def main():
features = lsi(tfidf('small_train.txt'), 100)
classes = get_classes('small_train.txt', 0.7, 5)
prob = svm_problem(classes, features)
param = svm_parameter('-t 0 -c 4 -b 1')
m = svm_train(prob, param)
test_features = lsi(tfidf('small_test.txt'), 100)
test_classes = get_classes('small_test.txt', 0.7, 5)
p_label, p_acc, p_val = svm_predict(test_classes, test_features, m)
def get_acc(trainfilename, testfilename, d, a, r):
features = lsi(tfidf('small_train.txt'), d)
classes = get_classes('small_train.txt', a, r)
prob = svm_problem(classes, features)
param = svm_parameter('-t 0 -c 4 -b 1')
m = svm_train(prob, param)
test_features = lsi(tfidf('small_test.txt'),d)
test_classes = get_classes('small_test.txt', a, r)
p_label, p_acc, p_val = svm_predict(test_classes, test_features, m)
return p_acc
def get_attendance(username, subject):
user = db.users.find_one({"username" : username})
attendance = user['attendance']
sub = get_classes(username)["subjects"]
req_sub = {}
for i in sub:
if i['name'] == subject:
req_sub = i
tot = len (req_sub["occurences"])
present = 0
absent = 0
for k in req_sub["occurences"]:
for i in attendance:
if i["id"] == k["id"]:
if i["presence"] == "p":
present += 1
elif i["absent"] == "a":
absent +=1
attendance_sub = {
"subject": subject ,
"attendance": {
"total" : tot,
"present": present,
"absent": absent,
"pending": tot - absent - present
}
}
print dumps(attendance_sub)
return dumps( attendance_sub )
def methods_trace(ctxt, expr=None):
'reports calls to several dalvik method'
cpt = 0
class_list = get_classes(ctxt, expr)
for cla in class_list:
method_list = get_methods(ctxt, cla)
for method in method_list:
cpath, mname, mjni = andbug.options.parse_mquery(".".join(method.split('.')[0:-1]), method.split('.')[-1])
cmd_hook_methods(ctxt, cpath, mname)
cpt += 1
#print(mname)
andbug.screed.item("%d functions hooked" % cpt)
def detect(self, img):
img_h, img_w = img.shape[:2]
inp, ox, oy, new_w, new_h = self.preprocess(img)
inp = np.expand_dims(inp,0)
outs = self.sess.run(self.net.prediction, feed_dict={self.inputs: inp})
outs = outs[0]
#print('out,',outs.shape)
boxes = decode_boxes(outs[:, :4],self.priors)
preds = np.argmax(outs[:, 4:], axis=1)
confidences = np.max(outs[:, 4:], axis=1)
#print('preds',preds.shape,confidences[:5])
#print(preds[:5])
# skip background class
mask = np.where(preds > 0)
#print('cls_mask',mask[0].shape)
boxes = boxes[mask]
preds = preds[mask]
confidences = confidences[mask]
mask = np.where(confidences >= cfgs.AnchorThreshold)
#print('confidence mask and threshold',mask[0].shape,cfgs.AnchorThreshold)
boxes = boxes[mask]
preds = preds[mask]
confidences = confidences[mask]
#print('final score:',confidences)
results = []
for box, clsid, conf in zip(boxes, preds, confidences):
xmin, ymin, xmax, ymax = box
left = int((xmin - ox) / new_w * img_w)
top = int((ymin - oy) / new_h * img_h)
right = int((xmax - ox) / new_w * img_w)
bottom = int((ymax - oy) / new_h * img_h)
conf = float(conf)
name, color = get_classes(clsid - 1)
results.append({
'left': left,
'top': top,
'right': right,
'bottom': bottom,
'name': name,
'color': color,
'confidence': conf,
})
#results = nms(results,1-self.threshold,'Min')
results = nms(results,self.threshold)
results = nms(results,1- self.threshold,'Min')
#results = soft_nms(results, self.threshold)
#print("after nms result num: ",len(results))
return results
args = parser.parse_args()
parser.print_help()
print('input args: ', args)
if __name__ == '__main__':
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=1.0)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
# sess = tf.compat.v1.Session()
# K.set_session(sess)
tf.compat.v1.keras.backend.set_session(sess)
data_dirs = args.data_dirs
output_representation = args.output_representation
sample_rate = args.sample_rate
print("sample rate", sample_rate)
batch_size = args.batch_size
classes = get_classes(wanted_only=True)
model_settings = prepare_model_settings(
label_count=len(prepare_words_list(classes)),
sample_rate=sample_rate,
clip_duration_ms=800,
window_size_ms=30.0,
window_stride_ms=10.0,
dct_coefficient_count=80,
num_log_mel_features=60,
output_representation=output_representation)
print(model_settings)
ap = AudioProcessor(data_dirs=data_dirs,
wanted_words=classes,
silence_percentage=13.0,
default=100,
help='num of epoch')
args = parser.parse_args()
parser.print_help()
print('input args: ', args)
if __name__ == '__main__':
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=1.0)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
