如何在模型内部更改Keras模型输出的阈值？

Question

我正在构建这样一个 Keras 模型：

Y = 1 for X >= 0.5
Y = 0 for X < 0.5

我的模特：

def define_model():
    model = Sequential()
    model.build(input_shape = (None, 1))
    model.add(Dense(1, activation = 'sigmoid'))

    opt = SGD(learning_rate = 0.01, momentum = 0.99)
    model.compile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy'])
    
    return model

创建模型后，我将权重设置为 [1] 并将偏差设置为 [-0.5] 现在，我获得了很高的准确性，但以下输入的输出错误：

[[0.50000006]
 [0.5       ]
 [0.50000001]
 [0.50000002]
 [0.50000007]
 [0.50000004]
 [0.50000001]
 [0.50000004]
 [0.50000004]
 [0.50000001]
 [0.50000003]
 [0.50000007]
 [0.50000008]
 [0.50000008]
 [0.50000004]
 [0.50000002]
 [0.50000006]
 [0.50000006]
 [0.5000001 ]
 [0.50000008]
 [0.50000002]
 [0.50000004]
 [0.50000006]
 [0.50000004]
 [0.5       ]
 [0.50000005]
 [0.50000003]
 [0.50000007]
 [0.50000004]
etc.

所以，模型已经知道 Y = 1 for only X > 0.5 但我需要 Y = 1 for X >= 0.5。

我知道这可以通过 pred = model.predict(X) 获取输出然后手动比较来完成。但我希望这在模型内部完成。 model.predict_classes应该有个门槛。我想改变这个门槛。我该怎么做？

Answer 1

predict_classes 不允许我们更改阈值。 keras是这样实现的

def predict_classes(self, x, batch_size=32, verbose=0):
    proba = self.predict(x, batch_size=batch_size, verbose=verbose)
    if proba.shape[-1] > 1:
      return proba.argmax(axis=-1)
    else:
      return (proba > 0.5).astype('int32')

如果你想有自己的阈值，那么你将不得不重载该方法。

代码

class MySequential(keras.models.Sequential):
  def __init__(self, **kwargs):
    super(MySequential, self).__init__(**kwargs)

  def predict_classes(self, x, batch_size=32, verbose=0):
    proba = self.predict(x, batch_size=batch_size, verbose=verbose)
    return (proba >= 0.6).astype('int32')

def define_model():
    model = MySequential()
    model.add(keras.layers.Dense(1, activation = 'sigmoid', input_shape=(None, 1)))

    opt = keras.optimizers.SGD(learning_rate = 0.01, momentum = 0.99)
    model.compile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy'])
    
    return model

# Test

model = define_model()
x = np.random.randn(5)
print (model.predict(x))
print (model.predict_classes(x))

如何在模型内部更改Keras模型输出的阈值？

How to change the threshold of output of Keras model inside the model?

python

predict

neural-network

keras

tensorflow

代码