Tensorflow 如何计算CNN模型训练过程的总运行时间

我需要获得CNN训练过程的总运行时间，但我不知道如何获得。是否有任何包可以获取总运行时间

我这里有一个CNN培训过程的结果示例。正如您所看到的，每个历元都有一个运行时进程，但我需要得到所有训练进程的总数。 有人能帮我吗

Epoch 1/50
250/250 [==============================] - 18s 71ms/step - loss: 4.0427 - accuracy: 0.0111 - val_loss: 3.8992 - val_accuracy: 0.0200
Epoch 2/50
250/250 [==============================] - 17s 70ms/step - loss: 3.9247 - accuracy: 0.0160 - val_loss: 3.6276 - val_accuracy: 0.0800
Epoch 3/50
250/250 [==============================] - 18s 70ms/step - loss: 3.3692 - accuracy: 0.0546 - val_loss: 2.9868 - val_accuracy: 0.1200
Epoch 4/50
250/250 [==============================] - 17s 70ms/step - loss: 2.8787 - accuracy: 0.1194 - val_loss: 2.5775 - val_accuracy: 0.2000
Epoch 5/50
250/250 [==============================] - 18s 71ms/step - loss: 2.4328 - accuracy: 0.2992 - val_loss: 2.0736 - val_accuracy: 0.3600
Epoch 6/50
250/250 [==============================] - 18s 70ms/step - loss: 1.8779 - accuracy: 0.4849 - val_loss: 1.6079 - val_accuracy: 0.4800
Epoch 7/50
250/250 [==============================] - 18s 71ms/step - loss: 1.4844 - accuracy: 0.6018 - val_loss: 1.3112 - val_accuracy: 0.6800
Epoch 8/50
250/250 [==============================] - 17s 69ms/step - loss: 1.2182 - accuracy: 0.6820 - val_loss: 1.1277 - val_accuracy: 0.6200
Epoch 9/50
250/250 [==============================] - 18s 70ms/step - loss: 1.0166 - accuracy: 0.7420 - val_loss: 0.9915 - val_accuracy: 0.7000
Epoch 10/50
250/250 [==============================] - 17s 70ms/step - loss: 0.8404 - accuracy: 0.8148 - val_loss: 0.8455 - val_accuracy: 0.7600
Epoch 11/50
250/250 [==============================] - 18s 71ms/step - loss: 0.6988 - accuracy: 0.8659 - val_loss: 0.7465 - val_accuracy: 0.7800
Epoch 12/50
250/250 [==============================] - 18s 71ms/step - loss: 0.5902 - accuracy: 0.8940 - val_loss: 0.6760 - val_accuracy: 0.7600
Epoch 13/50
250/250 [==============================] - 18s 71ms/step - loss: 0.5075 - accuracy: 0.9091 - val_loss: 0.5958 - val_accuracy: 0.8400
Epoch 14/50
250/250 [==============================] - 18s 71ms/step - loss: 0.4247 - accuracy: 0.9260 - val_loss: 0.5684 - val_accuracy: 0.8000
Epoch 15/50
250/250 [==============================] - 18s 71ms/step - loss: 0.3621 - accuracy: 0.9402 - val_loss: 0.5031 - val_accuracy: 0.8000
Epoch 16/50
250/250 [==============================] - 17s 70ms/step - loss: 0.3124 - accuracy: 0.9460 - val_loss: 0.4263 - val_accuracy: 0.9000
Epoch 17/50
250/250 [==============================] - 17s 69ms/step - loss: 0.2622 - accuracy: 0.9550 - val_loss: 0.3983 - val_accuracy: 0.8800
Epoch 18/50
250/250 [==============================] - 18s 72ms/step - loss: 0.2351 - accuracy: 0.9547 - val_loss: 0.4028 - val_accuracy: 0.8200
Epoch 19/50
250/250 [==============================] - 18s 72ms/step - loss: 0.2002 - accuracy: 0.9666 - val_loss: 0.3812 - val_accuracy: 0.8600
Epoch 20/50
250/250 [==============================] - 18s 71ms/step - loss: 0.1770 - accuracy: 0.9693 - val_loss: 0.3330 - val_accuracy: 0.9000
Epoch 21/50
250/250 [==============================] - 18s 71ms/step - loss: 0.1538 - accuracy: 0.9762 - val_loss: 0.4010 - val_accuracy: 0.8600
Epoch 22/50
250/250 [==============================] - 18s 71ms/step - loss: 0.1446 - accuracy: 0.9736 - val_loss: 0.3387 - val_accuracy: 0.8600
Epoch 23/50
250/250 [==============================] - 18s 70ms/step - loss: 0.1171 - accuracy: 0.9850 - val_loss: 0.2887 - val_accuracy: 0.9400
Epoch 24/50
250/250 [==============================] - 18s 72ms/step - loss: 0.1301 - accuracy: 0.9712 - val_loss: 0.2748 - val_accuracy: 0.9400
Epoch 25/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0966 - accuracy: 0.9846 - val_loss: 0.2883 - val_accuracy: 0.9200
Epoch 26/50
250/250 [==============================] - 18s 72ms/step - loss: 0.0936 - accuracy: 0.9853 - val_loss: 0.2930 - val_accuracy: 0.8800
Epoch 27/50
250/250 [==============================] - 18s 70ms/step - loss: 0.1104 - accuracy: 0.9761 - val_loss: 0.2626 - val_accuracy: 0.9000
Epoch 28/50
250/250 [==============================] - 18s 71ms/step - loss: 0.1067 - accuracy: 0.9762 - val_loss: 0.3085 - val_accuracy: 0.8800
Epoch 29/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0903 - accuracy: 0.9808 - val_loss: 0.3507 - val_accuracy: 0.8600
Epoch 30/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0792 - accuracy: 0.9886 - val_loss: 0.2796 - val_accuracy: 0.9000
Epoch 31/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0554 - accuracy: 0.9940 - val_loss: 0.2505 - val_accuracy: 0.9000
Epoch 32/50
250/250 [==============================] - 18s 72ms/step - loss: 0.0734 - accuracy: 0.9861 - val_loss: 0.2937 - val_accuracy: 0.9000
Epoch 33/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0446 - accuracy: 0.9949 - val_loss: 0.2639 - val_accuracy: 0.9000
Epoch 34/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0685 - accuracy: 0.9866 - val_loss: 0.3331 - val_accuracy: 0.9000
Epoch 35/50
250/250 [==============================] - 18s 71ms/step - loss: 0.1402 - accuracy: 0.9643 - val_loss: 0.2561 - val_accuracy: 0.9400
Epoch 36/50
250/250 [==============================] - 18s 72ms/step - loss: 0.0424 - accuracy: 0.9962 - val_loss: 0.3020 - val_accuracy: 0.8600
Epoch 37/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0356 - accuracy: 0.9969 - val_loss: 0.2670 - val_accuracy: 0.9200
Epoch 38/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0380 - accuracy: 0.9958 - val_loss: 0.3640 - val_accuracy: 0.8400
Epoch 39/50
250/250 [==============================] - 18s 70ms/step - loss: 0.0960 - accuracy: 0.9793 - val_loss: 0.3186 - val_accuracy: 0.8800
Epoch 40/50
250/250 [==============================] - 18s 72ms/step - loss: 0.0394 - accuracy: 0.9945 - val_loss: 0.2686 - val_accuracy: 0.9000
Epoch 41/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0364 - accuracy: 0.9955 - val_loss: 0.3073 - val_accuracy: 0.9000
Epoch 42/50
250/250 [==============================] - 18s 70ms/step - loss: 0.1673 - accuracy: 0.9567 - val_loss: 0.3678 - val_accuracy: 0.9000
Epoch 43/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0366 - accuracy: 0.9992 - val_loss: 0.3368 - val_accuracy: 0.8800
Epoch 44/50
250/250 [==============================] - 18s 72ms/step - loss: 0.0318 - accuracy: 0.9992 - val_loss: 0.3261 - val_accuracy: 0.8800
Epoch 45/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0338 - accuracy: 0.9969 - val_loss: 0.3418 - val_accuracy: 0.9000
Epoch 46/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0265 - accuracy: 0.9987 - val_loss: 0.3504 - val_accuracy: 0.8600
Epoch 47/50
250/250 [==============================] - 18s 71ms/step - loss: 0.0286 - accuracy: 0.9969 - val_loss: 0.3072 - val_accuracy: 0.9200
Epoch 48/50
250/250 [==============================] - 18s 72ms/step - loss: 0.1499 - accuracy: 0.9646 - val_loss: 0.2840 - val_accuracy: 0.9200
Epoch 49/50
250/250 [==============================] - 17s 70ms/step - loss: 0.0348 - accuracy: 0.9992 - val_loss: 0.2803 - val_accuracy: 0.9000
Epoch 50/50
250/250 [==============================] - 17s 70ms/step - loss: 0.0275 - accuracy: 0.9986 - val_loss: 0.3138 - val_accuracy: 0.9200

