_test_unseen, unique_labels_test_unseen = get_unique_vector(data.attributes_test_unseen, data.labels_test_unseen)
import pandas as pd
from sklearn.cross_validation import StratifiedShuffleSplit
import classifier
from sklearn.calibration import CalibratedClassifierCV
from sklearn.metrics import accuracy_score, roc_curve, auc
target_column_name = 'TARGET'
# point it to your training file
filename = '../data/public/train.csv'
if __name__ == '__main__':
df = pd.read_csv(filename)
y = df[target_column_name].values
X = df.drop(target_column_name, axis=1).values
skf = StratifiedShuffleSplit(y, n_iter=2, test_size=0.5, random_state=57)
for valid_train_is, valid_test_is in skf:
X_valid_train = X[valid_train_is]
y_valid_train = y[valid_train_is]
X_valid_test = X[valid_test_is]
y_valid_test = y[valid_test_is]
#clf = model.Classifier()
#clf_c = CalibratedClassifierCV(clf, cv=2, method='isotonic')
clf_c = classifier.Classifier()
clf_c.fit(X_valid_train, y_valid_train)
y_valid_pred = clf_c.predict(X_valid_test)
y_valid_proba = clf_c.predict_proba(X_valid_test)
fpr, tpr, _ = roc_curve(y_valid_test, y_valid_proba[:,1])
print 'auc = ', auc(fpr, tpr)
print 'accuracy = ', accuracy_score(y_valid_pred, y_valid_test)
def get_news(sources=['spiegel','faz','welt','zeit']):
'''
Collects all news articles from political ressort of major German newspapers
Articles are transformed to BoW vectors and assigned to a political party
For better visualization, articles' BoW vectors are also clustered into topics
INPUT
folder the model folder containing classifier and BoW transformer
sources a list of strings for each newspaper for which a crawl is implemented
default ['zeit','sz']
'''
import classifier
from bs4 import BeautifulSoup
from api import fetch_url
import urllib2
articles = []
# the classifier for prediction of political attributes
clf = classifier.Classifier(train=False)
for source in sources:
if source is 'spiegel':
# fetching articles from sueddeutsche.de/politik
url = 'http://www.spiegel.de/politik'
site = BeautifulSoup(urllib2.urlopen(url).read())
titles = site.findAll("div", { "class" : "teaser" })
urls = ['http://www.spiegel.de'+a.findNext('a')['href'] for a in titles]
if source is 'faz':
# fetching articles from sueddeutsche.de/politik
url = 'http://www.faz.net/aktuell/politik'
site = BeautifulSoup(urllib2.urlopen(url).read())
titles = site.findAll("a", { "class" : "TeaserHeadLink" })
urls = ['http://www.faz.net'+a['href'] for a in titles]
if source is 'welt':
# fetching articles from sueddeutsche.de/politik
url = 'http://www.welt.de/politik'
site = BeautifulSoup(urllib2.urlopen(url).read())
titles = site.findAll("a", { "class" : "as_teaser-kicker" })
urls = [a['href'] for a in titles]
if source is 'sz-without-readability':
# fetching articles from sueddeutsche.de/politik
url = 'http://www.sueddeutsche.de/politik'
site = BeautifulSoup(urllib2.urlopen(url).read())
titles = site.findAll("div", { "class" : "teaser" })
urls = [a.findNext('a')['href'] for a in titles]
if source is 'zeit':
# fetching articles from zeit.de/politik
url = 'http://www.zeit.de/politik'
site = BeautifulSoup(urllib2.urlopen(url).read())
urls = [a['href'] for a in site.findAll("a", { "class" : "teaser-small__combined-link" })]
print "Found %d articles on %s"%(len(urls),url)
# predict party from url for this source
print "Predicting %s"%source
for url in urls:
try:
title,text = fetch_url(url)
prediction = clf.predict(text)
prediction['url'] = url
prediction['source'] = source
articles.append((title,prediction))
except:
print('Could not get text from %s'%url)
pass
# do some topic modeling
topics = kpca_cluster(map(lambda x: x[1]['text'][0], articles))
# remove original article text for faster web-frontend
for a in articles:
a[1]['text'] = 'deleted'
# store current news and topics
json.dump(articles,open('news.json','wb'))
