InvalidArgumentError：reshape 的输入是一个具有 405000 个值的张量，但请求的形状有 1920000 [Op:Reshape]

Question

我在使用tensorflow2.0的时候遇到了一些问题，这是我的代码，控制台报错：

x=tf.reshape(x,shape=[BATCH_SIZE, self.lstm_step_num，self.input_length*self.vector_length])
File"/home/cyye/anaconda3/lib/python3.7/site-packages/te nsorflow/python/ops/array_ops.py", line 193, in reshape
result = gen_array_ops.reshape(tensor, shape, name)
File"/home/cyye/anaconda3/lib/python3.7/site-packages/te nsorflow/python/ops/gen_array_ops.py", line 8080, in reshape
tensor, shape, name=name, ctx=_ctx)
File"/home/cyye/anaconda3/lib/python3.7/site-packages/te nsorflow/python/ops/gen array ops.py", line 8107, in reshape eager fallback
ctx=ctx，name=name)
File"/home/cyye/anaconda3/lib/python3.7/site-packages/te nsorflow/python/eager/execute.py", line 60, in quick_execute
inputs, attrs, num_outputs)
tensorflow.python.framework.errors_impl.InvalidArgumenterro r: Input to reshape is a tensor with 405000 values, but the reques ted shape has 1920000[0p:Reshape]

我的数据集的shape=[4999990,10,15]，类型是list。在 reshape 层，input(128,10,15,100)(1920000) 的数据 rashape 为 data(BATCH_SIZE=128, self.lstm_step_num=10, self.input_length=15，self. vector_length=100)(1920000)。 我不知道错误在哪里，为什么不能转换形状。我非常感谢有关如何清理我的代码的任何提示！提前致谢！

# author:ZhuYuYing
# data:2021/12/9
# projectName:repetition-QSPC

import tensorflow as tf
import numpy as np
import tensorflow.keras as keras
from tensorflow.keras import layers
import time
tf.keras.backend.set_floatx('float32')

dataSet = np.load(r"../data/yuan/train_four/dataset_1.npy").astype('float32')
labels = np.load(r"../data/yuan/train_four/dataset_1_label_rt.npy").astype('float32')

data_rate = 0.5
split_data = int(len(dataSet)*data_rate)
train_data, test_data = dataSet[:split_data], dataSet[split_data:]
train_labels, test_labels = labels[:split_data], labels[split_data:]


BATCH_SIZE = 128


train_dataset = tf.data.Dataset.from_tensor_slices((train_data, train_labels))
train_dataset = train_dataset.shuffle(buffer_size=1024).batch(BATCH_SIZE)

test_dataset = tf.data.Dataset.from_tensor_slices((test_data, test_labels))
test_dataset = test_dataset.batch(BATCH_SIZE)



class TenModel(keras.Model):
    def __init__(self,UUID,vector_length,input_length,lstm_step_num,lstm_unit_num,
                 rate,perception_unit_num,ts_unit_num):
        super(TenModel, self).__init__()
        self.uuid = UUID
        self.vector_length = vector_length
        self.input_length = input_length
        self.lstm_step_num = lstm_step_num
        self.lstm_unit_num = lstm_unit_num
        self.rate = rate
        self.perception_unit_num = perception_unit_num
        self.ts_unit_num = ts_unit_num
    def call(self, inputs, training=None, mask=None):
        x = layers.Embedding(self.uuid + 1, self.vector_length, embeddings_initializer='normal', input_length=self.input_length)(inputs)  # 10*15*100
        #x = layers.Reshape((BATCH_SIZE, self.lstm_step_num, self.input_length*self.vector_length))(x)
        x = tf.reshape(x, shape=[BATCH_SIZE, self.lstm_step_num, self.input_length*self.vector_length])
        x = layers.Dense(512, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(x)
        x = tf.nn.dropout(x,rate=self.rate)
        x = layers.Dense(self.vector_length, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(x)
        x = layers.Dropout(self.rate)(x)

        lstm_x, _, _ = layers.LSTM(self.lstm_unit_num, return_sequences=True, return_state=True, activation=tf.nn.swish)(x)
        x = lstm_x[:, -1:, :]
        #x = layers.Reshape((BATCH_SIZE, self.lstm_unit_num))(x)
        x = tf.reshape(x, shape=[BATCH_SIZE, self.lstm_unit_num])
        x = layers.Dense(self.perception_unit_num, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(x)
        x = layers.Dropout(self.rate)(x)
        x = layers.Dense(self.perception_unit_num, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(x)
        x = layers.Dropout(self.rate)(x)

        x = layers.Dense(self.ts_unit_num, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(x)
        attention_x = layers.Dense(self.ts_unit_num, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(tf.concat([tf.reshape(lstm_x, [BATCH_SIZE, -1]), x], axis=1))
        attention_x = layers.Dense(self.ts_unit_num, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(attention_x)
        x = tf.multiply(x, tf.nn.softmax(attention_x))
        #x = tf.reduce_sum(x,axis=1)
        x = layers.Dense(self.ts_unit_num/2, activation=tf.nn.swish, bias_initializer=tf.random_normal_initializer())(x)
       

        return x






def loss_fn(y_true, y_pred):
    return tf.keras.losses.MSE(y_true, y_pred)

optimizer = tf.keras.optimizers.Adam()
train_loss = tf.keras.metrics.Mean(name='train_loss')
train_acc = tf.keras.metrics.SparseCategoricalAccuracy(name="train_acc")
test_loss = tf.keras.metrics.Mean(name='test_loss')
test_acc = tf.keras.metrics.SparseCategoricalAccuracy(name='test_acc')


'''
self,UUID,vector_length,input_length,lstm_step_num,lstm_unit_num,
                 rate,perception_unit_num,ts_unit_num
                 
11212,100,0.001,256,256,10,128,256
'''
Model = TenModel(11212,100,15,10,256,0.1,256,256)
tf.config.run_functions_eagerly(True)


@tf.function
def train_step(train_data, train_labels):
    with tf.GradientTape() as tape:
        y_pre = Model(train_data)
        print(y_pre)
        loss = loss_fn(train_labels, y_pre)
        total_loss = loss + sum(Model.losses)

    gradients = tape.gradient(total_loss, Model.trainable_variables)
    optimizer.apply_gradients(zip(gradients, Model.trainable_variables))
    train_loss(loss)
    train_acc(train_labels, y_pre)

@tf.function
def test_step(test_data, test_labels):
    y_pre = Model(test_data)
    loss = loss_fn(test_labels, y_pre)
    test_loss(loss)
    test_acc(test_labels, y_pre)

EPOCHS = 10
for epoch in range(EPOCHS):
    start = time.time()

    for train_data, train_labels in train_dataset:
        train_step(train_data, train_labels)

    for test_data, test_labels in test_dataset:
        test_step(test_data, test_labels)
    print ('Time for epoch {} is {} sec'.format(epoch + 1, time.time()-start))
    print('Epoch:{},Loss:{},Acc:{},Test Loss:{},Test Acc:{}'.format(epoch + 1, train_loss.result(),
                                                                   train_acc.result() * 100, test_loss.result(),
                                                                   test_acc.result() * 100))

print(Model.summary())
tf.keras.utils.plot_model(Model, "my_model_info.png", show_shapes=True)
Model.save("my_model")

Answer 1

您需要在 tensorflow 中删除少于一批数据中的数据。例如：

train_dataset = train_dataset.shuffle(buffer_size=1024).batch(BATCH_SIZE, **drop_remainder=True**)