当batchsize> 8时，InvalidArgumentError如何在generator中正确使用batchsize？

如何解决当batchsize> 8时，InvalidArgumentError如何在generator中正确使用batchsize？

当我以批处理大小大于1运行这些生成器（对于LSTM）时，出现以下错误：

InvalidArgumentError:  Incompatible shapes: [8,30,10] vs. [8,10]
     [[node sub (defined at ....py:72) ]] [Op:__inference_train_function_6469] 

Function call stack:
train_function


# EDIT:
# solverd
# But now I get:

InvalidArgumentError:  Incompatible shapes: [8] vs. [0]
     [[node gradient_tape/mean_squared_error/weighted_loss/BroadcastGradientArgs (defined at ....py:76) ]] [Op:__inference_train_function_400935]

Function call stack:
train_function

#for this setup:
#Input shape is:  (3,33)
model.add(LSTM( 32,input_shape= (3,33),return_sequences= True ))
model.add(Dropout( 0.3 ));
model.add(BatchNormalization())
model.add(LSTM( 16 )) 
model.add(Dropout( 0.2 ));
model.add(BatchNormalization())
model.add(Dense( 6,activation= tanh )) 
 

Model: "sequential_23"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm_46 (LSTM)               (None,3,32)             8448      
_________________________________________________________________
dropout_34 (Dropout)         (None,32)             0         
_________________________________________________________________
batch_normalization_43 (Batc (None,32)             128       
_________________________________________________________________
lstm_47 (LSTM)               (None,16)                3136      
_________________________________________________________________
dropout_35 (Dropout)         (None,16)                0         
_________________________________________________________________
batch_normalization_44 (Batc (None,16)                64        
_________________________________________________________________
dense_23 (Dense)             (None,6)                 102       
=================================================================
Total params: 11,878
Trainable params: 11,782
Non-trainable params: 96

“ MyGenerator_with_SampleWeights”使用（一维数组）作为每个样本的权重，而另一个则不使用。

如果您发现一些有助于提高这两个生成器效率的方法，我也感到非常高兴。

class MyGenerator_with_SampleWeights(tf.keras.utils.Sequence) :
    def __init__(self,list_x,labels,batch_size,sample_weights=None) : 
        self.labels         = labels
        self.batch_size     = batch_size
        self.list_x         = list_x
        self.sample_weights = sample_weights
        
    def __len__(self) :
        return (np.ceil(len(self.list_x) / float(self.batch_size))).astype(np.int)
    def __getitem__(self,idx) :
        batch_x      = self.list_x[idx * self.batch_size : (idx+1) * self.batch_size]
        batch_y      = self.labels[idx * self.batch_size : (idx+1) * self.batch_size]
        batch_weight = self.sample_weights[idx * self.batch_size : (idx+1) * self.batch_size]
        return np.array(batch_x),np.array(batch_y),np.array(batch_weight)

class MyGenerator(tf.keras.utils.Sequence) :
    def __init__(self,batch_size) : 
        self.labels         = labels
        self.batch_size     = batch_size
        self.list_x         = list_x 
        
    def __len__(self) :
        return (np.ceil(len(self.list_x) / float(self.batch_size))).astype(np.int)
    def __getitem__(self,idx) :
        batch_x      = self.list_x[idx * self.batch_size : (idx+1) * self.batch_size]
        batch_y      = self.labels[idx * self.batch_size : (idx+1) * self.batch_size] 
        return np.array(batch_x),np.array(batch_y)

这是一个最小的可重现样品： （还要从上方复制生成器）

import tensorflow as tf,numpy as np; 
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense,Dropout,LSTM,BatchNormalization
from tensorflow.keras.callbacks import TensorBoard,ModelCheckpoint 
path= r"C:\Users\User\documents\somefolder"  # <--- your path 

model = Sequential()
model.add(LSTM(500,input_shape=X_shapes,return_sequences=True))
model.add(Dropout(0.1)); model.add(BatchNormalization())   
model.add(LSTM(100))
model.add(Dropout(0.1)); model.add(BatchNormalization()) 
model.add(Dense(Y_cols,activation='tanh')); model.add(Dropout(0.2))

optimizer = tf.keras.optimizers.Adam(lr=0.005,decay=1e-7)
model.compile(loss='mean_squared_error',optimizer=optimizer,metrics=['mae'])

BATCH_SIZE  = 8
X_shapes = (30,5)
EPOCHS      = 3
Y_cols,len_X,len_Y = 10,1000,200
train_X,train_Y = np.asarray([np.random.random(X_shapes) for x in range(len_X)]),np.random.random((len_X,Y_cols))
validation_X,validation_Y = np.asarray([np.random.random(X_shapes) for x in range(len_Y)]),np.random.random((len_Y,Y_cols))
weight_arr = np.random.random((len_X,1))

tb = TensorBoard(path)
cp = ModelCheckpoint(path,monitor='val_loss',verbose=1,save_best_only=True,mode='min') 
    
if weight_arr is None:
    train_gen = MyGenerator(train_X,train_Y,BATCH_SIZE)
else:
    train_gen   = MyGenerator_with_SampleWeights(train_X,BATCH_SIZE,sample_weights=weight_arr) 
validation_gen  = MyGenerator(validation_X,validation_Y,BATCH_SIZE)
 
model.fit_generator(train_gen,steps_per_epoch=None,epochs=EPOCHS,validation_data = validation_gen,callbacks=[tb,cp])