json.dump(topics,open('topics.json','wb'))
import labeler
import classifier
l1 = labeler.Labeler()
list1 = l1.extractAndPrint(
"http://money.cnn.com/2018/03/21/technology/mark-zuckerberg-cambridge-analytica-response/index.html"
)
print(list1)
print("\n")
l2 = labeler.Labeler()
list2 = l2.extractAndPrint(
"https://www.nytimes.com/2018/03/27/world/europe/whistle-blower-data-mining-cambridge-analytica.html"
)
print(list2)
print("\n")
c1 = classifier.Classifier()
c1.readAndCompare(list1, list2)
#roc_auc = auc(fpr, tpr)
#mean_tpr /= len(cv)
#mean_tpr[-1] = 1.0
mean_auc = auc(mean_fpr, mean_tpr)
plt.plot(mean_fpr,
mean_tpr,
'k--',
label='Mean ROC (area = %0.2f) for diseases %d' %
(mean_auc, i),
lw=2)
#plt.legend(loc="lower right")
plt.xlim([-0.05, 1.05])
plt.ylim([-0.05, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('ROC and AUC for 3 diseases')
plt.show()
if __name__ == '__main__':
nb_clf = clf.Classifier('NB')
tree_clf = clf.Classifier('Tree')
nb_y_hat = cro_vld(nb_clf, 10)
tree_y_hat = cro_vld(tree_clf, 10)
plot_auc(nb_clf.y, nb_y_hat)
plot_auc(tree_clf.y, tree_y_hat)
print('--------------------------')
X_train_df = X_df.iloc[train_is].copy()
y_train_df = y_df.iloc[train_is].copy()
X_test_df = X_df.iloc[test_is].copy()
y_test_df = y_df.iloc[test_is].copy()
y_train_clf = y_train_df['molecule'].values
y_train_reg = y_train_df['concentration'].values
y_test_clf = y_test_df['molecule'].values
y_test_reg = y_test_df['concentration'].values
fe_clf = feature_extractor_clf.FeatureExtractorClf()
fe_clf.fit(X_train_df, y_train_clf)
X_train_array_clf = fe_clf.transform(X_train_df)
X_test_array_clf = fe_clf.transform(X_test_df)
clf = classifier.Classifier()
clf.fit(X_train_array_clf, y_train_clf)
y_proba_clf = clf.predict_proba(X_test_array_clf)
y_pred_clf = labels[np.argmax(y_proba_clf, axis=1)]
error = 1 - accuracy_score(y_test_clf, y_pred_clf)
print('error = %s' % error)
fe_reg = feature_extractor_reg.FeatureExtractorReg()
for i, label in enumerate(labels):
X_train_df.loc[:, label] = (y_train_df['molecule'] == label)
X_test_df.loc[:, label] = y_proba_clf[:, i]
fe_reg.fit(X_train_df, y_train_reg)
X_train_array_reg = fe_reg.transform(X_train_df)
X_test_array_reg = fe_reg.transform(X_test_df)
reg = regressor.Regressor()
def callback():
# Auth Step 4: Requests refresh and access tokens
access_token = request.args['access_token']
# Auth Step 6: Use the access token to access Spotify API
authorization_header = {"Authorization": "Bearer {}".format(access_token)}
params = {"limit": "50"}
# Get profile data
user_profile_api_endpoint = "{}/me/top/artists".format(SPOTIFY_API_URL)
profile_response = requests.get(user_profile_api_endpoint,
params=params,
headers=authorization_header)
profile_data = json.loads(profile_response.text)
print(profile_data)
# Combine profile and playlist data to display
display_arr = [profile_data]
list_of_genres = [a['genres'] for a in profile_data['items']]
BB = []
CC = []
for ll in list_of_genres:
LL = list(set(ll))
for kk in LL:
BB.append(kk.split())
CC.append(BB)
BB = []
flat_genres = [
item2 for sublist in CC for item in sublist for item2 in item
]
numOfGenres = len(genres)
mean_genre = [0] * numOfGenres
for genre in flat_genres:
genre = clean_genre(genre)
try:
idx = genres.index(genre)
except ValueError:
continue
mean_genre[idx] += 1
normalizer = sum(mean_genre)
if normalizer == 0:
normalizer = 1
for i in range(0, len(mean_genre)):
mean_genre[i] = mean_genre[i] / normalizer
clf = classifier.Classifier(genres, k)
maxScores, pred = clf.classify(mean_genre, X, Y)
result = []
for j in range(0, 20):
for i in range(0, k):
for track in Y[1][pred[1][0][i]]:
if track['id'] == maxScores[j][0]:
#print(track['permalink_url'])
#print(maxScores[j][1])
#print(track['id'])
#print(track['genre'])
result.append(track['stream_url'])
break
return {"data": result}