BaseCollectiveExecutor::StartAbort Out of range: End of sequence [[{{node IteratorGetNext}}问题

使用tensorflow2时训练epoch出现

Epoch 35/100
54/55 [============================>.] - ETA: 0s - loss: 0.5587 - accuracy: 0.70622020-01-07 15:36:38.360506: W tensorflow/core/common_runtime/base_collective_executor.cc:216] BaseCollectiveExecutor::StartAbort Out of range: End of sequence
	 [[{{node IteratorGetNext}}]]
	 [[Shape/_6]]

问题，出现问题的代码为：

...
...#导入数据
db = tf.data.Dataset.from_tensor_slices((x_train, y_train))
db = db.shuffle(6000).batch(batchsz)
ds_val = tf.data.Dataset.from_tensor_slices((x_val, y_val))
ds_val = ds_val.batch(batchsz)
db_test = tf.data.Dataset.from_tensor_slices((x_test, y_test))
db_test = db_test.batch(batchsz)
network = Sequential([layers.Conv1D(32, kernel_size=6,strides=3,padding='same',activation=tf.nn.relu),
                       layers.Dropout(0.5),
                       layers.Conv1D(64, kernel_size=6, strides=3,padding='same',activation=tf.nn.relu),
                       layers.Dropout(0.5),
                       layers.Conv1D(128, kernel_size=6, strides=3,padding='same',activation=tf.nn.relu),
                       layers.Dropout(0.5),
                       layers.Flatten(),
                       layers.Dense(64, activation=tf.nn.relu),
                       layers.Dropout(0.4),
                       layers.Dense(1,activation='sigmoid')])
network.build(input_shape=(None, 2000,4))
network.summary()
network.compile(optimizer=optimizers.Adam(lr=0.001),
                loss=tf.keras.losses.BinaryCrossentropy(),
                metrics=['accuracy'] )

network.fit(db, epochs=1000, validation_data=ds_val,
            validation_steps=2）

这是因为epoch在循环时，发现最后一个step数据不够一个batchsz，导致out-of-range错误，因此，需要在network.fit的时候加上steps_per_epoches的数据，例如如果你有1300个数据，batchsize=200，那么最后steps数目就为1300/200取整出6，然后在datasets中加repeat。

修改后的代码为：

db = tf.data.Dataset.from_tensor_slices((x_train, y_train))
db = db.shuffle(6000).batch(batchsz).repeat()                           # 加 repeat
ds_val = tf.data.Dataset.from_tensor_slices((x_val, y_val))
ds_val = ds_val.batch(batchsz)
db_test = tf.data.Dataset.from_tensor_slices((x_test, y_test))
db_test = db_test.batch(batchsz)
network = Sequential([layers.Conv1D(32, kernel_size=6, strides=3,padding='same',activation=tf.nn.relu),
                      layers.Dropout(0.5),
                      layers.Conv1D(64, kernel_size=6, strides=3,padding='same',activation=tf.nn.relu),
                      layers.Dropout(0.5),
                      layers.Conv1D(128, kernel_size=6, strides=3,padding='same',activation=tf.nn.relu),
                      layers.Dropout(0.5),
                      layers.Flatten(),
                      layers.Dense(64, activation=tf.nn.relu),
                      layers.Dropout(0.4),
                      layers.Dense(1,activation='sigmoid')])
network.build(input_shape=(None, 2000,4))
network.summary()
network.compile(optimizer=optimizers.Adam(lr=0.001),
                loss=tf.keras.losses.BinaryCrossentropy(),
                metrics=['accuracy']
                )

network.fit(db, epochs=1000, validation_data=ds_val,
            validation_steps=2,steps_per_epoch=x_train.shape[0]//batchsz）# 取整数steps的数目