tf.nn.dropout使用时输出相同的值

Question

当我不在模型中对猫和狗进行分类时，预测值保持正常，即所有图像的值都不相同。

但是，当我将tf.nn.dropout与keep_prob = 0.8一起用于我的模型时，为了规范化模型和提高准确性，建议使用tflearn，它会不断预测相同的值。我该如何解决？那里的每个教程或代码都使用array([[ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00]` [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], [ 2.20128131e+00, 1.78127408e+00], ，但这不会发生。

{{1}}