mohitprajapati

callbacks = [
            ReduceLROnPlateau(verbose=1),
            EarlyStopping(patience=3, verbose=1),
            ModelCheckpoint('/content/drive/MyDrive/Casestudy2/train/new_checkpoints/yolov3_train_{epoch}.tf',
                            verbose=1, save_weights_only=True),
            TensorBoard(log_dir='/content/drive/MyDrive/Casestudy2/train/new_logs')
        ]

mohitprajapati

model = Yolo_V3(size=416,channels=3,training=True,classes=4)
pretrained_model = Yolo_V3(size=416,channels=3,classes=80,training=True)
pretrained_model.load_weights(WEIGHTS_PATH)
model.get_layer('yolo_darknet').set_weights(pretrained_model.get_layer('yolo_darknet').get_weights())
Freeze_All(model.get_layer('yolo_darknet'))

mohitprajapati

def Yolo_Loss(anchors, classes=80, ignore_thresh=0.5):
    ''' Loss function for YOLOV3 '''
    def yolo_loss(y_true, y_pred):
        # 1. transform all pred outputs
        # y_pred: (batch_size, grid, grid, anchors, (x, y, w, h, obj, ...cls))
        pred_box, pred_obj, pred_class, pred_xywh = yolo_boxes(y_pred, anchors, classes)
        pred_xy = pred_xywh[..., 0:2]
        pred_wh = pred_xywh[..., 2:4]

        # 2. transform all true outputs

mohitprajapati

def yolo_boxes(pred, anchors, classes):
    ''' Function to get absolute coordinates of bounding boxes and objectness and class probs'''
    # pred: (batch_size, grid, grid, anchors, (x, y, w, h, obj, ...classes))
    Grd_Siz = tf.shape(pred)[1:3]
    Box_xy, Box_wh, Objectness, Class_Prob = tf.split(
        pred, (2, 2, 1, classes), axis=-1)
    #print(class_probs)
    Box_xy = tf.sigmoid(Box_xy)
    Objectness = tf.sigmoid(Objectness)
    Class_Prob = tf.sigmoid(Class_Prob)

mohitprajapati

def Create_TFrecord(path,xml,class_map,data,augmentation):
  ''' creating tf records '''
  images = os.path.join(path,'images',xml['filename'])
  img_raw = open(images,'rb').read()
  img = cv2.imread(images)
  width = int(xml['size']['width'])
  height = int(xml['size']['height'])
  if data == 'train' and augmentation:
    # for train dataset and if augmentation
    augseq = iaa.Sequential([

mohitprajapati

# stackoverflow for parsing the XML files
files = [file for file in os.listdir('/content/train/India/annotations/xmls')] 
cnt_dmg= [] # list to store the counts for the demage
for file in files:
  xmlfile = open('/content/train/India/annotations/xmls/' + file)
  tree = ET.parse(xmlfile)
  root = tree.getroot()
  for obj in root.iter('object'):
    dmg_typ = obj.find('name').text
    cnt_dmg.append(dmg_typ)

mohitprajapati

for i in range(len(label_count)-1):
    f1_state_terminal[i]/=273
print("The Combined Average F1 scores of train Terminals ",f1_state_terminal) # order available, passive, charging, offline, down

for i in range(len(label_count)-1):
    f1_state_terminal_test[i]/=273
print("The Combined Average F1 scores of test Terminals ",f1_state_terminal_test) # order available, passive, charging, offline,down

f1_final=0

mohitprajapati

# one hot encoding
week_number_ohe= pd.get_dummies(X['Week_number'])
week_number_test_ohe=pd.get_dummies(X_test['Week_number'])
print(week_number_ohe.shape)
print(week_number_test_ohe.shape)

mohitprajapati

# Is weekend
def IsWeekEnd(number):
    if number in [5,6]: # if staurday or sunday return 1
        return 1
    else: return 0    # weekdays=6
weekend=X['Week_number'].apply(IsWeekEnd)
weekend_test=X_test['Week_number'].apply(IsWeekEnd)
print(weekend.shape)
print(weekend_test.shape)

mohitprajapati

### DEFINE MODEL STRUCTURES ###

def T2V_NN(t2v_dim, dim):

    inp = layers.Input(shape=(dim,1))
    x = T2V(t2v_dim)(inp)
    x = layers.LSTM(5, activation='tanh', return_sequences=True)(x)
    
    x = layers.LSTM(5, activation='tanh')(x)