您的评估准确性为89.9％.我测试了我提供给您的用于打印结果的代码，这些代码对于我使用的模型和数据是正确的.请显示您用于model.predict的代码.应该像下面这样.不要使用预测生成器fit_generator或valuate_generator，因为在将来的tensorflow版本中，它将被折旧.只需使用拟合，预测或评估即可.

your evaluation accuracy is 89.9%. I tested the code I gave you to print the results and they were correct for the model and data I was using. Please show the code you used for model.predict. Should be like below. DO not use predict-generator fit_generator or evaluate_generator as the will be depreciated in future versions of tensorflow. Just use fit, predict or evaluate.

length=len(labels)
test_steps=  int(length/batch_size) +1
preds=model.predict(test_gen,  verbose=0, steps=test_steps)

看一下您的pred，第一个预测应该是对于索引= 4，这是E类，所以是正确的；第二个预测概率最高的概率是1，这是B类，所以是正确的，其余的也是.您确定标签为labels = test_gen.labels

Looking at your preds print out the first prediction should have been for index=4 which is class E so that is correct, second prediction index with highest probability is 1 which is class B so that is correct, so are the rest. Are you sure you have the labels as labels=test_gen.labels