K.set_session(sess)
data_dirs = args.data_dirs
output_representation = args.output_representation
sample_rate = args.sample_rate
batch_size = args.batch_size
epoch = args.epoch
classes = get_classes(wanted_only=True)
model_settings = prepare_model_settings(
label_count=len(prepare_words_list(classes)),
sample_rate=sample_rate,
clip_duration_ms=1000,
window_size_ms=30.0,
window_stride_ms=10.0,
dct_coefficient_count=80,
num_log_mel_features=60,
output_representation=output_representation)
print(model_settings)
ap = AudioProcessor(
data_dirs=data_dirs,
wanted_words=classes,
required=True,
help='<Required> The list of data directories. e.g., data/train')
args = parser.parse_args()
parser.print_help()
print('input args: ', args)
if __name__ == '__main__':
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=1.0)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
K.set_session(sess)
data_dirs = args.data_dirs
output_representation = args.output_representation
sample_rate = args.sample_rate
batch_size = args.batch_size
classes = get_classes(wanted_only=True)
model_settings = prepare_model_settings(
label_count=len(prepare_words_list(classes)),
sample_rate=sample_rate,
clip_duration_ms=1000,
window_size_ms=30.0,
window_stride_ms=10.0,
dct_coefficient_count=80,
num_log_mel_features=60,
output_representation=output_representation)
print(model_settings)
ap = AudioProcessor(
data_dirs=data_dirs,
wanted_words=classes,
if __name__ == '__main__':
test_fns = sorted(glob('data/test/audio/*.wav'))
sess = K.get_session()
K.set_learning_phase(0)
sample_rate = 16000
use_tta = True
use_speed_tta = False
if use_speed_tta:
tta_fns = sorted(glob('data/tta_test/audio/*.wav'))
assert len(test_fns) == len(tta_fns)
wanted_only = False
extend_reversed = False
output_representation = 'raw'
batch_size = 384
wanted_words = prepare_words_list(get_classes(wanted_only=True))
classes = get_classes(wanted_only=wanted_only,
extend_reversed=extend_reversed)
int2label = get_int2label(wanted_only=wanted_only,
extend_reversed=extend_reversed)
model_settings = prepare_model_settings(
label_count=len(prepare_words_list(classes)),
sample_rate=sample_rate,
clip_duration_ms=1000,
window_size_ms=25.0,
window_stride_ms=15.0,
dct_coefficient_count=80,
num_log_mel_features=60,
output_representation=output_representation)
ap = AudioProcessor(data_dirs=['data/train/audio'],
wanted_words=classes,
# np.log(32) ~ 3.5
# np.log(48) ~ 3.9
# 64727 training files
if __name__ == '__main__':
# restrict gpu usage: https://stackoverflow.com/questions/34199233/how-to-prevent-tensorflow-from-allocating-the-totality-of-a-gpu-memory # noqa
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.95)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
K.set_session(sess)
data_dirs = ['data/train/audio']
add_pseudo = True
if add_pseudo:
data_dirs.append('data/heng_pseudo')
output_representation = 'raw'
sample_rate = 16000
batch_size = 384
classes = get_classes(wanted_only=True, extend_reversed=False)
model_settings = prepare_model_settings(
label_count=len(prepare_words_list(classes)),
sample_rate=sample_rate,
clip_duration_ms=1000,
window_size_ms=30.0,
window_stride_ms=10.0,
dct_coefficient_count=80,
num_log_mel_features=60,
output_representation=output_representation)
ap = AudioProcessor(data_dirs=data_dirs,
wanted_words=classes,
silence_percentage=13.0,
unknown_percentage=60.0,
validation_percentage=10.0,
testing_percentage=0.0,
from tensorflow.python.ops import io_ops
from keras import backend as K
K.set_session(sess)
K.set_learning_phase(0)
from keras.models import load_model
from keras.applications.mobilenet import DepthwiseConv2D
from keras.activations import softmax
from model import relu6, overlapping_time_slice_stack
from classes import get_classes
from utils import smooth_categorical_crossentropy
FINAL_TENSOR_NAME = 'labels_softmax'
FROZEN_PATH = 'tf_files/frozen.pb'
OPTIMIZED_PATH = 'tf_files/optimized.pb'
wanted_classes = get_classes(wanted_only=True)
all_classes = get_classes(wanted_only=False)
model = load_model('checkpoints_186/ep-053-vl-0.2915.hdf5',
custom_objects={'relu6': relu6,
'DepthwiseConv2D': DepthwiseConv2D,
'overlapping_time_slice_stack':
overlapping_time_slice_stack,
'softmax': softmax,
'<lambda>':
smooth_categorical_crossentropy})
# rename placeholders for special prize:
# https://www.kaggle.com/c/tensorflow-speech-recognition-challenge#Prizes
# decoded_sample_data:0, taking a [16000, 1] float tensor as input,
# representing the audio PCM-encoded data.