`我正在将 tensorflow 2.4.1 升级到 2.15.0。我正在重新使用用 2.4.1 训练的模型。我无法在升级版本中进行批量预测。输入格式为 1 tensor.Model
`我正在将 tensorflow 2.4.1 升级为 2.15.0。我正在重新使用用 2.4.1 训练的模型。我无法在升级版本中进行批量预测。输入格式为 1 张量。模型输入形状：[(None, 46, 1), (None, 46, 1)]批量数据形状：[(200, 46, 1), (200, 46, 1)]
两个版本的模型摘要相同。
import tensorflow as tf
import numpy as np
class predictDataGenerator(tf.keras.utils.Sequence):
'Generates data for Keras'
def __init__(self, inData, nsamps, batch_size=250, dim=46,
n_channels=1, inData_ch2=None, n_classes=2,
avg_num=3, befN=18, shuffle=True):
'Initialization'
self.dim = dim
self.batch_size = batch_size
self.inData = inData
self.inData_ch2 = inData_ch2
self.n_channels = n_channels
self.n_classes = n_classes
self.avg_num = avg_num
self.befN = befN
self.part_start = (self.befN + 1) * self.avg_num + self.avg_num // 2 - (self.befN + 1)
self.part_window = self.dim + 1
self.obj_window = (self.dim + 1) * self.avg_num
self.part_tile = np.tile(np.arange(self.dim), (self.batch_size, 1))
self.part_tile += np.arange(self.batch_size).reshape(self.batch_size, 1)
self.shuffle = shuffle
self.nsamps = nsamps
self.on_epoch_end()
self.batch_count = self.__len__()
def __len__(self):
'Denotes the number of batches per epoch'
return int(np.ceil(self.nsamps / self.batch_size))
def __getitem__(self, index):
'Generate one batch of data'
# Generate indexes of the batch
batch_data = self.inData[index * self.batch_size:(index + 1) * self.batch_size + self.obj_window - 1]
num_points = batch_data.size - (self.obj_window - 1)
X_OBJ = np.empty((num_points, self.dim, self.n_channels))
X_PART = np.empty((num_points, self.dim, self.n_channels))
for i in range(self.n_channels):
X_obj, X_part = self.__data_generation(batch_data, num_points)
X_PART[:, :, i] = X_part
X_OBJ[:, :, i] = X_obj
return (X_OBJ, X_PART)
def on_epoch_end(self):
'Updates indexes after each epoch'
# Optional: Implement shuffling if needed
if self.shuffle:
self.indexes = np.arange(self.nsamps)
np.random.shuffle(self.indexes)
def get_part_data(self, batch_data, num_points):
'Prepare part data'
part_end = self.part_start + num_points + self.part_window
temp_data = batch_data[self.part_start:part_end] # Required samples
temp_data = np.diff(temp_data) # temp_data[1:] - temp_data[:-1]
X_part2 = temp_data[self.part_tile[:num_points, :]]
return X_part2
def get_obj_data(self, batch_data, num_points):
'Prepare object data'
X_OBJ2 = np.empty((num_points, self.dim))
for i in range(self.avg_num):
temp_end = batch_data.size - (batch_data.size - i) % self.avg_num
temp_data = batch_data[i:temp_end]
temp_data = temp_data.reshape(-1, self.avg_num)
temp_data = np.sum(temp_data, axis=1) / self.avg_num
temp_data = np.diff(temp_data)
num_windows = temp_data.size - self.dim + 1
obj_tile = np.tile(np.arange(self.dim), (num_windows, 1))
obj_tile += np.arange(num_windows).reshape(num_windows, 1)
X_OBJ2[i::self.avg_num] = temp_data[obj_tile]
return X_OBJ2
def __data_generation(self, batch_data, num_points):
'Generates data for one batch'
X_part = self.get_part_data(batch_data, num_points)
X_obj = self.get_obj_data(batch_data, num_points)
X_part = X_part - np.mean(X_part, axis=1).reshape(-1, 1)
X_part = X_part / ((np.std(X_part, axis=1)).reshape(-1, 1) + 1e-100)
X_obj = X_obj - np.mean(X_obj, axis=1).reshape(-1, 1)
X_obj = X_obj / ((np.std(X_obj, axis=1)).reshape(-1, 1) + 1e-100)
return X_obj, X_part
X_test = np.arange(1140)
pred_length = X_test.size-l_start-l_end
bsize = 200
testgen = predictDataGenerator(X_test.reshape(-1),
pred_length,batch_size=bsize,dim = Window,
avg_num = avgN,
befN = window1,shuffle=False)
print(len(testgen))
probs = bd_model.predict(testgen)
错误：
ValueError：层“模型”需要 2 个输入，但实际收到 1 个输入张量。收到的输入：[ ]`
层“模型”需要 2 个输入，但实际收到 1 个输入张量。输入记录：[
]
下载声明：
本站所有软件和资料均为软件作者提供或网友推荐发布而来，仅供学习和研究使用，不得用于任何商业用途。如本站不慎侵犯你的版权请联系我，我将及时处理，并撤下相关内容！