你可以试试这个

导入时间
开始=时间。时间（）
model.fit（）#培训声明
打印（“总时间：，time.time（）-开始，“秒”）

您需要添加回调方法，示例代码：

class timecallback(tf.keras.callbacks.Callback):
    def __init__(self):
        self.times = []
        # use this value as reference to calculate cumulative time taken
        self.timetaken = time.process_time()
    def on_epoch_end(self,epoch,logs = {}):
        self.times.append((epoch, time.process_time() - self.timetaken))
    def on_train_end(self,logs = {}):
        plt.xlabel('Epoch')
        plt.ylabel('Total time taken until an epoch in seconds')
        plt.plot(*zip(*self.times))
        plt.show()
        from operator import itemgetter
        previous_time = 0
        for item in self.times:
            print("Epoch ", item[0], " run time is: ", item[1]-previous_time)
            previous_time = item[1]
        print("Total trained time is: ", previous_time)

timetaken = timecallback()
callbacks = [
    # This callback saves a SavedModel every epoch
    # We include the current epoch in the folder name.
    keras.callbacks.ModelCheckpoint(
        filepath=os.path.join(os.getcwd(), "ckpt-{epoch}"), save_freq="epoch"
    ),
    timetaken
]

model.fit(train_dataset, epochs=epochs, callbacks=callbacks, validation_data=val_dataset)

---

我以前用过它，但运行时间不正确