Callback.py

num_epoch  = 50
print_freq = 10

class Histories(Callback):
    def on_train_begin(self, logs={}):
        self.epoch = []
        self.history = {}
        self.history["batch"] = []
 
    def on_train_end(self, logs={}):
        return
 
    def on_epoch_begin(self, epoch, logs={}):
        self.epoch.append(epoch)
        self.batch = []
        self.history['batch'] = {}
        self.history['batch']['acc']  = [0] * self.params['steps']
        self.history['batch']['loss'] = [0] * self.params['steps']
 
    def on_epoch_end(self, epoch, logs={}):
        for k, v in logs.items():
            self.history.setdefault(k, []).append(v)
 
    def on_batch_begin(self, batch, logs={}):
        self.batch.append(batch)
 
    def on_batch_end(self, batch, logs={}):
        self.history['batch']['acc'][batch]  = logs['acc']
        self.history['batch']['loss'][batch] = logs['loss']
        if (((batch+1) % print_freq) == 0) or (batch == self.params['steps']-1):
            num_batch  = self.params['steps']
            batch_loss = self.history['batch']['loss'][batch]
            batch_acc  = self.history['batch']['acc'][batch]
            train_loss = mean(self.history['batch']['loss'][:batch+1])
            train_acc  = mean(self.history['batch']['acc'][:batch+1])
            print('Ep %2d | Batch (%3d / %3d) | Trn [BLoss %.5f | BAcc %.5f | Loss %.5f | Acc %.5f]' 
              % (epoch+1, batch+1, num_batch, batch_loss, batch_acc, train_loss, train_acc), end='\r')

train_gen = genBatch(X_train, Y_train, batch_size,     num_batches)
val_gen   = genBatch(X_test,  Y_test,  batch_size_val, num_batches_val)
model.fit_generator(generator=train_gen, 
                    steps_per_epoch=num_batches, 
                    initial_epoch=epoch, 
                    epochs=epoch+1, 
                    validation_data=val_gen,
                    validation_steps=num_batches_val,
                    callbacks=[learning_rate_scheduler, h], 
                    verbose=0)