Question

我正在尝试生成序列以馈入喀拉拉邦的lstm。

下面显示了示例数据和数据生成器的代码（请注意，该函数中的打印语句是为了帮助从jupyter Notebook进行调试。

下面的代码片段可以复制/粘贴到jupyter中，以重现该错误。

下面是生成测试数据的代码

#Generating test data to check the data generator.
import pandas as pd
import numpy as np
a=np.arange(1,50)
b=np.arange(1,50)
c=np.arange(1,50)
y=np.random.randint(2, size=49)
data={'a':a,'b':b,'c':c,'y':y}
data=pd.DataFrame.from_dict(data)
colnames=list(data)[0:data.shape[1]-1]
col_y='y'

生成器本身的代码如下。我添加了打印语句以帮助调试。

import numpy as np
import keras
class DataGenerator(keras.utils.Sequence):
    def __init__(self, data_x,data_y, batch_size, lead, n_timesteps, n_features,  #dimensions are (samples,timesteps,features)
                 n_classes,shuffle=False):
        'Initialization'
        self.dim = n_features
        self.batch_size = batch_size
        self.lead=lead
        self.n_classes=n_classes
        self.labels = data_y
        self.predictor= data_x
        self.start_idx=0
        self.end_idx=len(data_x)
        self.n_timesteps = n_timesteps
        self.n_features = n_features
        self.shuffle = shuffle
        self.n_batches=0
        self.batch_num=1
        self.on_epoch_end()

    def __len__(self):
        'Denotes the number of batches per epoch'
        return int(np.ceil(len(self.labels) / self.batch_size))

    def __getitem__(self, index):
        'Generate one batch of data'
        idx = self.indexes[self.batch_num]
        print("batch is",self.batch_num)
        print("idx is",idx)
        indexes=np.arange(idx-self.n_timesteps,idx+self.batch_size, step=1)
        print("indexes are", indexes)
        x_temp=self.predictor[indexes,]
        y_temp=self.labels[indexes]
        # Generate data
        X, y = self.__data_generation(x_temp,y_temp)    
        self.batch_num+=1
        print("X is",X)
        print("y is", y)
        return X, y

    def on_epoch_end(self):
        'Updates indexes after each epoch'
        self.batch_num=1
        self.indexes = np.arange(self.start_idx,self.end_idx-self.batch_size,step=self.batch_size)
        if self.shuffle == True:
            np.random.shuffle(self.indexes)

    def __data_generation(self, x_temp,y_temp):
        'Generates data containing batch_size samples' # X : (n_samples, n_timesteps,n_features)
        # Initialization
        X = np.empty([self.batch_size, self.n_timesteps, self.n_features])
        y = np.empty((self.batch_size), dtype=int)
        # Generate data
        for rows in range(self.batch_size):
            # Store sample
            X[rows]=np.expand_dims(x_temp[rows:rows+self.n_timesteps,:][::-1], axis=0)
            # Store class
            y[rows] = y_temp[rows+self.n_timesteps]
        return X, keras.utils.to_categorical(y, num_classes=self.n_classes)

最后，我使用下面的代码来设置测试/训练数据。

import numpy as np

from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation
from keras.layers.recurrent import LSTM
from keras.regularizers import l1, l2
#from my_classes import DataGenerator

n_timesteps=10
n_features=3

# Parameters
params_train = {
          'batch_size': 10,
          'n_classes': 2,
          'n_timesteps': n_timesteps,
          'n_features': n_features,
          'lead': 0,
          'shuffle': False,
         }

params_val={ 
          'batch_size': 10,
          'n_classes': 2,
          'n_timesteps': n_timesteps,
          'n_features': n_features,
          'lead': 0,
          'shuffle': False,
         }


# Datasets

train_x=data.iloc[0:100,0:data.shape[1]-1].values
train_y=data.iloc[0:100,data.shape[1]-1].values
val_x=data.iloc[101:149,0:data.shape[1]-1].values
val_y=data.iloc[101:149,data.shape[1]-1].values

最后是一个示例网络。...

# Generators
training_generator = DataGenerator(train_x,train_y, **params_train)
validation_generator = DataGenerator(val_x,val_y, **params_val)

# Design model


model = Sequential()
model.add(LSTM(n_features,return_sequences=True,input_shape = (n_timesteps,n_features)))
model.add(Dropout(0.4))
model.add(LSTM(n_features,input_shape = (n_timesteps,n_features)))
model.add(Dense(n_features,activation='relu'))
model.add(Dense(2))
model.add(Activation('sigmoid'))
model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['categorical_accuracy'])

# Train model on dataset
model.fit_generator(generator=training_generator,
                    use_multiprocessing=False,
                    workers=1, epochs=1)

我将它限制为一个工作人员，并且不进行任何多进程处理，因此我可以轻松进行调试，请随时进行更改...

使用此设置，我得到的错误是

StopIteration: index 4 is out of bounds for axis 0 with size 4

似乎可以正常使用，直到生成第39-30个序列的第3批为止。从那以后，它会出错，并且不会获取49-40系列数据。

是什么原因引起的思考？谢谢！

keras序列生成器：错误，未使用提供的完整数据

0 个答案: