sadimanna

access_token = '***ENTER YOUR ACCESS TOKEN HERE***'

class botCallback(tf.keras.callbacks.Callback):
    def __init__(self,access_token):
        self.access_token = access_token
        self.ping_url = 'https://api.telegram.org/bot'+str(self.access_token)+'/getUpdates'
        self.response = requests.get(self.ping_url).json()
        self.chat_id = self.response['result'][0]['message']['chat']['id']

    def send_message(self,message):

sadimanna

checkpoint = tf.keras.callbacks.ModelCheckpoint(save_dir+model_name, monitor='val_accuracy', verbose=1, 
                             save_best_only=True, mode='max')

reduce_lr = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_accuracy', factor=0.5, patience=4, 
                                   verbose=1, mode='max', min_lr=0.00001)
                              
early_stop = tf.keras.callbacks.EarlyStopping(monitor="val_loss", mode="min", patience=5)

callbacks_list = [checkpoint, reduce_lr, early_stop, bot_callback, plotter]

sadimanna

history = model.fit_generator(train_data_gen, 
                              steps_per_epoch = TRAIN_STEPS_PER_EPOCH, 
                              epochs = 20, 
                              callbacks = callbacks_list, 
                              validation_data = validation_data_gen, 
                              validation_steps=VAL_STEPS_PER_EPOCH)

sadimanna

class Plotter(botCallback):
    def __init__(self,access_token):
        super().__init__(access_token)

    def on_train_begin(self,logs=None):
        self.batch = 0
        self.epoch = []
        self.train_loss = []
        self.val_loss = []
        self.train_acc = []

sadimanna

def TP(y, pred, th=0.5):
    pred_t = (pred > th)
    return np.sum((pred_t == True) & (y == 1))

sadimanna

def TN(y, pred, th=0.5):
    pred_t = (pred > th)
    return np.sum((pred_t == False) & (y == 0))

sadimanna

def FP(y, pred, th=0.5):
    pred_t = (pred > th)
    return np.sum((pred_t == True) & (y == 0))

sadimanna

def FN(y, pred, th=0.5):
    pred_t = (pred > th)
    return np.sum((pred_t == False) & (y == 1))

sadimanna

def precision(y, pred, th=0.5):
    tp = TP(y,pred,th)
    fp = FP(y,pred,th)
    return tp/(tp+fp)

sadimanna

def recall(y, pred, th=0.5):
    tp = TP(y,pred,th)
    fn = FN(y,pred,th)
    
    return tp/(tp+fn)